Iris · synthetic-character track.
Prompt, method & limits
Three subjects, two models, two requested durations: 12 submissions, all delivered. One attempt per setting; no retries, trimming or manual upscaling. Real-identity and synthetic results stay separate.
Kling used its 720p setting; H3 used the separately approved provider-2K setting. H3’s native/upscaled status and both models’ internal prompt processing are unverified. This compares these workflows, not equal-resolution model performance.
Check the face, hair, clothing, hands, table contact and background against the exact generated starting frame. This is first-frame continuity, not a measured match to Miguel’s original photos. Original photos and voice recordings were not submitted.
The prompt requests a closed mouth and no speaking. Native audio is retained but excluded from the visual test. This is not a controlled lip-sync test.
All files pass full decoding. Kling returned one extra frame; H3 returned three or four extra frames and fails the frozen one-frame duration tolerance. These delivery checks are separate from visual quality, AI judging and human acceptance.
Quotes use requested seconds: Kling $9.42/min; H3 $7.80/min. Combined generation estimate $12.915. Actual per-job charges, fees, taxes, refunds and cost per usable clip are unverified. One sample per subject/duration cannot establish a reliability rate or universal winner.
Exact supplied prompt
One continuous locked-camera podcast shot from the supplied first frame. Preserve the exact person, facial proportions, hair, beard if present, wardrobe, chair, microphone, table and lighting. The seated person naturally shifts their gaze from the off-camera interviewer toward the camera, turns the head slightly toward camera, blinks and gives one small nod. They lift one hand a few centimeters in a relaxed conversational gesture, then rest it back on the table. Natural breathing and subtle shoulder movement. Keep eyes and face visible, preserve fingers and table contact. Relaxed closed mouth; no speaking. No cuts, zooms, added people, text, logos or new objects.
