VIDEO COMPARISON · OCTOBER 13, 2026

Which model makes usable B-roll?

One starting image. Twelve video tests.

Sort

How it sorts

Within each test: highest complete AI rating, then lower historical cost. Partial panels stay unranked by rating. Lowest cost reverses that priority. Unknown values follow known ones. Ranges use their upper estimate. Human votes stay separate.

Video uses USD per delivered minute. Images use USD per generation until complete spend and human usability checks establish cost per usable image. Failed submissions and retries stay in the test history. Fees and taxes are unconfirmed.

Kandinsky 6 Lite · 10 steps ↗

AI 72.5/100Est. $2.02/min

Kandinsky 6 Lite · 10 steps · take 1

864 × 480 · 5.042s · 24 fps

References used

Est. $2.02/minEstimated · $0.17 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI score75/1005/5 reviewed

Judge-by-judge reviews

3 Yes · 2 No · 0 Unsure. Range 61.25–95/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeYes75/100
anthropic/claude-sonnet-5.5
A mostly faithful continuous shot with consistent identity, outfit and setting, a slow push-in and plausible typing. Minor issues: an initial smile, a vague trackpad edit and small background drift. Usable as is.
GeminiNo61.25/100
google/gemini-4-argon
The clip preserves the subject's likeness, lighting, and slow camera push well. However, as she lowers her left hand to type, the laptop shifts forward and her hands end up typing on the bare wooden table in front of the device rather than on the keyboard.
GrokNo65/100
x-ai/grok-4.7
The clip preserves the woman and Sunday-home setting with strong realism and continuity, but the required trackpad edit and typing correction are not clearly executed. It should not be used as-is for the supplied brief.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
The video accurately preserves the woman’s face, body, and setting from the starting image. She performs the specified editing sequence with a focused, tired expression under consistent Sunday-evening lighting. A restrained camera push enhances the shot without breaking continuity. Minor deviation: trackpad interaction is not visibly depicted, only typing. All other dimensions—visual quality,物理行为,
MistralYes75/100
mistralai/mistral-large-4-0
Output closely matches brief: single continuous shot preserves woman's face, dark curly bun, charcoal sweater, home dining table, laptop, and Sunday-evening lighting. She performs trackpad edit, types correction, and pauses to inspect screen with focused, mildly fatigued expression. Restrained slow camera push maintains face and hand readability. No new people, scene changes, speaking, music, capt

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: c4c105a54140699a588311b98a9d288707340b2d6d1f47ab4c3cd5eef9aaeb10

Fixed 5 sec / 480p / 10 steps; no upscale; prompt expansion disabled.

Kandinsky 6 Lite · 10 steps · take 2

864 × 480 · 5.042s · 24 fps

References used

Est. $2.02/minEstimated · $0.17 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI score70/1005/5 reviewed

Judge-by-judge reviews

2 Yes · 2 No · 1 Unsure. Range 65–95/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeNo65/100
anthropic/claude-sonnet-5.5
Identity, setting, and lighting are well preserved with a smooth push-in. But the required deliberate trackpad edit, typing, and inspection beat is barely visible. The woman mainly rests her chin on her hand with a slight smile, so it misses the action brief.
GeminiNo70/100
google/gemini-4-argon
The clip preserves the subject, environment, and slow camera push well, but misses the requested typing action as her left hand remains resting on her chin throughout.
GrokYes86.25/100
x-ai/grok-4.7
A continuous realistic Sunday-evening edit scene preserves the woman, setting, and restrained camera movement. The requested hand action is subtle but visible, with no major identity, continuity, or visual defects.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
The output meets the brief’s core requirements: consistent face, body, and environment; realistic Sunday-evening lighting; stable camera with readable hands. Minor uncertainty in explicitly showing the edit/type/pause sequence, but no major defects. No new people, logos, or audio. Temporal and physical coherence are strong. Suitable for use as-is.
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the core subject (woman, bun, charcoal sweater, dining table, laptop, evening lighting) with stable identity and no continuity breaks. However, the brief's specific action sequence—'small deliberate edit using trackpad, types a brief correction, then pauses to inspect'—is not clearly executed; the motion appears as generic typing without distinct phases. The required 'restrai

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: e9295b1a54fa5d6df619e74d1ae31976e5667a80d667c3ae95f2804aa185eef4

Fixed 5 sec / 480p / 10 steps; no upscale; prompt expansion disabled.

Settings & evidence

$0.17 estimated per take; taxes and invoice unknown.

Provider endpoint documentation ↗

fal-ai/kandinsky6-lite/image-to-video

Fixed 5 sec / 480p / 10 steps; no upscale; prompt expansion disabled.

Two playable submissions. Human acceptance is pending; settings and native sizes differ.

Take 1 silent-file SHA-256: c4c105a54140699a588311b98a9d288707340b2d6d1f47ab4c3cd5eef9aaeb10

Take 2 silent-file SHA-256: e9295b1a54fa5d6df619e74d1ae31976e5667a80d667c3ae95f2804aa185eef4

LongCat Video Distilled ↗

AI 68.8/100Cost/min unknown

Duration mismatch: 11.667s returned instead of 5.4s requested (350 vs 162 frames). Billing is unconfirmed.

LongCat Video Distilled · take 1

1280 × 704 · 11.667s · 30 fps

References used

Cost/min unknownCharge unverified

Cost basis

Duration/billing ambiguous; excluded from cost ordering.

AI score70/1005/5 reviewed

Judge-by-judge reviews

2 Yes · 1 No · 2 Unsure. Range 61.25–95/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure70/100
anthropic/claude-sonnet-5.5
Strong scene, identity and clothing preservation with a single continuous push-in and appropriate mood. Weaknesses are blurry or smeared hand motion, morphing screen content, and a fast push, which may need review before use as is.
GeminiNo61.25/100
google/gemini-4-argon
The shot preserves the subject, lighting, and slow camera push well, but fails the action brief and continuity when the woman reaches up to touch the non-touchscreen laptop display while the screen graphics warp unnaturally.
GrokYes86.25/100
x-ai/grok-4.7
The clip preserves the woman and Sunday-evening home setting while showing a restrained camera push, trackpad edit, brief typing, and final screen inspection. Minor display and hand-motion softness do not prevent using it as is.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
Output fully preserves the woman’s face, body, and attire. The specified action sequence—small edit, brief typing, pause—is executed with coherent physics and stable continuity. Visual quality matches the realistic, dim Sunday-evening home setting. The camera push is slightly more pronounced than 'restrained' but does not break framing or readability. No new people, scene changes, or extraneous UI
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the woman's identity, setting, and core narrative (agency owner working Sunday evening) with stable continuity and realistic visuals. However, critical brief elements are missing or weak: the specified 'restrained slow camera push' is absent (framing remains static), and the action sequence (trackpad edit → type correction → pause to inspect) is not clearly executed—hands move

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 70/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: 81887065724c2962a59eab594d9c10f69b41aa46050a9e3cedc6eda5271c566b

Requested 162 frames / 30 fps / 720p / 12+12 steps; prompt expansion disabled. Returned 350 frames.

LongCat Video Distilled · take 2

1280 × 704 · 11.667s · 30 fps

References used

Cost/min unknownCharge unverified

Cost basis

Duration/billing ambiguous; excluded from cost ordering.

AI score67.5/1005/5 reviewed

Judge-by-judge reviews

0 Yes · 3 No · 2 Unsure. Range 62.5–95/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeNo62.5/100
anthropic/claude-sonnet-5.5
Limited assessment: brief explanation shortened; full text retained. Consistent realistic scene and plausible typing, but the expression is a smile rather than focused fatigue, there is no clear trackpad edit or inspection pause, and the background and camera angle shift. Not usable as is.
GeminiNo67.5/100
google/gemini-4-argon
The clip executes the slow camera push, lighting, and general laptop interaction well, but suffers from a noticeable identity and posture shift around 8.5 seconds that degrades facial likeness and temporal continuity.
GrokNo77.5/100
x-ai/grok-4.7
The clip convincingly preserves the woman and Sunday-evening setting with stable continuity and a restrained push-in. However, the required deliberate trackpad edit, brief correction, and screen inspection are not clearly demonstrated, so it should not be used as-is.
QwenUnsure95/100
qwen/qwen3-vl-235b-a22b-instruct
The output delivers strong visual fidelity, stable identity, and coherent motion, meeting most brief targets. However, the core narrative action — a small deliberate edit followed by a pause to inspect — is not clearly depicted, and the camera push is negligible. Without these specific behavioral and framing cues, it’s uncertain if the output fully illustrates the intended story of an agency owner
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the core visual identity (face, bun, charcoal sweater, home setting, Sunday-evening lighting) and maintains temporal continuity without cuts or warping. However, the specific brief actions—'small deliberate edit using trackpad,' 'types a brief correction,' and 'pauses to inspect'—are not clearly observable in sampled frames. The screen content is too small to verify editing, 0

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 70/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: c00d996a5d3eeb5a56c8e234fbeb00c4ea8c7e91aad7b179bd82ebf38a8270f2

Requested 162 frames / 30 fps / 720p / 12+12 steps; prompt expansion disabled. Returned 350 frames.

Settings & evidence

~$0.117 estimated per take; taxes and invoice unknown.

Provider endpoint documentation ↗

fal-ai/longcat-video/distilled/image-to-video/720p

Requested 162 frames / 30 fps / 720p / 12+12 steps; prompt expansion disabled. Returned 350 frames.

Two playable submissions. Human acceptance is pending; settings and native sizes differ.

Take 1 silent-file SHA-256: 81887065724c2962a59eab594d9c10f69b41aa46050a9e3cedc6eda5271c566b

Take 2 silent-file SHA-256: c00d996a5d3eeb5a56c8e234fbeb00c4ea8c7e91aad7b179bd82ebf38a8270f2

Kandinsky 6 Lite · 4 steps ↗

5/5 AI reviewsEst. $1.01/min

Kandinsky 6 Lite · 4 steps · take 1

864 × 480 · 5.042s · 24 fps

References used

Est. $1.01/minEstimated · $0.085 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI scoreIncomplete4/5 scored · 5 reviewed

Judge-by-judge reviews

4 of 5 judges returned usable numeric scores. The combined score waits for five. Unknown results are not zero. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure75/100
anthropic/claude-sonnet-5.5
Limited assessment: body explanation shortened; full text retained. brief explanation shortened; full text retained. Largely faithful continuous shot with consistent identity, setting, and a slow push-in. Action is only partly as briefed (no clear trackpad edit or final inspection pause), and the end framing is tight. Usable but imperfect.
GeminiNo61.25/100
google/gemini-4-argon
The clip preserves the subject and environment well, but the camera push is overly aggressive rather than restrained, and the laptop screen changes abruptly before her hand reaches the keys.
GrokYes—
x-ai/grok-4.7
The clip convincingly shows the same woman working at her laptop during a restrained camera push, with stable continuity and realistic lighting. The requested correction sequence is slightly ambiguous, but there is no major defect preventing use as is.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
Output meets core brief: same woman, setting, and lighting. Face, body, and physics are consistent. Minor timing/pace deviation in action sequence prevents perfect brief rating. No scene changes, logos, or audio issues. Visually realistic and temporally stable. Suitable for intended use as is.
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the core visual identity (face, bun, sweater, setting, lighting) and maintains temporal stability. However, key brief requirements are weakly executed: the specified 'restrained slow camera push' is essentially absent (static framing), the 'small deliberate edit using the trackpad' is not clearly visible, and the 'pause to inspect the screen' moment is not distinctly captured.

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: 439245f3c9323bf73d54dbc72c3cfe78cb74ea12473cd3ac3a33b465b964597a

Fixed 5 sec / 480p / 4 steps; no upscale; prompt expansion disabled.

Kandinsky 6 Lite · 4 steps · take 2

864 × 480 · 5.042s · 24 fps

References used

Est. $1.01/minEstimated · $0.085 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI score65/1005/5 reviewed

Judge-by-judge reviews

0 Yes · 2 No · 3 Unsure. Range 57.5–90/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure65/100
anthropic/claude-sonnet-5.5
Well-preserved identity, scene and lighting with a smooth push-in. The required deliberate edit, typing correction and inspection pause are weakly executed, so use as-is for the brief is doubtful.
GeminiNo57.5/100
google/gemini-4-argon
The clip preserves the subject's likeness, environment, and slow camera push, but fails to cleanly execute the typing action due to a smeared, disappearing left arm at 1.49s.
GrokNo66.25/100
x-ai/grok-4.7
The clip preserves the woman and evening home setting convincingly, but the requested trackpad edit and typing are not clearly shown. An abrupt framing change and unstable laptop display undermine continuity, so it should not be accepted as-is.
QwenUnsure90/100
qwen/qwen3-vl-235b-a22b-instruct
The output preserves face, body, visual quality, and temporal continuity at a high level. However, the core action sequence (editing, typing, pausing) and the specified slow camera push are not clearly observable or verifiable from the sampled frames. Without clear evidence of the intended motion and framing change, it is uncertain whether the brief is fully met. No major defects, but insufficient
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the woman's identity, appearance, and setting well, with stable continuity and realistic lighting. However, the brief specifies a 'restrained slow camera push' which is not clearly executed—the framing appears largely static. The specific action sequence (trackpad edit → type correction → pause to inspect) is not distinctly readable; hand movements appear generic typing rather

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 70/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: 3c60023e8b9911ac337af122da1fbc3e6e6fc0de679ed269a2215363b5414b41

Fixed 5 sec / 480p / 4 steps; no upscale; prompt expansion disabled.

Settings & evidence

$0.085 estimated per take; taxes and invoice unknown.

Provider endpoint documentation ↗

fal-ai/kandinsky6-lite/image-to-video

Fixed 5 sec / 480p / 4 steps; no upscale; prompt expansion disabled.

Two playable submissions. Human acceptance is pending; settings and native sizes differ.

Take 1 silent-file SHA-256: 439245f3c9323bf73d54dbc72c3cfe78cb74ea12473cd3ac3a33b465b964597a

Take 2 silent-file SHA-256: 3c60023e8b9911ac337af122da1fbc3e6e6fc0de679ed269a2215363b5414b41

MiniMax H3 Max Turbo ↗

5/5 AI reviewsEst. $1.31–$1.44/min

MiniMax H3 Max Turbo · take 1

1344 × 768 · 6.583s · 24 fps

References used

Est. $1.31–$1.44/minEstimated · $0.144 – $0.158 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI scoreIncomplete4/5 scored · 5 reviewed

Judge-by-judge reviews

4 of 5 judges returned usable numeric scores. The combined score waits for five. Unknown results are not zero. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeNo61.25/100
anthropic/claude-sonnet-5.5
Limited assessment: brief explanation shortened; full text retained. Identity, scene and camera push are largely preserved, and the look is realistic. But the requested edit, typing and pause beats are weak, and the background has noticeable drift and added objects.
GeminiYes75/100
google/gemini-4-argon
The video preserves the reference subject and setting well, executing a smooth camera push and a natural hand transition to the laptop keyboard near the end.
GrokUnsure—
x-ai/grok-4.7
The clip convincingly shows the woman working at the dining table during a restrained push-in, but the requested trackpad action is not clearly demonstrated. Missing reference evidence prevents a likeness decision.
QwenYes90/100
qwen/qwen3-vl-235b-a22b-instruct
The output meets most brief requirements: face, body, and visual quality are preserved at target level. Brief execution is strong but lacks clear action delineation (edit/type/pause). Physics shows minor contact ambiguity. Temporal continuity is flawless. No new people, scene changes, or audio. Recommended AS IS for illustrating the agency owner’s Sunday work.
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the core visual identity (face, bun, charcoal sweater, setting, lighting) and maintains temporal continuity across the 6.5s clip. However, the brief's specific action sequence—deliberate trackpad edit, typing correction, pause to inspect—is not clearly executed; hand motion appears generic and repetitive rather than narratively specific. The required 'restrained slow camera p

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: b32657a59e54c5f991207e4ccb5a68bdd0b0e8a822a70517c8ed9ecb15303200

6 sec / 768P; balanced prompt expansion. Native audio removed for playback.

MiniMax H3 Max Turbo · take 2

1344 × 768 · 6.583s · 24 fps

References used

Est. $1.31–$1.44/minEstimated · $0.144 – $0.158 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI score70/1005/5 reviewed

Judge-by-judge reviews

2 Yes · 1 No · 2 Unsure. Range 65–95/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure70/100
anthropic/claude-sonnet-5.5
Limited assessment: body explanation shortened; full text retained. Good continuity of identity, setting and mood with a smooth slow push. The requested trackpad edit and typed correction are weakly shown, since the action is mostly a static pose. Acceptable but not clearly meeting the brief.
GeminiNo70/100
google/gemini-4-argon
The clip maintains the reference visual quality, character likeness, and slow camera push, but misses the requested typing correction action as the subject keeps one hand on her chin throughout.
GrokYes86.25/100
x-ai/grok-4.7
A coherent realistic push-in preserves the woman and Sunday-evening workspace while showing restrained laptop editing and inspection. Minor softness in the hands does not undermine the brief.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
Output meets most targets: face, body, visual, physics, and temporal continuity are fully preserved. Brief execution is strong but minor deviation in clearly showing the edit-typing-pause sequence. Camera push is restrained and framing remains readable. No new elements or scene changes. Suitable for use as is.
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the woman's identity, appearance, and setting well, with stable continuity and appropriate lighting. However, the specific action sequence—deliberate trackpad edit followed by typing a correction, then pausing to inspect—is not clearly distinguishable in the sampled frames. The hand motion appears present but lacks the precise narrative clarity requested. Without seeing the un

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: 3777edd7aec871f2d94bd492b3e5a47176d76ecada5d65e60a41651f7c5f082e

6 sec / 768P; balanced prompt expansion. Native audio removed for playback.

Settings & evidence

$0.144–0.158 estimated per take; taxes and invoice unknown.

Provider endpoint documentation ↗

minimax/h3-max-turbo/image-to-video

6 sec / 768P; balanced prompt expansion. Native audio removed for playback.

Two playable submissions. Human acceptance is pending; settings and native sizes differ.

Take 1 silent-file SHA-256: b32657a59e54c5f991207e4ccb5a68bdd0b0e8a822a70517c8ed9ecb15303200

Take 2 silent-file SHA-256: 3777edd7aec871f2d94bd492b3e5a47176d76ecada5d65e60a41651f7c5f082e

Grok Imagine 1.5 Lite ↗

5/5 AI reviewsEst. $1.89–$1.90/min

Grok Imagine 1.5 Lite · take 1

1264 × 720 · 6.042s · 24 fps

References used

Est. $1.89–$1.90/minEstimated · $0.19 – $0.191 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI score70/1005/5 reviewed

Judge-by-judge reviews

2 Yes · 1 No · 2 Unsure. Range 65–90/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure70/100
anthropic/claude-sonnet-5.5
Faithful scene, identity and a slow push-in, but the specified edit, typing and inspection beats are weakly performed; the pose is mostly static. It is usable only if subtle action is acceptable.
GeminiNo70/100
google/gemini-4-argon
The clip preserves the subject's likeness, setting, and lighting well with a smooth camera push. However, the requested typing action is incomplete as her left hand remains near her chin, with minor hand warping near the end.
GrokYes90/100
x-ai/grok-4.7
The clip preserves the woman and Sunday home setting while delivering a restrained camera push and subtle laptop editing. Minor softness in the hand action does not undermine the realistic continuity or brief.
QwenYes90/100
qwen/qwen3-vl-235b-a22b-instruct
Output meets most brief requirements: face, body, visual quality, and continuity are fully preserved. Minor deviations in physics (unclear hand contact) and brief execution (action not fully visible) prevent perfect score. Camera push and lighting match. No new elements or cuts. Suitable for use as is.
MistralUnsure65/100
mistralai/mistral-large-4-0
The output preserves the woman's identity, charcoal sweater, dark curly bun, home dining setting, and Sunday-evening lighting with a restrained slow push-in. Visual quality and temporal continuity are strong. However, the specific brief action sequence—deliberate trackpad edit, typing a correction, then pausing to inspect—is not clearly observable in sampled frames. Hand movements appear present

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: 3dce220c661924f7a1370ddaf243d81d2c335c2cda4448b01b1fe84d812d4a1a

6 sec / 720p; one image input included in estimate. Native audio removed for playback.

Grok Imagine 1.5 Lite · take 2

1264 × 720 · 6.042s · 24 fps

References used

Est. $1.89–$1.90/minEstimated · $0.19 – $0.191 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI scoreIncomplete4/5 scored · 5 reviewed

Judge-by-judge reviews

4 of 5 judges returned usable numeric scores. The combined score waits for five. Unknown results are not zero. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure71.25/100
anthropic/claude-sonnet-5.5
Good likeness and a mood-appropriate realistic push-in, but the scene changes: candles and a table are added and the screen swaps. The trackpad edit and the pause are weakly shown, so use as-is is uncertain.
GeminiYes70/100
google/gemini-4-argon
The clip preserves the subject's likeness, setting, and lighting while executing a smooth camera push. Hand movements on the laptop and an abrupt screen change at 4.47s are minor flaws, but the output is usable.
GrokNo—
x-ai/grok-4.7
The clip preserves the woman, home setting, and restrained camera push, but the required deliberate edit, brief typing correction, and inspection pause are not clearly readable. It should not be used as is.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
Output meets the brief's core requirements: continuous realistic shot, preserved face/body, correct actions, lighting, and camera movement. Minor timing ambiguity in action sequence doesn't break the narrative. No new people, scene changes, or audio issues. Visual and physical coherence is high. Suitable for illustrating an agency owner working late on Sunday.
MistralYes75/100
mistralai/mistral-large-4-0
Strong execution of the brief: single continuous shot preserves the woman, her dark curly bun, charcoal sweater, dining table setting, laptop, and warm Sunday-evening lighting. The slow camera push is restrained and keeps face and hands readable. The trackpad edit and typing action are present though subtle; the pause-to-inspect beat is visible. No new people, scene changes, speaking, music, capt,

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: 519ceba865cc9d571e56c44386aea92e254f6a0ef7b0aadf330711e504ff5edc

6 sec / 720p; one image input included in estimate. Native audio removed for playback.

Settings & evidence

$0.19–0.191 estimated per take; taxes and invoice unknown.

Provider endpoint documentation ↗

xai/grok-imagine-video/v1.5/lite/image-to-video

6 sec / 720p; one image input included in estimate. Native audio removed for playback.

Two playable submissions. Human acceptance is pending; settings and native sizes differ.

Take 1 silent-file SHA-256: 3dce220c661924f7a1370ddaf243d81d2c335c2cda4448b01b1fe84d812d4a1a

Take 2 silent-file SHA-256: 519ceba865cc9d571e56c44386aea92e254f6a0ef7b0aadf330711e504ff5edc

LTX 2.5 Fast ↗

5/5 AI reviewsEst. $5.36–$5.40/min

LTX 2.5 Fast · take 1

1280 × 720 · 6.042s · 24 fps

References used

Est. $5.36–$5.40/minEstimated · $0.54 – $0.544 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI score70/1005/5 reviewed

Judge-by-judge reviews

2 Yes · 2 No · 1 Unsure. Range 57.5–95/100. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure70/100
anthropic/claude-sonnet-5.5
Limited assessment: brief explanation shortened; full text retained. Good scene and identity preservation with a smooth push-in. The action is ambiguous: no clear trackpad edit or typing correction, and minor background drift (new vase and glass). Usable but not a clear match as is.
GeminiNo57.5/100
google/gemini-4-argon
The clip preserves the subject's likeness, environment, and slow camera push, but fails the action brief and exhibits temporal defects. Instead of editing via the trackpad and typing, the woman reaches out to touch the laptop screen with a warping hand while the screen interface disappears and reappears.
GrokNo61.25/100
x-ai/grok-4.7
The woman, home setting, lighting, and slow push are preserved convincingly, but the specified editing-and-typing action is unclear. Abrupt hand-motion and screen changes prevent using the clip as is.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
The output successfully preserves the woman’s face, body, and environment with high fidelity. The realistic lighting, textures, and slow camera push align with the brief. Minor deviation: typing motion is not clearly visible, though hand placement suggests it. No new people, scene changes, or audio issues. Physics and continuity are coherent. Suitable for illustrating an agency owner working late.
MistralYes90/100
mistralai/mistral-large-4-0
Strong execution of the brief. The generated video preserves the woman's face, dark curly bun, charcoal sweater, home dining table setting, laptop, and Sunday-evening lighting. The slow camera push is restrained and keeps face/hands readable. The action sequence—trackpad edit, typing correction, pause to inspect—is present though the hand gesture at ~2.5-3.5s slightly deviates from pure 'deliber

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: 65496f5f5de0ff728e2c2b6f2f7c02261f7e70eb6a2b039f1067bd91df75e0e2

6 sec / 720p / 24 fps; native audio disabled.

LTX 2.5 Fast · take 2

1280 × 720 · 6.042s · 24 fps

References used

Est. $5.36–$5.40/minEstimated · $0.54 – $0.544 for this test

Cost basis

Historical per-take estimate; normalized using delivered seconds.

AI scoreIncomplete4/5 scored · 5 reviewed

Judge-by-judge reviews

4 of 5 judges returned usable numeric scores. The combined score waits for five. Unknown results are not zero. AI diagnostics; community votes and Miguel’s verdict stay separate.

ClaudeUnsure75/100
anthropic/claude-sonnet-5.5
Largely faithful continuous shot with a stable woman, outfit and lighting plus a slow push-in. Weaknesses: minor background drift, a faster-than-restrained push, and an unclear trackpad edit and pause.
GeminiNo65/100
google/gemini-4-argon
The clip preserves the subject's likeness, setting, and slow camera push well, but misses the requested typing action after using the trackpad.
GrokNo—
x-ai/grok-4.7
The realistic scene and slow inward camera movement are convincing, but the requested sequence of a distinct trackpad correction, brief typing, and inspection pause is not clearly executed. It should not be used as-is for the brief.
QwenYes95/100
qwen/qwen3-vl-235b-a22b-instruct
The output largely fulfills the brief: face, body, lighting, and scene are preserved with high fidelity. The camera push is subtle but present. The core action sequence (edit, type, pause) is implied through hand motion and gaze but not explicitly detailed in the sampled frames. No new people, logos, or audio are introduced. Minor deviation in action clarity prevents a perfect score, but no major
MistralUnsure71.25/100
mistralai/mistral-large-4-0
The output preserves the core subject (woman, bun, charcoal sweater, dining table, laptop, Sunday lighting) and shows plausible trackpad/keyboard interaction with a slow camera push. However, the brief specifies a 'small deliberate edit' and 'brief correction' that are not clearly discernible in sampled frames—only generic typing is visible. More critically, temporal continuity shows background元素漂

Video frame reviews cover sampled visuals; they do not certify audio, lip sync or every frame.

Earlier reviews · 1

Gemini 3.8 Flash · 75/100
Earlier visual rubric · 2026-10-14. Earlier model/rubric; excluded from this five-judge score.

CommunityVotes are recorded separately below

Voting help

If your browser blocks embedded sign-in, open this review separately ↗. Votes and Credits use the same Navo account.

Prompt, settings & test details

Original silent clip ↗

Silent-file SHA-256: c0862a038268d3890c6a698464c0e63847dceec74d9e585ece76cd036f9bce99

6 sec / 720p / 24 fps; native audio disabled.

Settings & evidence

$0.54–0.544 estimated per take; taxes and invoice unknown.

Provider endpoint documentation ↗

lightricks/ltx-2.5/image-to-video/fast

6 sec / 720p / 24 fps; native audio disabled.

Two playable submissions. Human acceptance is pending; settings and native sizes differ.

Take 1 silent-file SHA-256: 65496f5f5de0ff728e2c2b6f2f7c02261f7e70eb6a2b039f1067bd91df75e0e2

Take 2 silent-file SHA-256: c0862a038268d3890c6a698464c0e63847dceec74d9e585ece76cd036f9bce99

AI picks · provisional

Kandinsky 4-step: lower estimated cost at 480p. H3 Turbo: provisional detail pick at 768p.

Editorial AI picks from one scene, separate from Gemini scores. LongCat’s duration and billing need checking.

Miguel’s pick

Not selected yet.

Starting image & shot brief

Exact opening image: fictional agency owner working on Sunday evening — Same starting image for every clip.

The shot

Sunday-night editing: trackpad, brief typing, then a pause. Gentle camera push-in.

Check face, hand contact, action and background. Silent clips retain native duration.

Prompt, method & costs

Same input; different native recipes

Twelve requests shared the exact image and supplied prompt: two takes per setting, no automatic retries. Kandinsky’s step settings are separate arms; seeds differ. H3 used balanced prompt expansion; Kandinsky and LongCat expansion was off. LTX and Grok’s internal rewriting is unverified.

Supplied prompt

A single continuous realistic shot. Preserve the woman in the starting image: same face, dark curly bun, charcoal sweater, home dining table, laptop and Sunday-evening lighting. She makes a small deliberate edit using the trackpad, types a brief correction, then pauses to inspect the screen. Focused expression with mild fatigue. A restrained slow camera push; keep face and hands readable. No new people, scene changes, speaking, music, captions or logos. This illustrates an agency owner fixing client work on Sunday, paying with her own time.

Opening-image SHA-256: 5c7e242d372f0143c72aeccefe9122dd8ad10e3466e09a6a9ad405f9394d335e

Costs are estimates

Displayed rates divide historical per-take estimates by delivered duration. The source quotes used requested durations of 5 or 6 seconds. LongCat’s 5.4s request returned 11.667s, so its range covers both interpretations. Billing remains unconfirmed.

Account-verified tariffs imply $2.49–$2.53 for all twelve attempts. Final charges, rounding, fees, taxes and refunds are unconfirmed. Human acceptance is pending, so cost per usable clip is unknown.

Review limits

Files passed full decoding and have distinct hashes. Review sampled eight frames per clip and selected playback, not every frame. Names are visible here; the A–L reel hides them until its reveal.

One scene; native duration, size, frame rate, steps and prompt rewriting vary. Clips have no AI upscaling, interpolation, speed changes or trims. This does not establish a universal winner or reliability rate.

Try your shot

Plan a shot and compare results before scaling up.

Open Navo