Can an AI presenter look real?

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Photographic human appearance maintained throughout video. (0, 6.016s)
• realism level 2: pass. Plausible skin texture, hair, and lighting without excessive synthetic artifacting. (0, 6.016s)
• realism level 3: pass. Natural facial movement and absence of doll-like eyes or CGI shading. (0, 6.016s)
• realism level 4: pass. Ordinary home-camera appearance holds up with no obvious avatar or CGI tells. (0, 6.016s)
• lip sync: pass. Mouth shapes align accurately with the spoken dialogue. (0.5, 5.5s)

openai/gpt-5.6-sol

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with natural facial structure, hair, clothing, and room lighting rather than a cartoon or obvious 3D-render look. (0, 3, 5.936s)
• realism level 2: pass. Across the sampled frames, skin texture and wrinkles, eyes behind glasses, lips, gray hair, teeth, and illumination remain mostly photographically plausible without major waxy or painted regions. (0.5, 2, 4.5, 5.5s)
• realism level 3: pass. Careful inspection of the visible frames shows consistent facial detail, plausible eye closures, natural hair strands, and believable expressions; no readily noticeable doll eyes, painted hair, CGI shading, or synthetic expression is visible. (1, 3.5, 4.5, 5.936s)
• realism level 4: pass. The sampled views convincingly resemble ordinary live-action home-camera footage: facial texture, glasses, hair, teeth, clothing, room lighting, and blink states show no meaningful visible synthetic tell. (0, 1.5, 3.5, 4.5, 5.936s)
• lip sync: unknown. Only sampled still frames are supplied, with no reliable audio-temporal evidence; mouth shapes at isolated timestamps cannot establish alignment with audible words. (0, 2.5, 5.936s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, skin texture, hair, and clothing consistent with a real person in a home setting. No predominant cartoon, illustration, or obvious 3D render artifacts are visible. (0, 5.936s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic conditions. Lighting is consistent with indoor ambient sources. Minor synthetic cues are not readily apparent; no major waxy, painted, or plastic appearance is observed. (0, 5.936s)
• realism level 3: pass. Under sampled frame inspection, the subject shows strong camera realism. Eyes, lips, skin, and hair exhibit natural texture and lighting. No readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading are visible. Minor deviations are not perceptible in sampled frames. (0, 5.936s)
• realism level 4: pass. The appearance is convincingly ordinary live-action home-camera footage under close inspection of sampled frames. No meaningful visible synthetic tells such as CGI/avatar-like eyes, lips, skin, hair, light, or expression are observed. Wrinkles and ambient clutter are natural and not indicative of synthetic artifacts. (0, 5.936s)
• lip sync: unknown. Sampled frames alone cannot prove lip sync with audible words. No audio is provided for comparison, and temporal continuity between frames is insufficient to confirm alignment of mouth movements with speech. (0, 5.936s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic human subject presented in an ordinary home setting. (1s)
• realism level 2: pass. Plausible skin texture, hair, lighting, and natural eye movements. (2.5s)
• realism level 3: pass. High camera realism with natural facial wrinkling and believable lighting. (3s)
• realism level 4: fail. Subtle synthetic smoothing and morphing visible around teeth and inner mouth motion. (2.8, 4.2s)
• lip sync: pass. Visible mouth shapes correspond accurately to the spoken phrases throughout. (0.5, 4.5s)

openai/gpt-5.6-sol

sampled_frames · 4/4 rubric index

• realism level 1: pass. The visible subject has a coherent photographic human face, hair, glasses, clothing, and home setting rather than a cartoon, illustration, or obvious 3D render. (0, 3, 5.5s)
• realism level 2: pass. Across sampled frames, skin texture and wrinkles, eyes behind glasses, teeth and lips, gray hair, and mixed window/interior lighting remain mostly photographically plausible without major waxy or plastic rendering. (0.5, 2, 4.5s)
• realism level 3: pass. Careful inspection of the sampled frames shows natural facial detail, hair strands, eyelid closures, mouth shapes, and scene lighting; no readily noticeable doll eyes, painted hair, CGI shading, or strongly synthetic expression is visible. (1, 3.5, 4.5, 5.936s)
• realism level 4: pass. The sampled views convincingly resemble ordinary live-action home-camera footage: facial texture, glasses, hair, teeth, clothing, and environmental lighting show no meaningful visible synthetic tell in the provided coverage. (0, 2.5, 4.5, 5.936s)
• lip sync: unknown. Only sampled still frames are supplied, and no audible words are available for temporal comparison; mouth positions in isolated frames cannot establish lip sync. (0, 3, 5.936s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin texture, realistic hair, and plausible facial structure without cartoon, illustration, or obvious 3D render artifacts. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic qualities observed across sampled frames. (0s)
• realism level 3: pass. Under careful inspection of sampled frames, only minor deviations are present; no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading. Expression and lighting appear natural. (0s)
• realism level 4: pass. The appearance is convincingly ordinary for a live-action home-camera recording under close inspection. No perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression detected in sampled frames. (0s)
• lip sync: unknown. Sampled frames alone cannot prove lip sync; audio is not provided for comparison, and temporal continuity between frames cannot confirm mouth alignment with speech. (0s)

google/gemini-3.8-flash

native_video_audio · 2/4 rubric index

• realism level 1: pass. Photographic human appearance maintained throughout without 2D cartoon rendering. (0, 7.2s)
• realism level 2: pass. Plausible skin texture, hair, and lighting despite minor mouth anomalies. (0, 7.2s)
• realism level 3: fail. Synthetic mouth deformations and garbled overlay text break realism. (1, 4s)
• realism level 4: fail. Clear AI artifacts including distorted subtitles and unnatural mouth kinematics. (1, 5s)
• identity: pass. Matches the reference woman with wavy gray hair, dark glasses, and identical clothing. (0, 7.2s)
• fixed camera: pass. Static framing and consistent room background throughout. (0, 7.2s)
• face stability: pass. Facial structure remains stable across speech without major warping. (0, 7.2s)
• speech: pass. Audible speech matches the exact required sentence cleanly. (0.2, 5.5s)
• lip sync: pass. Lip movements generally correspond to the spoken words. (0.2, 5.5s)

openai/gpt-5.6-sol

sampled_frames · 2/4 rubric index

• realism level 1: pass. The visible subject has a coherent photographic human appearance with natural room lighting, detailed hair, glasses, skin, and clothing rather than a cartoon or obvious 3D-render presentation. (0, 3, 6.5s)
• realism level 2: pass. Across the sampled frames, skin, lips, silver hair, glasses, and lighting remain mostly photographically plausible without major waxy, painted, or plastic regions. (0.5, 3.5, 6s)
• realism level 3: fail. Close inspection shows conspicuously bright, glassy blue eyes and somewhat smoothed facial texture and expression, especially in the later frontal frames, creating noticeable synthetic/avatar cues. (4, 5.5, 7.12s)
• realism level 4: fail. The unusually vivid, glassy eyes and polished facial rendering remain meaningful synthetic tells, so the subject does not fully resemble ordinary live-action home-camera footage under close inspection. (4, 5.5, 6.5s)
• identity: pass. The candidate visibly matches the reference's older feminine face, shoulder-length wavy silver hair, black rectangular glasses, navy cardigan, and light shirt in the same home-office setting. (0, 3, 6.5s)
• fixed camera: pass. All supplied frames retain the same frontal desk-camera composition and coherent room geometry, with no visible cut, zoom, or viewpoint change. (0, 3.5, 7.12s)
• face stability: pass. The facial identity, glasses, hairline, eyes, and mouth remain coherent across the sampled frames; visible mouth shapes and the briefly shown teeth are plausible in those frames. (0, 1, 3.5, 5.5, 7.12s)
• speech: unknown. No audible audio evidence is supplied with the sampled frames, and the distorted on-screen text cannot establish the exact spoken sentence. (1, 7.12s)
• lip sync: unknown. The sampled still frames show varying mouth positions but provide neither audible speech nor sufficient continuous temporal evidence to judge alignment with words. (0.5, 2, 4, 7.12s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 3/4 rubric index

• realism level 1: pass. Character exhibits coherent photographic human appearance: natural skin texture, realistic hair, and plausible facial structure without cartoon, illustration, or obvious 3D render cues. Visible at 0.0s. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic qualities. Minor synthetic cues may exist but are not dominant. Visible at 0.0s and 4.0s. (0, 4s)
• realism level 3: pass. Under careful inspection, face shows strong camera realism: eyes are not doll-like, skin lacks beauty-render smoothness, hair is not painted, shading is natural, and expressions are plausible. Visible at 0.0s, 2.0s, and 4.0s. (0, 2, 4s)
• realism level 4: fail. Perceptibly CGI/avatar-like qualities: eyes have unnatural blue glow and uniform reflection, skin texture lacks micro-variation, and lip movements appear slightly stiff and synthetic. Fails strict home-camera realism. Visible at 1.0s, 3.0s, and 5.0s. (1, 3, 5s)
• identity: pass. Character matches fictional reference: silver wavy hair, black-rimmed glasses, navy cardigan over white shirt, and seated at wooden desk. Facial structure and expression align with reference. Visible at 0.0s. (0s)
• fixed camera: pass. Frontal fixed framing throughout; background (bookshelves, window, printer, doorway) remains consistent with no artificial camera movement or cuts. Visible across all sampled frames. (0, 7.12s)
• face stability: pass. Face remains consistent in position and expression; eyes, mouth, and teeth are plausible and coherent across sampled frames. No visible distortions or flickering. Visible at 0.0s, 2.5s, 5.5s. (0, 2.5, 5.5s)
• speech: unknown. No audio provided; cannot verify if the exact sentence is spoken. Text overlay is garbled and not the requested sentence. (0s)
• lip sync: unknown. No audio provided; sampled frames alone cannot prove lip sync. Mouth movements are visible but cannot be correlated to speech without audio. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Photographic depiction of an adult woman in an interior setting without cartoon styling. (0, 6s)
• realism level 2: pass. Plausible skin texture, hair, lighting, and facial features throughout. (0, 6s)
• realism level 3: pass. Naturalistic expressions, eye movement, and lighting consistent with live camera footage. (0, 6s)
• realism level 4: pass. No clear synthetic artifacts observed in skin, hair, or motion during playback. (0, 6s)
• identity: pass. Matches the reference character's features, grey wavy hair, black glasses, and dark cardigan. (0, 6s)
• fixed camera: pass. Camera maintains a steady frontal position facing the desk without transitions. (0, 6s)
• face stability: pass. Facial geometry, mouth, teeth, and eyes remain coherent while speaking. (0, 6s)
• speech: pass. Audio clearly states: 'ChatGPT can help you reply to a personal letter without ever seeing the letter.' (0, 5.2s)
• lip sync: pass. Mouth shapes correlate accurately with the spoken phrases across the entire line. (0, 5.2s)

openai/gpt-5.6-sol

sampled_frames · 2/4 rubric index

• realism level 1: pass. The character has a coherent photographic human appearance across the sampled frames, rather than a cartoon, illustration, or obvious 3D render. (0, 3, 5.92s)
• realism level 2: pass. Skin texture, wrinkles, eyes behind glasses, lips, silver hair, and room lighting remain mostly photographically plausible in the visible samples, with only mild synthetic smoothness. (0.5, 2.5, 5s)
• realism level 3: fail. Under closer inspection, the unusually bright, glassy blue eyes and overly clean facial rendering give the face a noticeable generated beauty-render quality, especially during widened-eye expressions. (2.5, 4, 4.5s)
• realism level 4: fail. The sampled face does not fully resemble ordinary unprocessed home-camera footage: glassy saturated eyes, smooth skin rendering, and slightly synthetic mouth/expression shaping remain perceptible. (1, 3, 4s)
• identity: pass. The candidate visibly matches the reference character's facial structure, shoulder-length silver wavy hair, dark rectangular glasses, cream shirt, and navy cardigan. (0, 3, 5.92s)
• fixed camera: pass. All supplied frames retain the same frontal desk-camera composition and coherent room layout without a visible cut, zoom, or viewpoint change. (0, 3, 5.92s)
• face stability: pass. Across the sampled coverage, facial identity and features remain consistent; eyes, mouth, and visible teeth preserve plausible shapes through several expressions and blinks. (0, 1, 3.5, 5.5, 5.92s)
• speech: unknown. Only sampled frames are supplied, with no accessible audio; exact audible wording cannot be verified. (0, 5.92s)
• lip sync: unknown. The sampled frames show changing mouth shapes, but no accessible audio or continuous temporal coverage is available to establish alignment with audible words. (0, 2, 3.5, 5.92s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. Character exhibits coherent photographic human appearance with natural skin texture, hair, and lighting; no cartoon, illustration, or obvious 3D render artifacts visible in sampled frames. (0, 5.92s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic appearance observed across sampled frames. (0, 5.92s)
• realism level 3: pass. Under sampled inspection, no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading; minor synthetic cues not evident in visible coverage. (0, 5.92s)
• realism level 4: pass. Appearance convincingly matches ordinary live-action home-camera footage under close inspection; no perceptible CGI/avatar-like eyes, lips, skin, hair, light, or expression in sampled frames. (0, 5.92s)
• identity: pass. Face, silver wavy hair, glasses, and outfit match the fictional reference identity exactly across all sampled frames. (0, 5.92s)
• fixed camera: pass. Frontal fixed desk-camera framing maintained throughout; background room elements remain coherent and unchanged, indicating no artificial camera movement. (0, 5.92s)
• face stability: pass. Face, eyes, mouth, and teeth remain consistent and plausible across all sampled frames; no visible distortions or instability observed. (0, 5.92s)
• speech: unknown. Audio is not provided; therefore, it is impossible to verify whether the exact audible sentence was spoken. (0s)
• lip sync: unknown. Audio is not provided; sampled frames alone cannot confirm lip movement alignment with any audible words. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Depicts a photographic human appearance in a realistic home office. (1s)
• realism level 2: pass. Plausible skin texture, lighting, and hair rendering throughout. (2.5s)
• realism level 3: pass. Natural facial movement and believable environmental lighting with minimal digital artifacts. (3.5s)
• realism level 4: pass. Appears convincingly like standard live-action home-camera footage. (4s)
• identity: pass. Features match the reference character's wavy silver hair, glasses, face shape, and wardrobe. (0.5s)
• fixed camera: pass. Camera maintains a steady frontal angle with no cuts or frame shifts. (2s)
• face stability: pass. Facial structure, eyes, and mouth remain stable and coherent throughout speech. (3s)
• speech: pass. Spoken sentence matches the required prompt text accurately. (0, 5.5s)
• lip sync: pass. Mouth shapes align naturally with the spoken audio syllables. (1.5, 4.5s)

openai/gpt-5.6-sol

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance across the sampled frames, with natural room lighting and no predominantly illustrated or obvious 3D-rendered look. (0, 3, 5.957007s)
• realism level 2: pass. Skin texture, eyes behind glasses, lips, silver hair, and room illumination remain mostly photographically plausible; no major waxy, painted, or plastic appearance is visible. (0.5, 2.5, 4.5s)
• realism level 3: pass. Careful inspection of the sampled views shows detailed skin and hair, plausible eye reflections and facial shading, and natural-looking speech expressions without a readily noticeable CGI tell. (1, 3.5, 5.5s)
• realism level 4: pass. Across the supplied frames, the subject and home setting convincingly resemble ordinary live-action desk-camera footage; no meaningful visible synthetic defect in eyes, lips, skin, hair, lighting, or expression is apparent. (0, 2, 4, 5.957007s)
• identity: pass. The candidate visibly matches the reference's facial structure, shoulder-length wavy silver hair, dark rectangular glasses, navy cardigan, and light shirt. (0, 3, 5.957007s)
• fixed camera: pass. All sampled frames retain the same frontal desk-camera composition and coherent room geometry, with no visible reframing or artificial camera change. (0, 3, 5.957007s)
• face stability: pass. Facial proportions, glasses, hairline, eyes, and mouth remain consistent in the sampled coverage; open-mouth expressions and the briefly visible teeth appear plausible. (0, 1, 3.5, 5.957007s)
• speech: unknown. No audio is supplied in the sampled-frame evidence, so the exact audible sentence cannot be verified. (0, 5.957007s)
• lip sync: unknown. The sampled frames show varying mouth shapes but provide no audio or continuous temporal evidence, so alignment with audible words cannot be assessed. (0.5, 2.5, 4.5s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with no predominant cartoon, illustration or obvious 3D render cues across all sampled frames. (0, 5.957007s)
• realism level 2: pass. Skin, eyes, lips, hair and lighting appear mostly plausible with photographic qualities; no major waxy, painted or plastic appearance is visible in any frame. (0, 5.957007s)
• realism level 3: pass. Under careful inspection of sampled frames, only minor deviations are visible; no readily noticeable doll eyes, beauty-render skin, painted hair, CGI shading or synthetic expression is present. (0, 5.957007s)
• realism level 4: pass. The subject appears convincingly like ordinary live-action home-camera footage under close inspection; no perceptibly CGI/avatar-like eyes, lips, skin, hair, light or expression is visible in any sampled frame. (0, 5.957007s)
• identity: pass. The subject matches the fictional reference face, silver wavy hair, glasses and outfit across all sampled frames. (0, 5.957007s)
• fixed camera: pass. The framing remains frontal and fixed with coherent room background; no artificial camera movement or changes are visible across all frames. (0, 5.957007s)
• face stability: pass. Face, eyes, mouth and teeth remain consistent and plausible across all sampled frames; no distortions or incoherent changes are visible. (0, 5.957007s)
• speech: unknown. Audio is not provided in the input; therefore, it is impossible to verify whether the exact audible sentence was spoken. (0s)
• lip sync: unknown. Audio is not provided; sampled frames alone cannot prove lip sync alignment with spoken words. (0s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic human appearance maintained throughout without overt cartoon or 3D rendering stylization. (0, 3s)
• realism level 2: pass. Plausible photographic skin, hair, and ambient lighting present without plastic or waxy sheen. (1, 4s)
• realism level 3: pass. Strong camera realism with natural facial wrinkling, hair texture, and believable eye movements. (2, 5s)
• realism level 4: fail. Minor synthetic smoothing and mouth motion artifacts typical of AI generation around the lips during speech. (1.5, 3.5s)
• identity: pass. Character closely matches the reference in facial structure, wavy silver hair, glasses, and outfit. (0.5, 5.5s)
• fixed camera: pass. Camera maintains a steady, fixed frontal framing of the room with no erratic movements. (0, 6s)
• face stability: pass. Face geometry and features remain consistent across the video duration without major warping. (1, 5s)
• speech: fail. Initial word 'ChatGPT' is truncated, beginning abruptly at 'T can help you...' (0, 1.2s)
• lip sync: pass. Mouth articulations generally correspond well with the audible spoken words throughout. (1, 5.5s)

openai/gpt-5.6-sol

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance throughout the sampled frames, with natural room lighting and no predominantly cartoon or illustrated rendering. (0, 3, 5.957007s)
• realism level 2: pass. Skin texture, wrinkles, eyes behind glasses, lips, silver hair, clothing, and lighting are mostly photographically plausible in the visible coverage without major waxy or plastic appearance. (0.5, 2.5, 5s)
• realism level 3: pass. Across close, well-lit views, facial texture, hair strands, glasses, mouth shapes, and expressions remain strongly camera-realistic; no readily noticeable doll eyes, painted hair, or CGI shading is visible. (1, 3.5, 5.5s)
• realism level 4: pass. The sampled frames convincingly resemble ordinary live-action home-camera footage under close inspection, with coherent skin detail, eye reflections, hair, lighting, and expressions and no meaningful visible synthetic tell. (0, 1.5, 4, 5.957007s)
• identity: pass. The candidate visibly matches the fictional reference's facial structure, shoulder-length wavy silver hair, dark rectangular glasses, navy cardigan, and light shirt. (0, 3, 5.5s)
• fixed camera: pass. All supplied frames retain the same frontal desk-camera composition and coherent home-office background without a visible cut, zoom, or viewpoint change. (0, 3, 5.957007s)
• face stability: pass. The face remains consistent across sampled frames; eyes, mouth, and visible teeth retain plausible structure through several expressions. This establishes sampled-frame stability only. (0.5, 1.5, 3.5, 5.957007s)
• speech: unknown. No audio modality is supplied, so the exact audible sentence cannot be verified from frames. (0s)
• lip sync: unknown. Only sampled frames are available and no audio is supplied; varying mouth shapes cannot establish alignment with audible words. (0.5, 3, 5.957007s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with natural skin tones, hair texture, and clothing; no predominant cartoon, illustration, or obvious 3D render artifacts are visible in sampled frames. (0, 5.957007s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic qualities are observable in sampled frames. (0, 5.957007s)
• realism level 3: pass. Under sampled frame inspection, the subject shows strong camera realism with only minor deviations; no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading are visible. (0, 5.957007s)
• realism level 4: pass. The appearance is convincingly ordinary live-action under close inspection of sampled frames; no perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression are visible. (0, 5.957007s)
• identity: pass. The subject matches the fictional reference face, silver wavy hair, glasses, and outfit across all sampled frames. (0, 5.957007s)
• fixed camera: pass. The framing remains frontal and fixed throughout sampled frames; background elements (bookshelf, window, printer) remain spatially coherent with no artificial camera movement. (0, 5.957007s)
• face stability: pass. Face, eyes, mouth, and teeth remain consistent and plausible across all sampled frames; no distortions or unnatural shifts are observed. (0, 5.957007s)
• speech: unknown. Audio is not provided in the sampled frames; therefore, the exact audible sentence cannot be verified. (0s)
• lip sync: unknown. Audio is not available; sampled frames alone cannot prove lip sync alignment with speech. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Character has a natural photographic human presentation. (0, 3s)
• realism level 2: pass. Plausible skin texture, lighting, and hair throughout the clip. (1, 4s)
• realism level 3: pass. Facial features, lighting, and natural motion hold up under close viewing. (2, 5s)
• realism level 4: pass. Consistent home-camera look without notable digital artifacts. (0.5, 5.5s)
• identity: pass. Face, silver wavy hair, glasses, and outfit match reference. (0, 3s)
• fixed camera: pass. Camera framing and background remain stationary throughout. (0, 6s)
• face stability: pass. Face and mouth maintain temporal consistency across speech. (1, 5s)
• speech: pass. Audible sentence exactly matches requested transcript. (0, 5s)
• lip sync: pass. Mouth movement matches the spoken audio stream. (0, 5s)

openai/gpt-5.6-sol

sampled_frames · 4/4 rubric index

• realism level 1: pass. The sampled frames consistently depict a coherent photographic human rather than a cartoon, illustration, or obvious 3D render. (0, 3, 5.5s)
• realism level 2: pass. Skin texture, wrinkles, eyes behind glasses, lips, silver hair, and room lighting appear mostly photographically plausible across the samples, without major waxy or plastic rendering. (0.5, 2, 4.5s)
• realism level 3: pass. Close visible facial details remain strongly camera-realistic; hair strands, skin creases, eye reflections, and changing expressions show no readily noticeable doll-like or painted appearance in the supplied frames. (1, 3.5, 5s)
• realism level 4: pass. Across the supplied close, well-lit frames, the character and room convincingly resemble ordinary live-action home-camera footage, with no meaningful visible synthetic tell in eyes, lips, skin, hair, lighting, or expression. (0, 2.5, 4.5, 5.936s)
• identity: pass. The candidate closely matches the fictional reference in facial structure, shoulder-length wavy silver hair, dark rectangular glasses, cream shirt, and navy cardigan. (0, 3, 5.5s)
• fixed camera: pass. All samples retain the same frontal desk-camera composition, table edges, window, shelving, doorway, and ceiling light without a cut, zoom, or viewpoint change. (0, 3, 5.936s)
• face stability: pass. Facial proportions, glasses, hairline, eyes, and mouth remain consistent across samples; blinks and varied mouth shapes appear plausible, and visible teeth at 1.0 and 4.5 seconds are coherent. (1, 1.5, 3.5, 4.5, 5.5s)
• speech: unknown. Only sampled frames are supplied, with no audible track available to verify the exact sentence. (0, 5.936s)
• lip sync: unknown. The sampled still frames show changing mouth shapes but provide neither audible words nor sufficient continuous temporal coverage to assess alignment. (0, 2, 4.5, 5.936s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. Subject exhibits coherent photographic human appearance with natural skin tone, hair, and clothing; no cartoon, illustration, or obvious 3D render artifacts visible in sampled frames. (0, 5.936s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic appearance observed in sampled frames. (0, 5.936s)
• realism level 3: pass. Under sampled frame inspection, subject exhibits strong camera realism: natural wrinkles, lighting, and expression; no readily noticeable doll eyes, beauty-render skin, or CGI shading. (0, 5.936s)
• realism level 4: pass. Appearance is convincingly ordinary live-action home-camera under close inspection; no perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression in sampled frames. (0, 5.936s)
• identity: pass. Subject matches the fictional reference: silver wavy hair, black-rimmed glasses, navy cardigan over cream shirt, and seated position at wooden table with paper envelope. (0, 5.936s)
• fixed camera: pass. Frontal fixed desk-camera framing maintained throughout; background elements (bookshelf, window, printer, doorway) remain stationary and coherent. (0, 5.936s)
• face stability: pass. Face remains consistent in position and expression across sampled frames; eyes, mouth, and teeth appear plausible and coherent without distortion. (0, 5.936s)
• speech: unknown. Audio is not provided in the sampled frames; cannot verify the exact audible sentence. (0s)
• lip sync: unknown. Audio is not provided; sampled frames alone cannot prove lip sync alignment with speech. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Depicts a photographic human figure rather than a stylized cartoon or 3D render. (0, 7.2s)
• realism level 2: pass. Plausible skin, hair, and lighting without excessive waxy or synthetic artifacts. (0, 7.2s)
• realism level 3: pass. Natural live-action appearance with realistic facial expressions and textures. (0, 7.2s)
• realism level 4: pass. Convincingly ordinary home-camera footage without discernible synthetic tells. (0, 7.2s)
• identity: pass. Consistent face, wavy silver hair, glasses, and wardrobe matching the reference. (0, 7.2s)
• fixed camera: pass. Frontal framing from a fixed camera position in a consistent room. (0, 7.2s)
• face stability: pass. Facial structure, eyes, and mouth remain stable throughout the clip. (0, 7.2s)
• speech: pass. Exact requested sentence spoken clearly without omissions or additions. (0, 5.7s)
• lip sync: pass. Mouth articulations match the spoken audio sequence accurately. (0, 5.7s)

openai/gpt-5.6-sol

sampled_frames · 2/4 rubric index

• realism level 1: pass. Sampled frames consistently depict a coherent photographic older woman in a naturally lit home setting, not a cartoon or obvious 3D render. (0, 3, 6s)
• realism level 2: pass. Skin, wrinkles, glasses, silver hair, lips, and room lighting are mostly photographically plausible across the supplied frames, with only mild synthetic smoothness. (0.5, 3.5, 6.5s)
• realism level 3: fail. Under inspection, the unusually bright uniform blue irises, smoothed facial texture, and slightly synthetic mouth/expression are readily noticeable. (2.5, 4, 5.5s)
• realism level 4: fail. The conspicuously vivid doll-like blue eyes and polished facial rendering prevent a convincingly ordinary live-action home-camera appearance. (4, 5, 7.12s)
• identity: pass. The character visibly matches the reference's older face, shoulder-length wavy silver hair, dark rectangular glasses, navy cardigan, and light shirt. (0, 3, 6s)
• fixed camera: pass. All supplied frames retain the same frontal desk-camera composition and coherent bookshelf, window, doorway, table, and ceiling-light layout. (0, 3.5, 7.12s)
• face stability: pass. Across sampled frames, facial structure, glasses, hair, and features remain consistent; visible eye closures and mouth shapes are plausible, with no major deformation. (0, 1, 3.5, 6.5, 7.12s)
• speech: unknown. No audible audio modality is supplied, and the garbled on-screen text cannot establish that the exact sentence was spoken. (1, 7.12s)
• lip sync: unknown. Only sampled frames are supplied and no audible audio is available, so alignment of mouth motion with words cannot be assessed. (0.5, 2, 4, 7.12s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. Character exhibits coherent photographic human appearance with realistic skin texture, hair, and clothing; no cartoon, illustration, or obvious 3D render artifacts observed in sampled frames. (0, 7.12s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; minor synthetic cues are not readily apparent in sampled frames. (0, 7.12s)
• realism level 3: pass. Under sampled frame inspection, no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading; expression and lighting remain consistent with photographic realism. (0, 7.12s)
• realism level 4: pass. Appearance is convincingly ordinary live-action under close inspection; no perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression observed in sampled frames. (0, 7.12s)
• identity: pass. Character matches the fictional reference: silver wavy hair, glasses, blue cardigan over white shirt, and seated posture at desk are visually consistent with reference image. (0, 7.12s)
• fixed camera: pass. Frontal fixed desk-camera framing is maintained throughout; background elements (bookshelves, window, printer) remain stationary and coherent. (0, 7.12s)
• face stability: pass. Face remains consistent in position and expression across sampled frames; eyes, mouth, and teeth are plausible and coherent without distortion. (0, 7.12s)
• speech: unknown. Audio is not provided; no audible sentence can be verified from visual frames alone. (0, 7.12s)
• lip sync: unknown. Audio is not provided; sampled frames cannot prove lip movement aligns with any audible words. (0, 7.12s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic human appearance maintained throughout without cartoon or stylized rendering. (1.5s)
• realism level 2: pass. Skin, hair, eyes, and lighting appear mostly natural and photographic. (2s)
• realism level 3: pass. Realistic facial movement, natural skin texture, and lighting with only minor generative artifacts. (3s)
• realism level 4: fail. Subtle synthetic smoothing around the mouth and slightly stiff lower-face transitions reveal AI generation. (3.5s)
• identity: pass. Features, wavy silver hair, glasses, and outfit closely match the reference image. (1s)
• fixed camera: pass. Camera framing remains steady and fixed in the room without cuts or drifts. (2.5s)
• face stability: pass. Facial geometry, eyes, mouth, and teeth stay stable throughout the clip. (3s)
• speech: pass. Spoken line matches the required sentence verbatim. (0, 4.8s)
• lip sync: pass. Mouth phonemes visibly synchronize well with the spoken audio. (1, 4.5s)

openai/gpt-5.6-sol

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with natural room lighting and no predominantly illustrated or obvious 3D-render look. (0, 3, 5.961667s)
• realism level 2: pass. Skin, eyes, lips, silver hair, glasses, and lighting remain mostly photographically plausible across the supplied frames, without major waxy or plastic rendering. (0.5, 2.5, 5.5s)
• realism level 3: pass. Close visible views retain natural skin detail, eye appearance, hair texture, and facial shading; no readily noticeable doll eyes, painted hair, or CGI shading is evident in the samples. (3, 4.5, 5.961667s)
• realism level 4: pass. The sampled frames convincingly resemble ordinary live-action home-camera footage; even the closer late framing shows no meaningful visible synthetic tell in the face, hair, glasses, or expression. (0, 4.5, 5.961667s)
• identity: pass. The candidate recognizably matches the fictional reference through facial structure, shoulder-length silver wavy hair, dark rectangular glasses, cream shirt, and navy cardigan. (0, 3.5, 5.961667s)
• fixed camera: fail. Although the frontal desk-camera composition and room remain coherent, the framing progressively tightens: substantially less table and room are visible by the end, indicating an artificial zoom or camera change. (0, 4, 5.961667s)
• face stability: pass. Across the sampled frames, facial identity, glasses, eyes, mouth, and visible teeth remain coherent through varied expressions; continuous stability between samples is not established. (0.5, 1.5, 3, 5.5s)
• speech: unknown. No audio modality is supplied, so the exact audible sentence cannot be verified from mouth shapes in sampled frames. (0, 5.961667s)
• lip sync: unknown. Only sampled frames are supplied and no audio is available; visible mouth changes cannot establish alignment with audible words. (0.5, 3, 5.5s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with no predominant cartoon, illustration or obvious 3D render cues across all sampled frames. (0, 5.961667s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; minor synthetic cues such as slight uniformity in skin texture are present but do not dominate. (0, 5.961667s)
• realism level 3: pass. Under careful inspection, only minor deviations from realism are visible; no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading. Expression appears natural. (0, 5.961667s)
• realism level 4: pass. Convincingly ordinary live-action home-camera appearance under close inspection; no meaningful visible synthetic tell in eyes, lips, skin, hair, light or expression across sampled frames. (0, 5.961667s)
• identity: pass. Face, silver wavy hair, glasses, and outfit (navy cardigan over cream shirt) match the approved fictional reference identity across all sampled frames. (0, 5.961667s)
• fixed camera: pass. Frontal fixed desk-camera framing is maintained; background room elements remain coherent and unchanged, with no artificial camera motion or cuts. (0, 5.961667s)
• face stability: pass. Face remains consistent in position and structure; eyes, mouth, and teeth are plausible and coherent across all sampled frames, with no visible warping or distortion. (0, 5.961667s)
• speech: unknown. Audio is not provided in the sampled frames; therefore, the exact audible sentence cannot be verified. (0s)
• lip sync: unknown. Audio is not available in the sampled frames; therefore, visible mouth alignment with audible words cannot be assessed. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Plausible photographic human appearance throughout. (0, 3s)
• realism level 2: pass. Natural skin texture, hair, and lighting without major synthetic artifacts. (0, 3s)
• realism level 3: pass. Consistent realistic facial details, lighting, and expressions across playback. (0.5, 4s)
• realism level 4: pass. Convincing home-camera realism without obvious CGI or avatar cues. (1, 5s)
• identity: pass. Matches the reference woman, hair, glasses, and wardrobe. (0, 2s)
• fixed camera: pass. Camera framing and background remain fixed and stable. (0, 6s)
• face stability: pass. Facial features, mouth, and eyes remain structurally stable. (0.5, 5.5s)
• speech: pass. Audible sentence matches the requested script completely. (0, 4.8s)
• lip sync: pass. Mouth movement matches spoken words accurately. (0, 4.8s)

openai/gpt-5.6-sol

sampled_frames · 3/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance across all supplied frames, rather than a cartoon, illustration, or obvious 3D render. (0, 3, 5.92s)
• realism level 2: pass. Skin, eyes, lips, silver hair, glasses, and room lighting are mostly photographically plausible in the sampled views, with no major waxy or painted appearance. (0.5, 2.5, 5.5s)
• realism level 3: pass. Close sampled views retain convincing skin detail, hair strands, reflections, and expression; only slight smoothness and unusually vivid eyes appear as minor deviations. (1, 3.5, 5s)
• realism level 4: fail. Under close inspection, the eyes remain unusually bright and glassy and some speaking expressions around the mouth look subtly synthetic, preventing fully ordinary live-action realism. (0, 3, 4.5s)
• identity: pass. The candidate recognizably matches the reference's older face, shoulder-length wavy silver hair, dark rectangular glasses, navy cardigan, and light shirt. (0.5, 3, 5.92s)
• fixed camera: pass. All sampled frames retain the same frontal desk-camera composition and coherent home-office background, without a visible camera change. (0, 3, 5.92s)
• face stability: pass. Facial proportions, glasses, hairline, eyes, and mouth remain consistent across the sampled frames; open-mouth and blink states are plausible in the visible coverage. (0, 1, 3.5, 5.5s)
• speech: unknown. No audio is supplied, so the exact audible sentence cannot be verified from sampled frames. (0, 5.92s)
• lip sync: unknown. Sampled still frames show changing mouth shapes, but without audio and continuous temporal coverage they cannot establish alignment with audible words. (0, 2, 4, 5.92s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 3/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with natural skin tone, hair, and clothing; no cartoon, illustration, or obvious 3D render artifacts are visible in any sampled frame. (0, 5.92s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; minor synthetic cues (e.g., slight uniformity in skin texture) are present but do not dominate the appearance. (0, 5.92s)
• realism level 3: pass. Under sampled inspection, no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading; expression and lighting remain consistent and natural across frames. (0, 5.92s)
• realism level 4: fail. Perceptibly synthetic appearance in eyes (slightly glossy, uniform reflection) and subtle plasticity in lip movement; fails strict live-action home-camera realism under close inspection. (0, 5.92s)
• identity: pass. Recognizably matches the fictional reference: silver wavy hair, black-rimmed glasses, navy cardigan over cream shirt, and seated desk posture are consistent across all frames. (0, 5.92s)
• fixed camera: pass. Frontal fixed framing is maintained; background elements (bookshelves, window, printer) remain static with no artificial camera movement or cuts. (0, 5.92s)
• face stability: pass. Face remains consistent in position and structure; eyes, mouth, and teeth are plausible and coherent across all sampled frames without distortion or flicker. (0, 5.92s)
• speech: unknown. Audio is not provided; cannot verify whether the exact sentence 'ChatGPT can help you reply to a personal letter without ever seeing the letter' is audible. (0s)
• lip sync: unknown. Audio is not provided; sampled frames alone cannot confirm alignment between mouth movements and any specific spoken words. (0s)

google/gemini-3.8-flash

still_image · 4/4 rubric index

• realism level 1: pass. Presents a coherent photographic human appearance rather than an illustration or obvious 3D model. (0s)
• realism level 2: pass. Plausible photographic representation of skin, hair, and lighting without excessive waxy or plastic texture. (0s)
• realism level 3: pass. Strong camera realism with natural facial lighting, realistic hair strands, and no obvious synthetic rendering artifacts. (0s)
• realism level 4: pass. Believable home-camera realism across facial details, natural skin texture, clothing folds, and ambient lighting. (0s)

openai/gpt-5.6-sol

still_image

• realism level 1: pass. The visible frame shows a coherent photographic older adult with natural human proportions, recognizable skin, hair, eyes, clothing, and a camera-like home interior rather than cartoon or overt 3D styling. (0s)
• realism level 2: pass. Skin texture and age lines, gray hair variation, eyeglass reflections, facial features, hands, clothing, and room lighting are mostly photographically plausible; no major waxy, painted, or plastic treatment is visible at this scale. (0s)
• realism level 3: pass. In the supplied frame, facial texture, hair strands, expression, hands, shadows, and indoor illumination maintain strong camera realism without a readily noticeable doll-eyed, beauty-rendered, painted-hair, or CGI-shaded appearance. (0s)
• realism level 4: unknown. Only one medium-wide sampled frame is available from the 0.05-second clip; it appears plausibly live-action, but the limited facial detail and lack of additional temporal views are insufficient to exclude subtle synthetic tells under strict close inspection. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin tone, visible wrinkles, realistic hair texture, and plausible lighting. No cartoon, illustration, or obvious 3D render artifacts are present. (0s)
• realism level 2: pass. Skin shows natural texture and pores, eyes have realistic reflections and depth, lips exhibit subtle shading and moisture, hair has varied strand detail, and lighting appears consistent with indoor ambient sources. No major waxy, painted, or plastic qualities are visible. (0s)
• realism level 3: pass. Under close inspection, the subject displays strong camera realism: eyes have natural corneal reflections and iris detail, skin shows fine wrinkles and uneven tone, hair has realistic flyaways and volume, and lighting creates plausible shadows and highlights. No doll eyes, beauty-render skin, painted hair, or CGI shading are apparent. (0s)
• realism level 4: pass. The image convincingly matches an ordinary live-action home-camera appearance: skin texture, hair detail, eyewear reflections, and ambient lighting are consistent with real-world capture. No perceptibly CGI/avatar-like eyes, lips, skin, hair, or expression are visible. Minor clutter and natural blur do not constitute synthetic tells. (0s)

google/gemini-3.8-flash

still_image

• realism level 1: pass. Clear photographic representation of an older woman in a room. (0s)
• realism level 2: pass. Plausible skin texture, hair, lighting, and facial features without obvious plastic appearance. (0s)
• realism level 3: pass. Naturalistic home-camera lighting, hair strand detail, and subtle skin folds. (0s)
• realism level 4: unknown. Extremely brief clip duration (0.05s) prevents evaluating temporal realism and natural motion. (0s)

openai/gpt-5.6-sol

still_image

• realism level 1: pass. The sampled frame shows a coherent photographic older woman in a naturally lit home-office setting, not a cartoon, illustration, or obvious 3D render. (0s)
• realism level 2: pass. Visible skin texture, eyes behind glasses, lips, gray hair, and room lighting are mostly photographically plausible, with no major waxy, painted, or plastic appearance. (0s)
• realism level 3: pass. Under inspection of the available frame, facial features, hair strands, glasses, clothing, lighting, and background retain strong camera realism without a readily noticeable doll-like or CGI-rendered feature. (0s)
• realism level 4: unknown. The sole 0.0-second sampled frame appears convincingly live-action, but the extremely brief 0.05-second clip and single-frame coverage are insufficient to exclude meaningful synthetic tells that could appear through motion or changing expression. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin texture, realistic hair strands, plausible facial features, and ambient indoor lighting consistent with a home camera. No cartoon, illustration, or obvious 3D render artifacts are present. (0s)
• realism level 2: pass. Skin shows natural variation and subtle pores; eyes have realistic reflections and depth; lips show natural texture and slight asymmetry; hair has varied strand direction and lighting response; ambient light casts soft shadows consistent with indoor environment. No major waxy, painted, or plastic appearance observed. (0s)
• realism level 3: pass. Under close inspection, the image maintains strong camera realism: no doll-like eyes, no beauty-render skin smoothing, no painted hair, no CGI shading artifacts. Minor deviations such as slight softness in hair edges or minor noise are consistent with consumer-grade home-camera footage, not synthetic tells. (0s)
• realism level 4: pass. The frame exhibits convincingly ordinary live-action home-camera appearance: natural wrinkles, realistic lighting falloff, ambient clutter (books, cables, phone stand), and no perceptibly CGI/avatar-like eyes, lips, skin, hair, or expression. Minor blur or noise is consistent with real camera capture and does not indicate synthetic origin. (0s)

google/gemini-3.8-flash

Native video/audio supplied; internal model sampling unknown · 3/4 rubric index

• realism level 1: pass. Clear photographic rendering of a human subject with realistic lighting and environment. (0, 7.2s)
• realism level 2: pass. Facial features, skin texture, and hair show plausible photographic detail without major 3D or waxy artifacts. (0, 7.2s)
• realism level 3: pass. Natural facial movements, subtle blinking, and organic head motion consistent with camera recording. (1, 5s)
• realism level 4: fail. Slight synthetic mouth articulation and static eye gaze typical of AI video generation. (2.5, 4.5s)
• identity: pass. Matches the reference character with gray hair, black glasses, and similar facial features. (0, 7.2s)
• fixed camera: pass. Steady frontal shot positioned at eye level in an office/home setting throughout the clip. (0, 7.2s)
• face stability: pass. Facial structure remains stable with no warping or distortion across the duration. (0, 7.2s)
• speech: pass. Audio clearly speaks the exact target phrase without dropouts. (0.5, 5.5s)
• lip sync: pass. Mouth movement matches the timing and cadence of the spoken audio track. (0.6, 5.4s)

openai/gpt-5.6-sol

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject consistently has a coherent photographic human appearance rather than an illustrative or overtly rendered one. (0, 3, 6.5s)
• realism level 2: pass. Skin, eyes, lips, glasses, and silver-gray hair are mostly photographically plausible, with coherent warm room lighting and no major waxy or plastic finish dominating the face. (0.5, 3, 5.5s)
• realism level 3: fail. Careful inspection shows noticeable synthetic smoothness and uniformity around the cheeks, mouth, and under-eye areas; several speaking expressions also look slightly avatar-like rather than fully natural. (1.5, 2, 4, 7.12s)
• realism level 4: fail. The face retains meaningful generated-video tells under close inspection, especially overly clean skin rendering, slightly artificial eye/eyelid appearance, and subtly synthetic mouth expressions. (1, 2.5, 4.5, 7.12s)
• identity: pass. The visible character preserves the reference's recognizable older-woman presentation, silver-gray shoulder-length wavy hair, black rectangular glasses, long narrow face, and similar eye and mouth placement. (0, 3, 6s)
• fixed camera: pass. All sampled frames maintain the same eye-level frontal medium shot and the same coherent book-lined home room, with no visible camera-angle change. (0, 3.5, 7.12s)
• face stability: pass. The same facial structure, glasses, hairline, and features persist across the sampled coverage; no clear sliding features, deformed eyes, or exposed synthetic teeth are visible. (0, 2, 4.5, 7.12s)
• speech: unknown. Only sampled visual frames are supplied; they show mouth changes but cannot establish whether the specified sentence is audible. (0, 7.12s)
• lip sync: unknown. Sampled frames at roughly half-second intervals and no observable audio evidence cannot establish alignment between mouth motion and individual audible words. (1, 2.5, 4, 7.12s)

qwen/qwen3-vl-235b-a22b-instruct

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified; visual-only subset: rejected audio inference excluded · 3/4 rubric index

• realism level 1: pass. The subject has coherent photographic human appearance: visible skin texture, natural hair strands, realistic eye reflections, and plausible lighting. No predominant illustration, cartoon, or 3D rendering artifacts are present. (0, 7.12s)
• realism level 2: pass. Face exhibits mostly plausible photographic qualities: skin shows natural tonal variation and fine wrinkles, eyes have realistic reflections and depth, lips move naturally, hair has strand-level detail and volume, and lighting is soft and consistent with indoor ambient sources. No major waxy, painted, or plastic finish is evident. (0, 7.12s)
• realism level 3: pass. Under careful inspection, only minor photographic deviations are visible: slight softness in hair edges and minimal uniformity in skin tone, but no doll-like eyes, beauty-render skin, painted hair, CGI shading, or artificial edge separation. Expression and anatomy across sampled frames appear natural and consistent. (0, 7.12s)
• realism level 4: fail. Perceptibly CG/3D-like qualities remain: skin lacks micro-texture and pore detail under close inspection, eye reflections are overly uniform and lack depth variation, and hair edges show slight haloing and lack fine strand separation. Lighting, while plausible, is too even and lacks the subtle imperfections expected from an ordinary home camera. (0, 7.12s)
• identity: pass. Character matches reference identity: silver-gray wavy shoulder-length hair, black rectangular glasses, and facial structure consistent with the reference image. Wardrobe and setting align with the fictional character’s visual identity. (0, 7.12s)
• fixed camera: pass. Eye-level frontal framing is consistent throughout. Background elements (bookshelves, lamp, curtains) remain static and coherent, indicating no camera movement or generated angle changes. (0, 7.12s)
• face stability: pass. Face remains consistent across all sampled frames: no sliding features, rubbery mouth, synthetic teeth, or eye deformation. Expression changes are natural and anatomically plausible. (0, 7.12s)

google/gemini-3.8-flash

Native video/audio supplied; internal model sampling unknown · 3/4 rubric index

• realism level 1: pass. Coherent photographic human appearance throughout. (0, 5.8s)
• realism level 2: pass. Skin texture, hair, and lighting appear mostly photographic without heavy 3D rendering. (1, 4s)
• realism level 3: pass. Minor synthetic cues only; realistic facial expressions and natural ambient lighting. (1.5, 4.5s)
• realism level 4: fail. Slightly stiff mouth articulation and subtle synthetic smoothing around jawline under scrutiny. (1.8, 3.2s)
• identity: pass. Recognizable match to reference with silver-gray hair and black rectangular glasses. (0.5, 5s)
• fixed camera: pass. Static eye-level frontal shot in a coherent room setting with no camera moves. (0, 5.8s)
• face stability: pass. Facial features, glasses, and teeth remain spatially stable without deformation. (0, 5.8s)
• speech: pass. Speaks the required sentence clearly without alterations. (0, 4.8s)
• lip sync: pass. Mouth movement matches the spoken phonemes across the audio track. (0.1, 4.6s)

openai/gpt-5.6-sol

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 3/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with natural facial anatomy, clothing, and room lighting rather than an illustrative or overt 3D-rendered look. (0, 2.5, 5.7325s)
• realism level 2: pass. Across sampled views, skin, eyes, lips, gray hair, glasses, and lighting are mostly photographically plausible without a dominant waxy, plastic, or painted finish. (0.5, 2.5, 4.5s)
• realism level 3: pass. Careful inspection of the samples shows plausible wrinkles, hair strands, reflections, blinking, and varied mouth shapes; no readily noticeable doll-like eyes, painted hair, CGI shading, or malformed sampled expression is visible. (0, 1.5, 3.5, 5s)
• realism level 4: fail. The face remains unusually pristine and uniformly lit for ordinary home-camera footage, with subtly smoothed skin and highly clean facial/glasses edges that give the sampled portrait a perceptibly generated, studio-composited finish under close inspection. (1, 3, 5.7325s)
• identity: pass. The visible character matches the reference's recognizable older feminine face presentation, shoulder-length silver-gray wavy hair, and black rectangular glasses. (0, 2.5, 5.7325s)
• fixed camera: pass. All supplied samples retain the same eye-level frontal medium framing and the same coherent book-lined home room, with no visible camera-angle change. (0, 2.5, 5.7325s)
• face stability: pass. The same facial structure, glasses, hairline, and features persist across the sampled coverage; varied mouth and eyelid states remain anatomically coherent without visible sliding, malformed teeth, or eye deformation. (0, 0.5, 1.5, 3.5, 5.7325s)
• speech: unknown. Only sampled frames are supplied here, so the requested audible sentence cannot be verified. (0, 5.7325s)
• lip sync: unknown. Sampled still frames show changing mouth shapes but cannot establish alignment with audible words or continuous temporal synchronization. (0, 1.5, 3.5, 5.7325s)

qwen/qwen3-vl-235b-a22b-instruct

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject has coherent photographic human appearance: visible skin texture, natural hair strands, realistic eye reflections, and plausible lighting/shadow transitions. No predominant illustration, cartoon, or obvious 3D rendering cues are present. (0, 5.7325s)
• realism level 2: pass. Face exhibits mostly plausible photographic qualities: skin shows natural wrinkles and pores, eyes have realistic highlights and depth, lips and hair appear textured and non-painted. Minor synthetic cues (e.g., slight uniformity in hair strand rendering) are present but do not dominate. (0, 5.7325s)
• realism level 3: fail. Noticeable synthetic cues remain: eyes appear slightly doll-like with overly uniform sclera and corneal reflections; skin has subtle airbrushed smoothness in cheek and forehead regions; hair edges show slight artificial separation from background. Expression changes (e.g., 0.0 to 1.5s) appear anatomically plausible but lack micro-textural fidelity expected in high realism. (0, 5.7325s)
• realism level 4: fail. Perceptibly CG-like cues persist: eyes lack natural iris microstructure and tear film variation; lips show slightly plastic sheen during motion; lighting is uniformly soft without realistic specular highlights or shadow gradients. These deviations prevent convincing ordinary home-camera appearance under close inspection. (0, 5.7325s)
• identity: pass. Visible features match reference: silver-gray wavy shoulder-length hair, black rectangular glasses, hazel-brown eyes, and overall facial structure consistent with reference image. Wardrobe (dark cardigan over patterned blouse) also aligns. (0, 5.7325s)
• fixed camera: pass. Eye-level frontal framing remains consistent across all frames; background elements (bookshelves, lamp, curtains) show no perspective or parallax changes. Room appears ordinary with natural clutter and lighting, no generated camera movement. (0, 5.7325s)
• face stability: pass. Face remains consistent in structure across all frames; no sliding features, rubbery mouth deformation, or synthetic teeth. Expression changes (e.g., smiling at 0.0s, pursed lips at 1.5s, open mouth at 3.0s) are anatomically coherent without distortion. (0, 5.7325s)

google/gemini-3.8-flash

Native video/audio supplied; internal model sampling unknown · 3/4 rubric index

• realism level 1: pass. Shows coherent photographic appearance rather than an illustration or obvious 3D render. (1.5s)
• realism level 2: pass. Natural photographic textures on skin, hair, and clothing without dominant plastic finish. (2.5s)
• realism level 3: pass. Lighting and skin details remain largely plausible with subtle facial motion. (3.5s)
• realism level 4: fail. Slight synthetic mouth movement and digital smoothing visible under close inspection. (2s)
• identity: pass. Features match reference with gray wavy hair, glasses, and facial structure. (0.5s)
• fixed camera: pass. Camera maintains a fixed eye-level position in a room setting throughout. (3s)
• face stability: pass. Facial geometry and features stay stable without warping or sliding. (4s)
• speech: pass. Audible sentence matches the requested phrase accurately. (2.5s)
• lip sync: pass. Mouth shapes correspond well with the spoken audio track. (3s)

qwen/qwen3-vl-235b-a22b-instruct

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject has coherent photographic human appearance: skin texture, hair strands, wrinkles, and lighting are consistent with a photograph rather than illustration or cartoon. No predominant synthetic rendering artifacts are visible. (0, 5.92s)
• realism level 2: pass. Face shows mostly plausible photographic skin, eyes, lips, and hair. Lighting appears natural with soft shadows. Minor synthetic cues (e.g., slight uniformity in hair highlights) are present but do not dominate; no major waxy, painted, or plastic finish observed. (0, 5.92s)
• realism level 3: fail. Under close inspection, minor synthetic cues remain: skin lacks micro-pores and fine texture variation; hair has uniform wave pattern and sheen; eye reflections are slightly too consistent across frames; lighting is overly even with no natural falloff or specular highlights. These are noticeable deviations from strong camera realism. (0, 5.92s)
• realism level 4: fail. Perceptibly CG-like qualities persist: eyes have a doll-like gloss and uniform reflection; skin appears airbrushed with no natural pores or imperfections; hair lacks individual strand variation; lighting lacks dynamic shadow transitions. These are meaningful synthetic tells under close inspection. (0, 5.92s)
• identity: pass. The character matches the reference: silver-gray wavy hair, black rectangular glasses, and facial structure align with the approved fictional identity. Clothing and expression are consistent with the reference gallery. (0, 5.92s)
• fixed camera: pass. Framing is eye-level frontal with no camera movement or angle changes. Background (bookshelves, lamp, curtains) remains static and coherent, consistent with an ordinary home-camera setup. (0, 5.92s)
• face stability: pass. Face remains consistent across all sampled frames; no sliding features, rubbery mouth, synthetic teeth, or eye deformation observed. Expressions change naturally with speech but anatomy is stable. (0, 5.92s)
• speech: unknown. Audio is not provided in the supplied media. Speech cannot be confirmed from sampled frames alone. (0s)
• lip sync: unknown. Audio is not provided; lip movement cannot be aligned with speech. Sampled frames alone cannot prove or disprove lip sync. (0s)

openai/gpt-5.6-sol

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with consistent facial anatomy, clothing, and room lighting across sampled frames. (0, 3, 5.92s)
• realism level 2: pass. Skin, eyes, lips, silver hair, glasses, and illumination appear mostly photographic; no major waxy, plastic, painted, or overtly 3D finish dominates. (0.5, 2.5, 4.5s)
• realism level 3: fail. Careful inspection shows noticeable generation-like polish: facial skin and lighting are unusually uniform, hair strands have a rendered regularity, and open-mouth/teeth appearances look slightly artificial. (1, 3.5, 4.5s)
• realism level 4: fail. The highly polished, uniformly lit face, crisply separated hair, and slightly synthetic mouth/teeth prevent a convincingly ordinary live-action home-camera appearance under close inspection. (1, 3.5, 4.5s)
• identity: pass. The visible character retains the reference's recognizable older-woman presentation, shoulder-length wavy silver-gray hair, black rectangular glasses, and similar facial proportions. (0, 3, 5.92s)
• fixed camera: pass. All supplied frames retain the same eye-level frontal composition and coherent book-lined home room, with no visible camera-angle change. (0, 3, 5.92s)
• face stability: pass. Across sampled frames, facial proportions, glasses, eyes, and hair remain consistent; varied mouth shapes do not show clear feature sliding or deformation in the visible samples. (0, 1.5, 3.5, 5.92s)
• speech: unknown. Only sampled visual frames are supplied here; they cannot establish whether the specified sentence is audible. (0, 5.92s)
• lip sync: unknown. Sampled frames at 0.5-second intervals and no assessable audio alignment cannot establish word-level mouth synchronization or continuous motion between samples. (0, 1, 3, 5.92s)

openai/gpt-5.6-sol

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance, with consistent facial anatomy, clothing, glasses, hair, and room lighting across sampled frames. (0, 3.5, 7.12s)
• realism level 2: pass. Skin, eyes, lips, silver-gray hair, glasses, and illumination are mostly photographically plausible; no major waxy, plastic, painted, or overtly 3D finish dominates the sampled views. (0, 2.5, 5s)
• realism level 3: fail. Careful inspection shows noticeable synthetic polish: facial skin is unusually smooth and evenly rendered, hair strands and edges have a generated look, and the broad smile/teeth appear overly uniform. (1, 4.5, 6.5s)
• realism level 4: fail. The repeated beauty-render smoothness of the face, highly uniform lighting and texture, and polished hair and teeth remain meaningful visible synthetic tells rather than an unequivocal ordinary home-camera recording. (1, 4.5, 6.5s)
• identity: pass. The visible character retains the reference's recognizable older feminine face, shoulder-length wavy silver-gray hair, dark rectangular glasses, and similar facial proportions. (0, 3, 6.5s)
• fixed camera: pass. All supplied frames maintain the same eye-level frontal composition and coherent study setting, with unchanged desk, window, shelves, curtains, and lamp placement. (0, 3.5, 7.12s)
• face stability: pass. The same face and glasses remain spatially consistent through the sampled expressions; no clear sliding features, deformed eyes, rubbery mouth, or malformed visible teeth appear in these frames. (0, 1.5, 3.5, 5.5, 7.12s)
• speech: unknown. Only sampled visual frames are supplied; they show changing mouth shapes but cannot establish that the specified sentence is audible. (0.5, 3.5, 7.12s)
• lip sync: unknown. Sampled frames at roughly half-second intervals and no assessable audio cannot establish alignment between mouth motion and individual audible words. (0.5, 2.5, 5.5, 7.12s)

google/gemini-3.8-flash

Native video/audio supplied; internal model sampling unknown · 2/4 rubric index

• realism level 1: pass. Photographic human likeness is maintained rather than an overt cartoon or 3D render. (1s)
• realism level 2: pass. Visible skin texture, hair, and lighting appear generally plausible without overwhelming plastic artifacts. (2.5s)
• realism level 3: fail. Mouth movement and teeth deformation exhibit noticeable synthetic cues during speech. (3.8s)
• realism level 4: fail. Slightly artificial rendering around animated facial features prevents convincing live-action realism. (4.5s)
• identity: pass. Features match the reference subject with gray wavy hair and black rectangular glasses. (0.5s)
• fixed camera: pass. Camera remains frontal, static, and eye-level within a home interior setup. (2s)
• face stability: fail. Lower facial structure and mouth region display warping during dynamic speech expressions. (4s)
• speech: pass. Spoken sentence matches the prompt clearly without missing words. (3.5s)
• lip sync: pass. Lip movements generally correspond to the timing of the audible speech track. (3s)

qwen/qwen3-vl-235b-a22b-instruct

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with recognizable facial features, skin texture, hair, and clothing. No predominant illustration, cartoon, or obvious 3D rendering artifacts are visible across sampled frames. (0, 7.12s)
• realism level 2: pass. The visible face exhibits mostly plausible photographic skin, eyes, lips, and hair. Lighting appears natural with soft shadows. Minor synthetic cues may exist but do not dominate; no major waxy, painted, or plastic finish is evident. (0, 7.12s)
• realism level 3: fail. Under careful inspection, minor synthetic cues remain: skin texture appears overly smooth and uniform, lacking natural pore variation; eye reflections and lip contours have slight CGI-like consistency across expressions; hair strands show minimal individual variation and edge separation from background. (0, 7.12s)
• realism level 4: fail. Perceptibly CG/3D-like qualities persist: skin lacks micro-texture and natural imperfections, eyes have uniform glossy reflection, and hair exhibits artificial uniformity. Expression changes are visible but lack the subtle anatomical micro-movements expected from a live home-camera recording. (0, 7.12s)
• identity: pass. The character matches the fictional reference: silver-gray wavy hair, black rectangular glasses, and overall facial structure consistent with the approved identity board. No real person is identified. (0, 7.12s)
• fixed camera: pass. The framing is eye-level and frontal throughout all sampled frames. Background elements (bookshelf, desk, lamp, window) remain spatially coherent with no camera angle changes or generated motion. (0, 7.12s)
• face stability: pass. The same face is present throughout all sampled frames. No sliding features, rubbery mouth, synthetic teeth, or eye deformation are observed. Expressions change naturally without structural instability. (0, 7.12s)

openai/gpt-5.6-sol

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with recognizable skin, hair, eyes, glasses, clothing, and room lighting rather than a predominantly illustrated or overtly rendered look. (0, 2.5, 5.7325s)
• realism level 2: pass. Across the sampled frames, facial skin, wrinkles, eyes, lips, gray hair, and illumination are mostly photographically plausible without a major waxy, plastic, or painted finish dominating. (0, 2.5, 4.5s)
• realism level 3: fail. Careful inspection shows noticeable synthetic polish: facial texture and lighting remain unusually even, hair contours are overly pristine, and several open-mouth expressions look avatar-like rather than ordinary camera capture. (0, 3.5, 4s)
• realism level 4: fail. The highly controlled facial detail, pristine hair separation, uniformly polished room image, and synthetic-looking mouth expressions are meaningful visible tells, so it is not fully convincing as ordinary live-action under close inspection. (0, 1.5, 3.5, 4.5s)
• identity: pass. The visible character consistently resembles the fictional reference through an older feminine face, shoulder-length silver-gray wavy hair, black rectangular glasses, and similar facial proportions. (0, 2.5, 5.7325s)
• fixed camera: pass. Every supplied frame retains the same eye-level frontal medium framing and a spatially coherent home-office background, with no visible camera-angle change. (0, 3, 5.7325s)
• face stability: pass. The same face and glasses persist across all sampled frames; eyes, nose, jaw, and mouth remain spatially consistent, with no clear feature sliding, malformed teeth, or eye deformation visible in the samples. (0, 1.5, 3.5, 5.7325s)
• speech: unknown. Only sampled visual frames are supplied here; they show changing mouth shapes but cannot establish that the specified sentence is audible. (0, 2.5, 5s)
• lip sync: unknown. Sampled frames at roughly half-second intervals cannot reliably establish alignment between individual audible words and mouth motion, and no usable audio evidence is supplied in this coverage. (0, 1.5, 3.5, 5s)

google/gemini-3.8-flash

Native video/audio supplied; internal model sampling unknown · 2/4 rubric index

• realism level 1: pass. Photographic human appearance maintained without cartoon or overt 3D rendering stylization. (0s)
• realism level 2: pass. Mostly plausible skin texture, hair, and lighting without dominant plastic or painted finish. (1.5s)
• realism level 3: fail. Unnatural lip deformations and slightly rigid, synthetic ocular focus visible during speech. (1.2, 3.4s)
• realism level 4: fail. Perceptible generative animation artifacts around the mouth and lower jaw area during speaking. (2s)
• identity: pass. Matches the reference character's features, wavy silver-gray hair, and black rectangular glasses. (0.5s)
• fixed camera: pass. Static frontal eye-level framing in a home room setting without camera movement. (0s)
• face stability: pass. Face structure remains consistent throughout without warping or drifting features. (3s)
• speech: pass. Fully audible and intelligible prompt sentence matches the requested dialogue accurately. (0, 5s)
• lip sync: pass. Lip movements correspond reasonably well with the spoken audio syllables. (1, 4s)

qwen/qwen3-vl-235b-a22b-instruct

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, skin texture, and clothing; no predominant illustration, cartoon, or 3D rendering artifacts are visible across sampled frames. (0, 5.7325s)
• realism level 2: pass. Face shows mostly plausible photographic skin, eyes, lips, and hair under natural lighting; minor synthetic cues may exist but are not dominant. No major waxy, painted, or plastic finish is observable. (0, 5.7325s)
• realism level 3: fail. Under careful inspection, minor synthetic cues remain: skin texture appears overly smooth and uniform, lacking natural pore variation; eye reflections and eyelid transitions show subtle doll-like rendering; hair strands lack individual strand realism and exhibit slight haloing against background. (0, 5.7325s)
• realism level 4: fail. Perceptibly CG-like eyes with uniform iris shading and artificial catchlights; lips show overly consistent texture and motion; skin lacks micro-variation expected in live-action home-camera footage. These cues prevent convincingly ordinary appearance under close inspection. (0, 5.7325s)
• identity: pass. Character matches reference identity: silver-gray wavy hair, black rectangular glasses, and facial structure consistent with the provided fictional reference image. (0, 5.7325s)
• fixed camera: pass. Eye-level frontal framing remains consistent; background elements (bookshelf, lamp, window) show no camera movement or angle changes across sampled frames. (0, 5.7325s)
• face stability: pass. Same face is visible throughout; no sliding features, rubbery mouth deformation, synthetic teeth, or eye distortion observed in any sampled frame. (0, 5.7325s)

openai/gpt-5.6-sol

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 2/4 rubric index

• realism level 1: pass. The woman, face, hair, glasses, clothing, and room have a coherent photographic appearance despite a conspicuous torso artifact. (0, 2.5, 5.92s)
• realism level 2: pass. Visible facial skin, eyes, lips, gray hair, and lighting are mostly photographically plausible; the major neon distortion is concentrated on the sweater rather than dominating the face. (0.5, 2.5, 4.5s)
• realism level 3: fail. A large bright pink/green pixelated form with vertical streaking repeatedly covers the chest and reaches the chin/mouth area, while some mouth shapes and teeth look unnaturally rendered. (0, 1, 3.5, 5.5s)
• realism level 4: fail. The persistent luminous torso/chin corruption is an unmistakable synthetic tell and prevents an ordinary live-action home-camera appearance under close inspection. (0, 2, 4, 5.92s)
• identity: pass. The visible character consistently resembles the fictional reference through an older face, shoulder-length silver-gray wavy hair, black rectangular glasses, and similar facial proportions. (0, 3, 5.92s)
• fixed camera: pass. All sampled frames retain the same eye-level frontal medium shot and coherent study setting, with the window, desk, shelves, curtains, and lamp fixed in place. (0, 3, 5.92s)
• face stability: fail. Although overall identity remains stable, vertical distortion crosses the lower face and sampled speaking poses show unnaturally striped lips/teeth and malformed mouth rendering. (0, 1, 3.5, 5s)
• speech: unknown. Only sampled visual frames are supplied; they cannot establish whether the requested sentence is audible. (0, 5.92s)
• lip sync: unknown. Sampled frames show varying mouth positions but provide neither reliable continuous timing nor accessible audio evidence needed to judge alignment with words. (0, 1, 3.5, 5.92s)

google/gemini-3.8-flash

Native video/audio supplied; internal model sampling unknown · 1/4 rubric index

• realism level 1: pass. The character presents a coherent photographic human appearance overall. (0s)
• realism level 2: fail. Severe digital rendering glitch and bright pixelated artifact overlaying the chest area. (1.5s)
• realism level 3: fail. Obvious synthetic digital visual artifact on the body and somewhat artificial eye rendering. (2.5s)
• realism level 4: fail. Prominent synthetic visual corruption and unnatural head motion dynamics. (3.5s)
• identity: pass. Features match reference with gray wavy hair, glasses, and older woman's likeness. (0.5s)
• fixed camera: pass. Camera maintains a steady, eye-level frontal shot throughout. (3s)
• face stability: pass. Facial structure remains consistent without warping or feature drift. (4s)
• speech: pass. The audible spoken sentence matches the requested target sentence. (2s)
• lip sync: pass. Mouth shapes generally track the spoken words across the clip. (2.5s)

qwen/qwen3-vl-235b-a22b-instruct

2fps frames and end sample at768px; continuous motion/audio/lip sync cannot be certified · 1/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with recognizable facial features, hair, and clothing. It does not resemble a cartoon, illustration, or obvious 3D render. (0s)
• realism level 2: fail. Major synthetic cues dominate: skin has an overly smooth, waxy finish with uniform lighting and lack of microtexture; hair appears painted with uniform strands; eyes lack realistic specular highlights and depth; overall appearance is doll-like rather than photographic. (0s)
• realism level 3: fail. Noticeable synthetic cues remain: eyes are doll-like with flat irises and no realistic wetness; skin lacks pores and natural shading; hair has artificial edge separation from background; lighting is implausibly uniform across face and torso. (0s)
• realism level 4: fail. Perceptibly CG/3D-avatar-like eyes, lips, skin, and hair; uniform lighting and lack of natural micro-movements or imperfections; overall appearance does not resemble an ordinary live-action home-camera recording. (0s)
• identity: pass. The character displays the requested fictional reference identity: silver-gray wavy hair, black rectangular glasses, and facial structure consistent with the reference board. (0s)
• fixed camera: pass. Eye-level frontal framing remains consistent; background elements (bookshelf, lamp, window) show no camera movement or angle changes; room appears ordinary and coherent. (0s)
• face stability: fail. Face exhibits synthetic deformation: mouth shape and teeth appear rubbery and inconsistent across frames; lips lack natural muscle definition; eye shape subtly warps during expression changes. (0, 1, 2, 3, 4, 5s)

google/gemini-3.8-flash

One resized still preview; motion/audio not reviewed · 4/4 rubric index

• realism level 1: pass. Presents a coherent photographic human appearance without cartoon or 3D rendering stylization. (0s)
• realism level 2: pass. Facial features, skin texture, lighting, and hair appear plausibly photographic without heavy synthetic cues. (0s)
• realism level 3: pass. Shows natural skin details, eyes, hair strands, and home-office ambient lighting without prominent CGI artifacts. (0s)
• realism level 4: pass. Appears convincingly like an ordinary live-action webcam capture with realistic texture, depth of field, and lighting. (0s)
• identity: pass. Matches the reference character's features, silver-gray hair, black glasses, and dark blue sweater. (0s)
• fixed camera: pass. Stable eye-level frontal desk framing set in a coherent home office environment. (0s)

openai/gpt-5.6-sol

One resized still preview; motion/audio not reviewed · 2/4 rubric index

• realism level 1: pass. The woman, clothing, desk, and room form a coherent photographic-looking scene rather than a predominantly illustrated or cartoon image. (0s)
• realism level 2: pass. Visible skin, eyes, lips, gray hair, glasses, and illumination are mostly photographically plausible, without a major waxy, plastic, or painted finish dominating the face. (0s)
• realism level 3: fail. Careful inspection shows noticeable synthetic polish: facial skin and lighting are unusually smooth and uniform, the eyes have a slightly doll-like clarity, and some hair and glasses boundaries look artificially clean. (0s)
• realism level 4: fail. The highly uniform face rendering, polished eyes, and overly clean hair/glasses edges remain perceptible synthetic tells, so it is not fully convincing as ordinary home-camera footage under close inspection. (0s)
• identity: pass. The visible woman resembles the fictional reference through her older facial appearance, shoulder-length silver-gray wavy hair, black rectangular glasses, and similar facial proportions. (0s)
• fixed camera: pass. The single supplied frame has eye-level, centered frontal framing in a coherent home office with desk, monitor, lamps, windows, curtains, and bookshelves; angle changes are inapplicable to this still. (0s)

qwen/qwen3-vl-235b-a22b-instruct

One resized still preview; motion/audio not reviewed · 3/4 rubric index

• realism level 1: pass. The image shows a coherent photographic human appearance: a woman with recognizable facial structure, skin tone, hair, and clothing under natural lighting. No predominant illustration, cartoon, or obvious 3D rendering artifacts are visible. (0s)
• realism level 2: pass. The visible face exhibits mostly plausible photographic qualities: skin texture appears natural with subtle variation, eyes have realistic reflections and depth, lips show natural shading, hair has strand-level detail and volume, and lighting is consistent with indoor ambient sources. Minor synthetic cues (e.g., slight uniformity in hair highlights) do not dominate. (0s)
• realism level 3: pass. Under careful inspection, only minor photographic deviations are present: slight softness in skin pores and minor edge haloing around hair against the background. No doll-like eyes, beauty-render skin, painted hair, CGI shading, or implausibly uniform lighting are evident. Anatomy and expression appear natural for a still frame. (0s)
• realism level 4: fail. Perceptibly CG/3D-like qualities remain: the skin has a slightly airbrushed, overly smooth finish lacking microtexture; the eye reflections are too uniform and lack natural corneal imperfections; hair strands show repetitive pattern artifacts; lighting on the face is unnaturally even for a home setting. These cues prevent convincing ordinary live-action home-camera appearance under close inspection. (0s)
• identity: pass. The character matches the fictional reference: silver-gray wavy hair, black rectangular glasses, and facial structure consistent with the approved identity board. Wardrobe (dark sweater) and setting (home office) also align. (0s)
• fixed camera: pass. The framing is eye-level and frontal, consistent with an ordinary home camera. The background (bookshelf, lamp, window) is coherent and static. No camera angle changes or generated motion are present in the single frame. (0s)

google/gemini-3.8-flash

One resized still preview; motion/audio not reviewed · 3/4 rubric index

• realism level 1: pass. Image exhibits photographic human likeness without cartoon or illustration styling. (0s)
• realism level 2: pass. Skin texture, hair strands, and lighting appear plausibly photographic without heavy CGI finish. (0s)
• realism level 3: pass. Facial features, eye reflections, and ambient room lighting show no major synthetic artifacts. (0s)
• realism level 4: fail. Slight digital smoothing around the face and hair edges reveals synthetic generation under close inspection. (0s)
• identity: pass. Subject matches reference with wavy silver-gray hair, facial structure, and dark glasses. (0s)
• fixed camera: pass. Framing is an eye-level frontal shot in an ordinary home office setting. (0s)

qwen/qwen3-vl-235b-a22b-instruct

One resized still preview; motion/audio not reviewed · 4/4 rubric index

• realism level 1: pass. The image shows a coherent photographic human appearance: a woman with recognizable facial features, skin texture, hair, and clothing under natural lighting. No predominant illustration, cartoon, or obvious 3D rendering artifacts are visible. (0s)
• realism level 2: pass. The face exhibits mostly plausible photographic qualities: skin shows natural tonal variation and fine wrinkles, eyes have realistic reflections and depth, lips have subtle texture, hair has natural wave and strand detail, and lighting appears ambient and consistent with indoor environment. No major waxy, painted, or plastic finish is evident. (0s)
• realism level 3: pass. Under careful inspection, only minor photographic deviations are present (e.g., slight softness in hair edges, minimal noise). No readily noticeable doll-like eyes, beauty-render skin, painted hair, CGI shading, or artificial edge separation. Lighting and anatomy appear plausible for a still photo. No motion to evaluate. (0s)
• realism level 4: pass. The image convincingly resembles an ordinary live-action home-camera photo: skin pores and wrinkles are visible, hair strands vary naturally, glasses reflect ambient light plausibly, and the background (bookshelves, lamp, curtains) shows depth and natural clutter. No perceptibly CG/3D/avatar-like cues in eyes, lips, skin, hair, lighting, or expression are present in the still frame. (0s)
• identity: pass. The subject matches the fictional reference: silver-gray wavy shoulder-length hair and black rectangular glasses are clearly visible. Facial structure and expression align with the reference images in the approved identity board. (0s)
• fixed camera: pass. The framing is eye-level and frontal, consistent with a fixed home camera. The background shows a coherent room with bookshelves, curtains, and a lamp, with no evidence of camera movement or generated angle changes. Set is ordinary, not overly polished. (0s)

openai/gpt-5.6-sol

One resized still preview; motion/audio not reviewed · 3/4 rubric index

• realism level 1: pass. The still has a coherent photographic human appearance with anatomically consistent facial features, hair, clothing, and room lighting rather than a predominantly illustrated or overtly 3D-rendered look. (0s)
• realism level 2: pass. Visible skin texture, wrinkles, eyes, lips, silver-gray hair, glasses, and warm window/lamp illumination are mostly photographically plausible without a major waxy or plastic finish. (0s)
• realism level 3: pass. Careful inspection shows generally natural facial texture, hair detail, eyeglass integration, anatomy, and coherent directional light; no readily noticeable doll-like eyes, painted hair, or artificial edge separation dominates the still. (0s)
• realism level 4: fail. The face remains unusually polished and symmetrical, with slightly smooth skin transitions and very clean eye/eyeglass rendering that create a subtle generated portrait impression under close inspection rather than entirely ordinary camera capture. (0s)
• identity: pass. The visible character retains the reference's recognizable older feminine face, shoulder-length wavy silver-gray hair, black rectangular glasses, and similar central facial proportions. (0s)
• fixed camera: pass. The single supplied frame uses an eye-level, frontal, centered home-camera composition in a coherent room with bookshelves, windows, curtains, and a lamp; no angle change is present in the available still. (0s)

google/gemini-3.8-flash

One resized still preview; motion/audio not reviewed · 4/4 rubric index

• realism level 1: pass. Image exhibits clear photographic features and coherent human anatomy. (0s)
• realism level 2: pass. Plausible photographic skin, hair, and lighting without major 3D or waxy rendering artifacts. (0s)
• realism level 3: pass. Natural texture on skin, hair strands, and fabric with realistic depth of field. (0s)
• realism level 4: pass. Convincing home interior scene with natural lighting, subtle imperfections, and no obvious synthetic cues. (0s)
• identity: pass. Matches the reference character's age, wavy silver-gray hair, facial structure, and dark rectangular glasses. (0s)
• fixed camera: pass. Eye-level frontal composition showing a coherent home office setting. (0s)

openai/gpt-5.6-sol

One resized still preview; motion/audio not reviewed · 2/4 rubric index

• realism level 1: pass. The still has a coherent photographic human appearance with anatomically consistent facial features, hands, clothing, and room lighting rather than predominantly illustrative or cartoon rendering. (0s)
• realism level 2: pass. Visible skin texture, wrinkles, eyes behind the lenses, lips, gray hair, and warm directional illumination are mostly photographically plausible without a major waxy or plastic finish. (0s)
• realism level 3: fail. Careful inspection shows noticeable synthetic polish: facial texture and lighting are unusually uniform, hair strands and glasses edges look overly clean, and the hands have softened, slightly malformed detail. (0s)
• realism level 4: fail. The portrait does not remain fully convincing under close inspection because the face, hair, eyewear boundaries, and hands retain perceptible generated-image smoothness and detail irregularities. (0s)
• identity: pass. The subject visibly matches the reference's defining fictional identity cues: older feminine face, shoulder-length wavy silver-gray hair, black rectangular glasses, and similar facial proportions. (0s)
• fixed camera: pass. The sole frame uses a stable, near-eye-level frontal composition in a coherent home-office/living-room setting; as a still, it contains no generated angle change. (0s)

qwen/qwen3-vl-235b-a22b-instruct

One resized still preview; motion/audio not reviewed · 4/4 rubric index

• realism level 1: pass. The image shows a coherent photographic human appearance: a woman with visible skin texture, hair strands, clothing weave, and natural lighting/shadows. It does not resemble illustration, cartoon, or obvious 3D rendering. (0s)
• realism level 2: pass. The face exhibits plausible photographic skin texture, natural eye highlights, realistic lips, and wavy hair with varied lighting. No major waxy, painted, or plastic finish is visible. Minor synthetic cues (if any) do not dominate. (0s)
• realism level 3: pass. Under careful inspection, the image shows minor photographic deviations (e.g., slight softness in hair edges, uniform lighting), but no readily noticeable doll-like eyes, beauty-render skin, painted hair, CGI shading, or artificial edge separation. Anatomy and expression appear plausible. (0s)
• realism level 4: pass. The image convincingly resembles an ordinary live-action home-camera capture: natural skin pores, realistic hair strands, coherent lighting from window and lamp, and no perceptibly CG/3D/avatar-like eyes, lips, skin, hair, lighting, or expression. Minor blur and depth-of-field are consistent with consumer cameras. (0s)
• identity: pass. The subject matches the fictional reference: silver-gray wavy hair, black rectangular glasses, and facial features consistent with the approved identity board. No real person is identified. (0s)
• fixed camera: pass. The framing is eye-level and frontal, showing a coherent home office setting with desk, lamp, window, and bookshelf. No camera angle changes are present (still image). (0s)

google/gemini-3.8-flash

One resized still preview; motion/audio not reviewed

• realism level 1: pass. Image exhibits a photographic human appearance with realistic skin texture and environment. (0s)
• realism level 2: pass. Face displays natural skin details, believable lighting, and lacks severe 3D or painted artifacts. (0s)
• realism level 3: pass. Eyes, hair strands, and soft directional home lighting appear consistent with minimal synthetic cues. (0s)
• realism level 4: unknown. A single still frame is insufficient to fully confirm absence of subtle generative tells across motion. (0s)
• identity: pass. Subject matches reference with wavy silver-gray hair, rectangular black glasses, and similar facial features. (0s)
• fixed camera: pass. Eye-level frontal framing seated in a coherent home office setting. (0s)

openai/gpt-5.6-sol

One resized still preview; motion/audio not reviewed · 2/4 rubric index

• realism level 1: pass. The subject has a coherent photographic human appearance with recognizable facial anatomy, hair, glasses, clothing, and room lighting rather than a predominantly illustrated or cartoon presentation. (0s)
• realism level 2: pass. The face is mostly photographically plausible, with visible age lines, skin variation, eye reflections, individual hair strands, and coherent illumination; no major waxy, plastic, or painted finish dominates. (0s)
• realism level 3: fail. Careful inspection shows noticeable synthetic polish: unusually smooth and uniform facial texture, glassy symmetrical eyes, highly sculpted hair strands, and clean subject/background edge separation. (0s)
• realism level 4: fail. The portrait does not remain fully convincing under close inspection because the face, eyes, hair, and evenly balanced lighting retain a perceptibly generated, beauty-rendered quality. (0s)
• identity: pass. The image preserves the reference character's recognizable older face, center-parted silver-gray wavy hair, black rectangular glasses, and gentle neutral expression. (0s)
• fixed camera: pass. The single supplied frame uses an eye-level, centered frontal composition in a coherent home office with desk, shelves, window, curtains, and lamp; the still image provides no conflicting camera angle. (0s)

qwen/qwen3-vl-235b-a22b-instruct

One resized still preview; motion/audio not reviewed · 4/4 rubric index

• realism level 1: pass. The image shows a coherent photographic human appearance: a woman with defined facial features, hair, clothing, and background objects (bookshelves, lamp, window) that resemble real-world photography rather than illustration, cartoon, or obvious 3D rendering. (0s)
• realism level 2: pass. The visible face exhibits mostly plausible photographic qualities: skin shows natural texture and lighting, eyes have realistic catchlights and depth, lips have subtle shading, hair has strand-level variation, and lighting appears consistent with indoor ambient + lamp source. Minor synthetic cues may exist but do not dominate. (0s)
• realism level 3: pass. Under careful inspection, only minor photographic deviations are visible. No readily noticeable doll-like eyes, beauty-render skin, painted hair, CGI shading, or artificial edge separation. Lighting is plausible with soft shadows and highlights consistent with a home office setting. (0s)
• realism level 4: pass. The image convincingly mimics an ordinary live-action home-camera appearance. No meaningful visible synthetic tells: eyes, lips, skin, hair, and lighting appear naturally photographed. Wrinkles, pores, and ambient blur are consistent with a real camera capture, not added post-processing. Expression and anatomy are plausible for a still frame. (0s)
• identity: pass. The subject matches the fictional reference: silver-gray wavy shoulder-length hair and black rectangular glasses are clearly visible. Facial structure and expression align with the reference identity board’s visual identity section. (0s)
• fixed camera: pass. The framing is eye-level and frontal, consistent with a fixed home camera. Background elements (bookshelves, desk, lamp, window) form a coherent room setting with no evidence of generated camera angle changes or dynamic motion. (0s)

google/gemini-3.8-flash

One resized still preview; motion/audio not reviewed · 3/4 rubric index

• realism level 1: pass. Coherent photographic human appearance present. (0s)
• realism level 2: pass. Plausible skin texture, hair, and lighting without dominant 3D finish. (0s)
• realism level 3: pass. Subtle natural lighting and skin texture with minimal digital rendering cues. (0s)
• realism level 4: fail. Slight artificial smoothness and symmetry around eye and skin rendering under close scrutiny. (0s)
• identity: pass. Matches reference with silver-gray wavy hair, dark glasses, and similar facial features. (0s)
• fixed camera: pass. Eye-level frontal shot in a coherent, ordinary living room setting. (0s)

openai/gpt-5.6-sol

One resized still preview; motion/audio not reviewed

• realism level 1: pass. The still shows a coherent photographic human face and body with natural-looking room detail, rather than predominantly illustration, cartoon, or obvious 3D rendering. (0s)
• realism level 2: pass. Visible skin texture, wrinkles, eyes, lips, gray hair, glasses, and warm indoor lighting are mostly photographically plausible without a major waxy or plastic finish. (0s)
• realism level 3: pass. Careful inspection shows largely natural facial anatomy, age detail, hair variation, reflections, and coherent lighting; no readily noticeable doll-like eyes, painted hair, or strong artificial edge separation is visible. (0s)
• realism level 4: unknown. The single supplied frame is highly photographic, but one still at 0.0 seconds provides insufficient evidence to establish that there is no meaningful synthetic tell under broader close inspection. (0s)
• identity: pass. The visible woman matches the reference's salient fictional facial presentation, including silver-gray shoulder-length wavy hair, black rectangular glasses, mature facial features, and similar face shape. (0s)
• fixed camera: pass. The still uses eye-level, frontal, centered home-camera framing in a coherent living room with sofa, shelves, windows, plants, and lamps; as a still, it shows no generated angle change. (0s)

qwen/qwen3-vl-235b-a22b-instruct

One resized still preview; motion/audio not reviewed · 4/4 rubric index

• realism level 1: pass. The image displays a coherent photographic human appearance: a woman with recognizable facial features, skin texture, hair, and clothing under natural lighting. No predominant illustration, cartoon, or obvious 3D rendering cues are present. (0s)
• realism level 2: pass. The visible face exhibits mostly plausible photographic qualities: skin shows natural variation and fine lines, eyes have realistic catchlights and depth, lips are not overly smooth or painted, hair has natural wave and strand detail, and lighting appears consistent with indoor ambient sources. Minor synthetic cues (if any) do not dominate. (0s)
• realism level 3: pass. On close inspection, the image shows only minor photographic deviations: skin pores and wrinkles are visible, hair strands are individually defined, eye reflections are consistent, and lighting falls naturally across the face and background. No doll-like eyes, beauty-render skin, painted hair, or artificial edge separation is apparent. (0s)
• realism level 4: pass. The image convincingly resembles an ordinary live-action home-camera capture: skin texture, hair detail, glasses reflection, and ambient lighting are consistent with a real photograph. No perceptible CG/3D/avatar-like eyes, lips, skin, hair, lighting, or expression cues are visible under close inspection. (0s)
• identity: pass. The subject matches the fictional reference: silver-gray wavy shoulder-length hair, hazel-brown eyes, and black rectangular glasses. Facial structure and expression align with the reference identity board. (0s)
• fixed camera: pass. The framing is eye-level and frontal, consistent with a fixed home camera. The background shows a coherent room with furniture, bookshelves, and natural lighting, with no evidence of generated camera angle changes. (0s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic depiction of an older woman in a domestic environment throughout the clip. (1s)
• realism level 2: pass. Natural skin texture, hair, and ambient lighting without significant synthetic plastic sheen. (2.5s)
• realism level 3: pass. Consistent facial expressions and natural ocular motion with minimal CGI artifacts. (3.5s)
• realism level 4: fail. Subtle digital smoothing around mouth articulation and minor synthetic rendering cues during speech. (1.8s)
• lip sync: pass. Lip movements correspond directly with spoken words throughout the sentence. (2, 4.5s)

openai/gpt-5.6-sol

sampled_frames · 4/4 rubric index

• realism level 1: pass. Sampled frames show a coherent photographic older adult with consistent facial anatomy, clothing, and a plausible home interior rather than cartoon or obvious 3D rendering. (0, 3, 5.92s)
• realism level 2: pass. Skin texture, wrinkles, eyes behind glasses, teeth, lips, gray hair, and room lighting remain mostly photographic across varied expressions; no major waxy or painted appearance is visible. (0.5, 1, 2.5, 4.5s)
• realism level 3: pass. Across the visible samples, facial details, hair strands, reflections in the glasses, and changing expressions retain strong camera realism without readily noticeable doll-like eyes, painted hair, or CGI shading. (0, 1.5, 3.5, 5.5s)
• realism level 4: pass. Close visible inspection of the sampled frames shows a convincingly ordinary home-camera portrait with plausible skin detail, eyes, mouth, hair, illumination, and background; no meaningful visible synthetic tell is apparent. (1, 2, 4, 5.92s)
• lip sync: unknown. Only sampled still frames are supplied and no audible words are available for comparison, so mouth-to-audio alignment cannot be established. (0, 3, 5.92s)

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with natural skin tones, hair texture, and facial features, not resembling cartoon, illustration, or obvious 3D render. Observed across all sampled frames (0.0–5.92s). (0, 5.92s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic qualities are visible. Minor synthetic cues cannot be ruled out but are not evident in sampled frames (0.0–5.92s). (0, 5.92s)
• realism level 3: pass. Under careful inspection of sampled frames (0.0–5.92s), no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading are present. Expression and lighting appear consistent with natural human behavior and home-camera conditions. (0, 5.92s)
• realism level 4: pass. The appearance is convincingly ordinary for a live-action home-camera setting under close inspection. No perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression are visible in any sampled frame (0.0–5.92s). Wrinkles and ambient clutter are natural, not synthetic tells. (0, 5.92s)
• lip sync: unknown. No audio is provided; sampled frames alone cannot prove or disprove mouth alignment with audible words. Lip motion is visible but cannot be validated without audio. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Coherent photographic human appearance throughout the video. (0s)
• realism level 2: pass. Plausible skin, hair, and lighting without excessive synthetic smoothing. (1.5s)
• realism level 3: pass. Naturalistic facial features and environmental lighting under close inspection. (3s)
• realism level 4: pass. Convincingly ordinary live-action webcam appearance with no obvious synthetic artifacts. (4.5s)
• lip sync: fail. The subject's mouth remains completely closed while speech audio is playing. (1, 4s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 3/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, hair, clothing, and natural lighting; no predominant cartoon, illustration, or obvious 3D render characteristics are visible. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; minor synthetic cues may exist but no major waxy, painted, or plastic appearance is evident across sampled frames. (0s)
• realism level 3: pass. Under sampled inspection, the subject shows strong camera realism: no readily noticeable doll eyes, beauty-render skin, painted hair, CGI shading, or synthetic expression; minor deviations are not visually apparent. (0s)
• realism level 4: fail. Perceptibly CGI/avatar-like eyes and skin texture are visible; the subject’s facial rendering lacks the micro-textural detail and lighting interaction expected of ordinary live-action home-camera footage, failing the strict target. (0s)
• lip sync: unknown. Sampled frames cannot prove lip sync; no audio is provided and temporal continuity of mouth movement relative to speech cannot be assessed from still images. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Coherent photographic human appearance throughout the clip. (2.5s)
• realism level 2: pass. Plausible skin, hair, and lighting without major plastic or painted artifacts. (3s)
• realism level 3: pass. Naturalistic textures and convincing camera capture without noticeable CGI rendering. (4s)
• realism level 4: pass. Ordinary home-camera appearance maintained with realistic fine details and natural motion. (2s)
• lip sync: pass. Mouth movements correspond consistently with the spoken audio track. (1.5, 4.5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin texture, realistic hair strands, and plausible lighting. No cartoon, illustration, or obvious 3D render artifacts are visible in sampled frames. (0, 5.92s)
• realism level 2: pass. Skin, eyes, lips, and hair show mostly plausible photographic qualities. Minor synthetic cues such as slight uniformity in skin tone and subtle edge softness are present but do not dominate; no major waxy, painted, or plastic appearance is observed. (0, 5.92s)
• realism level 3: pass. Under sampled frame inspection, the subject displays strong camera realism: wrinkles, natural hair flyaways, and ambient lighting are consistent. Minor deviations such as slight gloss on skin and uniform eye reflection are present but not readily noticeable as doll-like or beauty-rendered. (0, 5.92s)
• realism level 4: pass. The appearance is convincingly ordinary live-action home-camera under close inspection of sampled frames. No perceptibly CGI/avatar-like eyes, lips, skin, hair, or light. Wrinkles and background clutter are natural; no synthetic tells are visible. (0, 5.92s)
• lip sync: unknown. Audio is not provided; sampled frames alone cannot prove lip sync. Mouth shape changes are visible across frames but cannot be correlated with speech without audio. (0, 5.92s)

google/gemini-3.8-flash

native_video_audio · 2/4 rubric index

• realism level 1: pass. Coherent photographic human appearance throughout the video. (2.5s)
• realism level 2: pass. Skin, hair, eyes, and lighting appear mostly natural and photographic. (3s)
• realism level 3: fail. Unnatural synthetic smoothing and slight warping around the mouth during speech. (1.5, 4s)
• realism level 4: fail. Visible AI generation artifacts including unstable lower facial movement and garbled subtitle text overlay. (2, 5s)
• lip sync: pass. Mouth movements generally synchronize with the spoken audio words. (1, 4.5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, hair, and clothing, without predominant cartoon, illustration, or obvious 3D render characteristics. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic appearance is visible in sampled frames. (0s)
• realism level 3: pass. Under careful inspection of sampled frames, the subject shows strong camera realism with only minor deviations; no readily noticeable doll eyes, beauty-render skin, painted hair, or CGI shading is present. (0s)
• realism level 4: pass. The appearance is convincingly ordinary for a live-action home-camera recording under close inspection; no perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression is visible in sampled frames. (0s)
• lip sync: unknown. Sampled frames cannot prove lip sync; audio is not provided, and mouth movements cannot be reliably correlated to speech without temporal audio-visual alignment. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Coherent photographic human appearance throughout the clip. (0, 3s)
• realism level 2: pass. Natural skin texture, hair, and lighting without plastic rendering. (1, 4s)
• realism level 3: pass. Consistent realistic features and lighting without synthetic artifacts. (2, 5s)
• realism level 4: pass. Plausible live-action home camera quality with natural micro-movements. (1.5, 4.5s)
• lip sync: pass. Mouth movement corresponds accurately with the spoken speech audio. (0.5, 3.5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, skin texture, hair, and clothing. No predominant cartoon, illustration, or obvious 3D render artifacts are visible in any sampled frame. (0, 5.92s)
• realism level 2: pass. Skin, eyes, lips, and hair show mostly plausible photographic qualities. Minor synthetic cues may be present under close inspection, but there is no major waxy, painted, or plastic appearance across the sampled frames. (0, 5.92s)
• realism level 3: pass. Under careful inspection of sampled frames, the subject displays strong camera realism. Minor deviations may exist, but there are no readily noticeable doll eyes, beauty-render skin, painted hair, CGI shading, or synthetic expression. (0, 5.92s)
• realism level 4: pass. The appearance is convincingly ordinary live-action home-camera under close inspection. No meaningful visible synthetic tell is present in sampled frames — eyes, lips, skin, hair, lighting, and expression appear naturally photographic without CGI/avatar-like artifacts. (0, 5.92s)
• lip sync: unknown. Sampled frames cannot prove lip sync. Audio is not provided, and temporal continuity between frames is insufficient to verify mouth alignment with audible words. (0s)

google/gemini-3.8-flash

still_image · 4/4 rubric index

• realism level 1: pass. Shows a photographic human presentation rather than an illustration or 3D render. (0s)
• realism level 2: pass. Natural skin, hair, and lighting without excessive waxy or painted artifacts. (0s)
• realism level 3: pass. Strong camera realism with realistic facial features and balanced ambient illumination. (0s)
• realism level 4: pass. Convincing live-action home camera quality with no obvious synthetic anomalies. (0s)

openai/gpt-5.6-sol

still_image

• realism level 1: pass. The visible frame shows a coherent photographic older adult with anatomically consistent facial features, hair, clothing, and indoor lighting rather than a cartoon, illustration, or obvious 3D render. (0s)
• realism level 2: pass. Skin texture, eyes behind the glasses, lips, gray hair, and soft room illumination appear mostly photographically plausible, with no major waxy, painted, or plastic region visible. (0s)
• realism level 3: pass. At the supplied resolution, facial texture, eye reflections, hair strands, glasses, garment texture, and background depth remain camera-like; no readily noticeable doll eyes, painted hair, CGI shading, or synthetic expression is visible. (0s)
• realism level 4: unknown. The sole sampled frame looks convincingly live-action, but 0.05 seconds of sampled-frame coverage is insufficient to establish the strict absence of meaningful visible synthetic tells under close inspection beyond this instant. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin texture, realistic hair strands, and plausible facial structure without cartoonish, illustrated, or obvious 3D render characteristics. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair show mostly plausible photographic qualities under ambient indoor lighting. No major waxy, painted, or plastic appearance is evident; minor synthetic cues are not readily observable. (0s)
• realism level 3: pass. Under visible inspection, the image maintains strong camera realism: skin texture and wrinkles appear natural, eyes are not doll-like, hair has realistic strand variation, and lighting is consistent with indoor ambient sources. No obvious CGI shading or synthetic expression is present. (0s)
• realism level 4: pass. The subject and setting convincingly resemble an ordinary live-action home-camera recording. No perceptibly CGI or avatar-like features are visible in eyes, lips, skin, hair, or expression. Minor background clutter and natural lighting do not detract from the realism. (0s)

google/gemini-3.8-flash

still_image · 4/4 rubric index

• realism level 1: pass. Photographic depiction of an older adult woman in a domestic environment. (0s)
• realism level 2: pass. Natural skin texture, hair, and lighting consistent with camera footage. (0s)
• realism level 3: pass. Convincing facial details and environment without obvious CGI shading or artificial smoothing. (0s)
• realism level 4: pass. Plausible live-action home-camera realism across visible features and backdrop. (0s)

openai/gpt-5.6-sol

still_image · 4/4 rubric index

• realism level 1: pass. The visible frame shows a coherent photographic older woman in a naturally depicted living-room setting, not a cartoon, illustration, or obvious 3D render. (0s)
• realism level 2: pass. Skin texture, eyes behind glasses, closed lips, gray hair, clothing, and room lighting are mostly photographically plausible, without major waxy, painted, or plastic appearance. (0s)
• realism level 3: pass. In the supplied frame, facial texture, eye reflections, hair strands, expression, and soft indoor illumination remain camera-realistic under inspection; no readily noticeable doll-like or CGI-rendered feature is visible. (0s)
• realism level 4: pass. The visible frame convincingly resembles an ordinary live-action home-camera portrait: facial aging detail, glasses, hair, clothing, subject lighting, and background depth are mutually consistent, with no meaningful visible synthetic tell. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin tone, visible wrinkles, realistic hair texture, and plausible clothing and background. No cartoon, illustration, or obvious 3D render artifacts are present. (0s)
• realism level 2: pass. Skin shows natural texture and pores, eyes have realistic reflections and depth, lips have subtle shading and creases, hair has strand-level variation and natural lighting, and ambient lighting matches indoor environment. No major waxy, painted, or plastic appearance. (0s)
• realism level 3: pass. Under close inspection, the subject displays strong camera realism: eyes lack doll-like gloss, skin shows natural aging and microtexture, hair has realistic flyaways and lighting interaction, and expression appears organic. No readily noticeable CGI shading, beauty-render skin, or synthetic cues. (0s)
• realism level 4: pass. The image convincingly mimics ordinary live-action home-camera footage: skin, eyes, hair, and lighting show no perceptible CGI/avatar-like artifacts. Wrinkles, ambient blur, and background clutter are consistent with real-world capture and do not indicate synthetic origin. (0s)

google/gemini-3.8-flash

native_video_audio · 2/4 rubric index

• realism level 1: pass. Character has a photographic human appearance rather than an overt cartoon or 3D render. (2s)
• realism level 2: pass. Skin texture, hair, and lighting appear plausibly photographic without heavy waxy artifacts. (3.5s)
• realism level 3: fail. Visible digital animation softening and unnatural rigidity in facial deformation during speech. (4s)
• realism level 4: fail. Avatar-like lip sync warping and synthetic motion inconsistencies show clear AI generation cues. (5s)
• lip sync: pass. Mouth shapes roughly align with the audible spoken dialogue throughout the clip. (2.5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 3/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, hair, clothing, and environment; no predominant cartoon, illustration, or obvious 3D render artifacts are visible. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair show mostly plausible photographic qualities under sampled frames. Minor synthetic cues may exist, but no major waxy, painted, or plastic appearance is evident. (0s)
• realism level 3: pass. Under sampled inspection, the subject exhibits strong camera realism: skin texture, hair strands, lighting, and expression appear natural with only minor deviations. No readily noticeable doll eyes, beauty-render skin, or CGI shading. (0s)
• realism level 4: fail. Perceptibly synthetic cues are present: eyes lack micro-reflections and depth, skin appears overly smooth and uniform, and facial expression transitions appear slightly plastic or avatar-like under close inspection, failing the strict home-camera realism standard. (0s)
• lip sync: unknown. Audio is not provided; sampled frames alone cannot prove or disprove mouth alignment with audible words. (0s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic depiction of an older woman in a study context. (1s)
• realism level 2: pass. Mostly plausible facial textures, hair, and lighting throughout. (2.5s)
• realism level 3: pass. Convincing natural lighting, detailed skin, and realistic subtle expressions. (4s)
• realism level 4: fail. Slight synthetic rigidity around mouth and eye blinking mechanics. (3.2s)
• lip sync: pass. Viseme movements generally track the spoken phrases cleanly. (1.5, 4.5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 3/4 rubric index

• realism level 1: pass. The character exhibits coherent photographic human appearance with recognizable facial features, hair, clothing, and environment, without predominant cartoon, illustration, or obvious 3D render artifacts across all sampled frames. (0, 7.92s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic appearance is visible. Minor synthetic cues may be present but are not dominant in the sampled frames. (0, 7.92s)
• realism level 3: pass. Under careful inspection of sampled frames, the character displays strong camera realism: skin texture, eye reflections, hair strands, and lighting are consistent with live photography. No readily noticeable doll eyes, beauty-render skin, or CGI shading is evident. (0, 7.92s)
• realism level 4: fail. Perceptibly synthetic appearance in eyes and skin: the eyes exhibit uniform, overly glossy reflections and lack natural micro-movements or depth; skin has a smooth, airbrushed quality inconsistent with ordinary home-camera footage. These are meaningful visible synthetic tells under close inspection. (0, 7.92s)
• lip sync: unknown. No audio is provided; sampled frames alone cannot prove or disprove mouth alignment with audible words. Temporal coverage is insufficient to assess lip sync. (0s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic presentation with recognizable human anatomy and real-world environment. (0, 2s)
• realism level 2: pass. Plausible skin texture, hair strands, and lighting throughout the scene. (1, 3.5s)
• realism level 3: pass. Convincing facial features and lighting without obvious CGI artifacts or doll-like eyes. (2, 4.5s)
• realism level 4: fail. Subtle AI motion smoothing and synthetic rendering cues around mouth movement and fingers. (3, 5.5s)
• lip sync: pass. Lip movements correspond closely with the spoken dialogue throughout. (0.5, 5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 2/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, skin texture, hair, and clothing. No predominant cartoon, illustration, or obvious 3D render characteristics are visible across sampled frames. (0, 7.961667s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting. Minor synthetic cues may be present (e.g., slight uniformity in skin texture), but no major waxy, painted, or plastic appearance is evident. (0, 7.961667s)
• realism level 3: fail. Under careful inspection, the eyes exhibit a slightly doll-like quality with overly uniform catchlights and lack of micro-texture. Skin appears smoothed beyond natural variation, and hair lacks fine strand detail, suggesting a beauty-render or CGI aesthetic rather than strong camera realism. (0, 7.961667s)
• realism level 4: fail. The subject displays perceptibly synthetic characteristics: eyes have an avatar-like sheen and uniformity, skin lacks natural pore and wrinkle variation, and hair appears painted rather than fibrous. These deviations prevent convincingly ordinary live-action home-camera appearance under close inspection. (0, 7.961667s)
• lip sync: unknown. Audio is not provided; sampled frames alone cannot prove mouth alignment with audible words. Temporal coverage is insufficient to determine lip sync. (0s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic depiction of an older woman in an indoor setting. (1s)
• realism level 2: pass. Skin texture, hair strands, and lighting appear plausibly photographic. (2.5s)
• realism level 3: pass. Naturalistic facial features and subtle micro-expressions without excessive rendering artifacts. (4s)
• realism level 4: fail. Rigid torso posture and slightly mechanical mouth motion betray synthetic generation. (3, 5s)
• lip sync: pass. Lip movements generally synchronize with the spoken English sentences. (0.5, 4.5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 2/4 rubric index

• realism level 1: pass. The character exhibits coherent photographic human appearance with recognizable features (face, hair, clothing, hands) and plausible lighting/shadows. No predominant cartoon, illustration, or obvious 3D render artifacts are visible across sampled frames. (0, 7.92s)
• realism level 2: pass. Skin texture, eye reflection, lip shape, and hair strands appear mostly plausible under photographic conditions. Minor synthetic cues such as uniform skin smoothness and slightly idealized lighting are present but do not rise to major waxy, painted, or plastic appearance. (0, 7.92s)
• realism level 3: fail. Under careful inspection, the eyes exhibit a slightly doll-like or beauty-render quality with overly uniform catchlights and minimal scleral detail. Skin lacks micro-texture and pore variation typical of live-action footage. Expression appears subtly synthetic, especially in the mouth and brow area during transitions. (0, 7.92s)
• realism level 4: fail. Perceptibly CGI/avatar-like qualities are present: eyes have synthetic catchlights and lack natural scleral texture; skin appears airbrushed with no visible pores or fine wrinkles; hair strands lack individual strand variation and shadowing; overall expression is overly consistent and lacks micro-muscular realism expected in ordinary home-camera footage. (0, 7.92s)
• lip sync: unknown. Audio is not provided. Sampled frames alone cannot prove or disprove mouth alignment with audible words. Temporal coverage is insufficient to infer continuous motion or phoneme alignment. (0, 7.92s)

google/gemini-3.8-flash

native_video_audio · 3/4 rubric index

• realism level 1: pass. Photographic human appearance maintained throughout without overt cartoon rendering. (1.5s)
• realism level 2: pass. Plausible skin texture, hair, and lighting without excessive waxy or plastic artifacts. (3s)
• realism level 3: pass. Convincing facial movement, realistic skin creasing, and natural lighting integration. (4.5s)
• realism level 4: fail. Subtle AI warping around the jawline and mouth boundary during speaking transitions. (3.8s)
• lip sync: pass. Visible mouth shapes correspond appropriately to spoken words across the sentence. (2.2s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 3/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, skin texture, hair, and clothing. No predominant cartoon, illustration, or obvious 3D render artifacts are visible in any sampled frame. (0, 7.961667s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting. Minor synthetic cues may be present in fine texture or specular highlights, but no major waxy, painted, or plastic appearance is evident across sampled frames. (0, 7.961667s)
• realism level 3: pass. Under careful inspection of sampled frames, the subject displays strong camera realism. Minor deviations may exist in microtexture or lighting falloff, but there are no readily noticeable doll eyes, beauty-render skin, painted hair, CGI shading, or synthetic expression. (0, 7.961667s)
• realism level 4: fail. Perceptibly synthetic cues are present: eyes have an unnaturally uniform gloss and lack micro-reflections typical of live human eyes; skin texture appears overly smooth and uniform, especially around the forehead and cheeks; hair strands show repetitive, slightly artificial wave patterns. These cues violate the strict target of convincingly ordinary live-action home-camera appearance under close inspection. (0, 7.961667s)
• lip sync: unknown. No audio is provided; sampled frames alone cannot prove or disprove mouth alignment with audible words. Temporal coverage is insufficient to assess lip sync. (0s)

google/gemini-3.8-flash

native_video_audio · 4/4 rubric index

• realism level 1: pass. Photographic human appearance maintained without obvious cartoon rendering. (0, 2.5s)
• realism level 2: pass. Plausible skin texture, hair, and lighting without severe plastic look. (1, 3.5s)
• realism level 3: pass. Naturalistic facial movement, subtle wrinkles, and coherent lighting under close view. (2, 4.5s)
• realism level 4: pass. Convincing live-action webcam appearance without noticeable synthetic artifacts. (0.5, 5s)
• lip sync: pass. Mouth shapes align with spoken speech syllables throughout. (0.5, 5.5s)

openai/gpt-5.6-sol

not_reviewed

qwen/qwen3-vl-235b-a22b-instruct

sampled_frames · 3/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance with recognizable facial features, skin texture, hair, and clothing. No predominant cartoon, illustration, or obvious 3D render characteristics are visible in any sampled frame. (0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 5.961667s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting. Minor synthetic cues such as slight uniformity in skin tone or hair strand definition are present but do not rise to major waxy, painted, or plastic appearance across sampled frames. (0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 5.961667s)
• realism level 3: pass. Under careful inspection of sampled frames, only minor deviations are visible (e.g., slight smoothing in skin texture, uniform lighting). No readily noticeable doll eyes, beauty-render skin, painted hair, CGI shading, or synthetic expression is evident. (0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 5.961667s)
• realism level 4: fail. Perceptibly synthetic cues remain under close inspection: eyes exhibit slight uniform gloss and lack micro-reflections typical of live-action; skin texture appears overly smooth and uniform, lacking fine pores and micro-variations; hair strands show repetitive patterns and lack individual strand dynamics. These deviations prevent convincingly ordinary live-action home-camera appearance. (0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 5.961667s)
• lip sync: unknown. Sampled frames cannot prove lip sync. No audio is provided for comparison, and temporal continuity between frames is insufficient to verify mouth alignment with audible words. (0s)

google/gemini-3.8-flash

still_image · 2/4 rubric index

• realism level 1: pass. Shows coherent photographic human features rather than a cartoon or render. (0s)
• realism level 2: pass. Plausible hair, skin, and natural lighting without extreme plastic sheen. (0s)
• realism level 3: fail. Facial skin texture and hair boundaries appear subtly smoothed and synthetic. (0s)
• realism level 4: fail. Noticeable AI softening and synthetic hair strands fall short of raw camera capture. (0s)

openai/gpt-5.6-sol

still_image

• realism level 1: pass. The sampled frame shows a coherent photographic older woman with natural human facial structure, clothing, and a plausible home interior rather than cartoon or obvious 3D-render styling. (0s)
• realism level 2: pass. Skin texture, eyes behind glasses, closed lips, gray hair, and soft indoor illumination are mostly photographically plausible, without major waxy or painted regions. (0s)
• realism level 3: pass. At the visible scale, facial features, hair strands, eyeglass reflections, fabric, and scene lighting remain camera-like; no readily noticeable doll eyes, painted hair, CGI shading, or synthetic expression is visible. (0s)
• realism level 4: unknown. The only supplied coverage is one 768×432 frame from a 0.05-second clip; it appears plausibly live-action, but this limited spatial and temporal coverage is insufficient to establish the absence of any meaningful synthetic tell under close inspection. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin tone, realistic hair texture, and plausible facial features without cartoon, illustration, or obvious 3D render artifacts at 0.0s. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic lighting; no major waxy, painted, or plastic qualities are visible at 0.0s. (0s)
• realism level 3: pass. Under visible inspection at 0.0s, the subject shows strong camera realism: natural skin texture, realistic eye reflection, hair strands, and ambient lighting with no readily noticeable doll eyes, beauty-render skin, or CGI shading. (0s)
• realism level 4: pass. At 0.0s, the subject exhibits convincingly ordinary live-action home-camera appearance: natural wrinkles, ambient lighting, and no perceptible CGI/avatar-like eyes, lips, skin, hair, or expression; clutter and depth of field are consistent with real-world capture. (0s)

google/gemini-3.8-flash

still_image

• realism level 1: pass. Shows coherent photographic human appearance without cartoon or 3D render styling. (0s)
• realism level 2: pass. Skin, hair, eyes, and lighting appear photographic without prominent waxy or plastic artifacts. (0s)
• realism level 3: pass. Natural facial features, realistic hair texture, and believable lighting with no obvious synthetic shading. (0s)
• realism level 4: unknown. Extremely brief duration limits verification of temporal artifacts or natural motion under close inspection. (0s)

openai/gpt-5.6-sol

still_image

• realism level 1: pass. The visible frame has a coherent photographic human appearance with natural room perspective and illumination, not a predominantly illustrated or obvious 3D-rendered look. (0s)
• realism level 2: pass. Skin, eyes, lips, gray hair, glasses, and window-lit shading are mostly photographically plausible; no major waxy, painted, or plastic treatment is visible. (0s)
• realism level 3: pass. At the available resolution, facial features, hair strands, skin variation, glasses, and indoor lighting retain strong camera realism without a readily noticeable doll-eye, painted-hair, or CGI-shading defect. (0s)
• realism level 4: unknown. Only one sampled frame from a 0.05-second clip is available; it appears plausibly live-action, but this limited coverage is insufficient to establish absence of any meaningful synthetic tell under close inspection. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin texture, realistic hair strands, plausible facial structure, and consistent lighting. No cartoon, illustration, or obvious 3D render artifacts are visible. (0s)
• realism level 2: pass. Skin shows natural pores and subtle imperfections; eyes have realistic reflections and depth; lips display natural texture and slight asymmetry; hair has varied strand direction and volume; lighting creates soft shadows consistent with indoor ambient sources. No major waxy, painted, or plastic appearance. (0s)
• realism level 3: pass. Under close inspection, the subject maintains strong camera realism: eyes show natural corneal reflections and scleral detail; skin exhibits fine wrinkles and uneven tone; hair has realistic flyaways and volume; expression appears organic with slight asymmetry. No doll eyes, beauty-render skin, painted hair, or CGI shading are visible. (0s)
• realism level 4: pass. The subject appears convincingly like ordinary live-action home-camera footage: skin shows natural aging texture including fine lines and pores; hair has realistic frizz and strand variation; lighting is consistent with ambient indoor sources; expression is neutral and natural. No perceptibly CGI/avatar-like eyes, lips, skin, hair, or light. Wrinkles and background clutter are natural, not synthetic tells. (0s)

google/gemini-3.8-flash

still_image · 3/4 rubric index

• realism level 1: pass. Presents a coherent photographic human form in a domestic interior rather than a stylized render. (0s)
• realism level 2: pass. Natural facial features, hair strands, and ambient indoor lighting without obvious plastic texture. (0s)
• realism level 3: pass. Subtle skin texture, natural eye reflections, and plausible home-camera lighting with no overt CGI shading. (0s)
• realism level 4: fail. Slightly over-smoothed skin blending and static, portrait-like framing characteristic of synthetic generation. (0s)

openai/gpt-5.6-sol

still_image

• realism level 1: pass. The sampled frame shows a coherent photographic older woman with natural human facial structure, hair, clothing, and a camera-like domestic background rather than cartoon or obvious 3D-render styling. (0s)
• realism level 2: pass. Skin texture, eyes behind glasses, closed lips, gray hair strands, and soft window/interior lighting are mostly photographically plausible, with no major waxy, painted, or plastic appearance. (0s)
• realism level 3: pass. Under inspection of the sole frame, facial texture and age lines, eyeglass reflections, layered hair, knit clothing, and scene lighting remain strongly camera-realistic; no readily noticeable doll eyes, painted hair, CGI shading, or synthetic expression is visible. (0s)
• realism level 4: unknown. The single sampled frame appears convincingly live-action and has no clear visible synthetic tell, but one frame from a 0.05-second clip is insufficient to establish the strict no-meaningful-tell standard across expression or motion. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. Frame at 0.0s shows a coherent photographic human appearance: natural skin tones, realistic hair texture, and plausible facial structure without cartoon, illustration, or obvious 3D render artifacts. (0s)
• realism level 2: pass. At 0.0s, skin shows natural texture and pores, eyes have realistic reflections and depth, lips exhibit subtle moisture and creases, hair has varied strand detail, and lighting appears consistent with indoor ambient sources — no major waxy, painted, or plastic appearance. (0s)
• realism level 3: pass. Frame at 0.0s shows strong camera realism: skin texture, eye reflections, hair strand variation, and lighting gradients appear natural. Minor deviations (if any) are not readily noticeable; no doll eyes, beauty-render skin, painted hair, or CGI shading evident. (0s)
• realism level 4: pass. At 0.0s, the image convincingly mimics ordinary live-action home-camera footage: natural wrinkles, realistic fabric texture, ambient lighting with soft shadows, and no perceptible CGI/avatar-like eyes, lips, skin, hair, or expression. Background clutter and depth of field are consistent with casual home video. (0s)

google/gemini-3.8-flash

still_image · 3/4 rubric index

• realism level 1: pass. Displays a photographic human likeness without illustrative styling. (0s)
• realism level 2: pass. Skin texture, hair, and lighting appear generally photographic. (0s)
• realism level 3: pass. Naturalistic facial features and lighting hold up under standard inspection. (0s)
• realism level 4: fail. Subtle synthetic smoothing and symmetrical studio-like rendering cues on skin and hair detract from an authentic home-camera capture. (0s)

openai/gpt-5.6-sol

still_image

• realism level 1: pass. The visible frame presents a coherent photographic older woman with anatomically consistent face, hands, clothing, and a plausible home interior rather than cartoon or obvious 3D styling. (0s)
• realism level 2: pass. Skin texture, eyes behind glasses, lips, gray hair, and window illumination are mostly photographically plausible, with no major waxy or painted region visible. (0s)
• realism level 3: pass. The single visible frame has strong camera realism: facial aging detail, hair strands, eyeglass reflections, fabric texture, and lighting remain coherent, without a readily noticeable doll-eye, painted-hair, or CGI-shading defect. (0s)
• realism level 4: unknown. The frame appears convincingly photographic, but coverage is limited to one sampled frame from a 0.05-second clip; this is insufficient to establish the strict absence of meaningful synthetic tells under close inspection across the clip. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. Frame at 0.0s shows a human figure with coherent photographic appearance: recognizable facial features, hair, clothing, and indoor setting without cartoon, illustration, or obvious 3D render artifacts. (0s)
• realism level 2: pass. At 0.0s, skin texture, eyes, lips, and hair exhibit mostly plausible photographic qualities. Minor synthetic cues may exist but no major waxy, painted, or plastic appearance is evident in this single frame. (0s)
• realism level 3: pass. At 0.0s, under careful inspection, the figure displays strong camera realism: skin shows natural texture and wrinkles, eyes are not doll-like, hair has realistic volume and shading, and lighting appears consistent with indoor ambient sources. No readily noticeable CGI shading or synthetic expression. (0s)
• realism level 4: pass. At 0.0s, the figure exhibits convincingly ordinary live-action home-camera appearance: natural skin pores and wrinkles, realistic glasses reflection, fabric texture on clothing, and ambient room lighting. No perceptibly CGI/avatar-like eyes, lips, skin, hair, or expression is visible in this frame. (0s)

google/gemini-3.8-flash

still_image · 2/4 rubric index

• realism level 1: pass. Coherent photographic human depiction is maintained without illustrated or cartoon appearance. (0s)
• realism level 2: pass. Plausible skin texture, hair, lighting, and environmental integration without heavy waxy artifacts. (0s)
• realism level 3: fail. Noticeable digital smoothing across facial features and somewhat stylized lighting cues. (0s)
• realism level 4: fail. Render-like skin smoothness and subtle synthetic sheen preclude indistinguishable raw home-camera realism. (0s)

openai/gpt-5.6-sol

still_image · 4/4 rubric index

• realism level 1: pass. The visible frame presents a coherent photographic older woman in a naturally lit home-office setting, not a cartoon, illustration, or obvious 3D render. (0s)
• realism level 2: pass. Skin texture, eyes behind glasses, lips, gray hair, clothing, and window lighting are mostly photographically plausible without major waxy, painted, or plastic appearance. (0s)
• realism level 3: pass. Careful inspection of the sole frame shows consistent facial detail, hair strands, reflections, shadows, and a natural expression; no readily noticeable doll eyes, painted hair, or CGI shading is visible. (0s)
• realism level 4: pass. The supplied frame convincingly resembles an ordinary live-action home-camera image: facial features, aging cues, hair, glasses, hands, lighting, and surrounding objects show no meaningful visible synthetic tell in this frame. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin tone, visible wrinkles, realistic hair texture, and plausible clothing. No cartoon, illustration, or obvious 3D render artifacts are present in the single frame. (0s)
• realism level 2: pass. Skin shows natural texture and subtle shading, eyes have realistic reflections and depth, lips have natural coloration and creasing, hair appears volumetric with individual strands, and lighting is consistent with indoor ambient sources. No major waxy, painted, or plastic appearance is evident. (0s)
• realism level 3: pass. Under close inspection, the image maintains strong camera realism: skin shows fine pores and natural variation, eyes lack doll-like uniformity, hair has realistic flyaways and shadowing, and expression appears natural without synthetic stiffness. Minor deviations are not readily noticeable. (0s)
• realism level 4: pass. The frame exhibits convincingly ordinary live-action home-camera appearance: wrinkles are naturally distributed, lighting matches ambient indoor sources, skin and hair show micro-texture without CGI shading, and expression is consistent with a relaxed human pose. No perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression are visible. (0s)

google/gemini-3.8-flash

still_image · 2/4 rubric index

• realism level 1: pass. Shows a coherent photographic human appearance in an office environment. (0s)
• realism level 2: pass. Features generally plausible skin, hair, and lighting without severe plastic artifacts. (0s)
• realism level 3: fail. Noticeable synthetic smoothing on facial skin and airbrushed hair rendering. (0s)
• realism level 4: fail. Lacks ordinary home-camera texture, showing clear AI beauty-filter smoothing and artificial lighting. (0s)

openai/gpt-5.6-sol

still_image · 2/4 rubric index

• realism level 1: pass. The sampled frame shows a coherent photographic older woman with natural human proportions and a plausible home-office setting, rather than cartoon or obvious 3D-render styling. (0s)
• realism level 2: pass. Visible skin texture, eyes behind glasses, closed lips, gray hair, clothing, and window illumination are mostly photographically plausible without major waxy or plastic appearance. (0s)
• realism level 3: fail. Under inspection, the face appears noticeably beauty-rendered: skin is unusually smooth and evenly modeled, the eyes/reflections have a polished synthetic quality, and several hair contours look overly shaped. (0s)
• realism level 4: fail. The single visible frame does not achieve convincingly ordinary home-camera realism under close inspection because the polished facial skin, glassy eye area, and highly sculpted hair remain perceptible synthetic tells. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 3/4 rubric index

• realism level 1: pass. The subject exhibits coherent photographic human appearance: natural skin texture, realistic hair strands, plausible facial features, and consistent lighting. No cartoon, illustration, or obvious 3D render artifacts are present in the sampled frame. (0s)
• realism level 2: pass. Skin, eyes, lips, and hair appear mostly plausible under photographic conditions. Minor synthetic cues may exist (e.g., slight uniformity in skin tone), but no major waxy, painted, or plastic appearance is evident in the frame. (0s)
• realism level 3: pass. Under careful inspection of the single frame, the subject displays strong camera realism: eyes are not doll-like, skin lacks beauty-render smoothness, hair shows natural variation, lighting is consistent with indoor ambient sources, and expression appears natural. No readily noticeable CGI shading or synthetic expression is visible. (0s)
• realism level 4: fail. Perceptibly CGI/avatar-like qualities are present: the skin exhibits an overly uniform, slightly airbrushed texture; the hair lacks fine strand-level randomness; and the overall lighting and depth-of-field feel slightly rendered rather than captured by a live-action home camera. These deviations prevent convincing ordinary live-action appearance under close inspection. (0s)

google/gemini-3.8-flash

still_image · 3/4 rubric index

• realism level 1: pass. Photographic visual appearance of a woman in an office setting. (0s)
• realism level 2: pass. Plausible skin texture, lighting, and hair rendering throughout. (0s)
• realism level 3: pass. Naturalistic facial features and environmental light without pronounced CGI artifacts. (0s)
• realism level 4: fail. Subtle digital smoothing on skin and slightly stylized hair consistency indicate AI generation. (0s)

openai/gpt-5.6-sol

still_image · 4/4 rubric index

• realism level 1: pass. The sampled frame shows a coherent photographic older woman in a naturally lit home-office scene, not a cartoon, illustration, or obvious 3D render. (0s)
• realism level 2: pass. Visible skin texture, eyes behind glasses, lips, gray hair, clothing, and window lighting are mostly photographically plausible without major waxy or plastic rendering. (0s)
• realism level 3: pass. Under inspection of the single frame, facial features, hair strands, skin shading, expression, hands, and scene lighting remain strongly camera-realistic, with no readily noticeable doll-like or CGI feature. (0s)
• realism level 4: pass. The visible frame convincingly resembles an ordinary live-action home-camera image; the face, glasses, hair, hands, clothing, window light, and office background show no meaningful visible synthetic tell at the supplied resolution. (0s)

qwen/qwen3-vl-235b-a22b-instruct

still_image · 4/4 rubric index

• realism level 1: pass. Frame at 0.0 s shows a coherent photographic human appearance: recognizable facial features, natural skin tone, hair texture, and clothing consistent with a real person in a home setting. No cartoon, illustration, or obvious 3D render artifacts are present. (0s)
• realism level 2: pass. At 0.0 s, skin shows natural texture and subtle imperfections, eyes have realistic reflections and depth, lips show natural color and contour, hair has varied strand detail, and lighting appears consistent with indoor ambient sources. No major waxy, painted, or plastic appearance is visible. (0s)
• realism level 3: pass. At 0.0 s, under close inspection, the subject exhibits strong camera realism: skin pores and fine wrinkles are visible, eye reflections are natural, hair strands are individually defined, and lighting creates realistic shadows and highlights. No doll eyes, beauty-render skin, painted hair, or CGI shading is apparent. (0s)
• realism level 4: pass. At 0.0 s, the image convincingly matches an ordinary live-action home-camera appearance: skin texture, eye detail, hair strand variation, and lighting are all consistent with unaltered video capture. No perceptibly CGI/avatar-like eyes, lips, skin, hair, light, or expression are visible. Clutter and blur are not present as synthetic tells. (0s)