H3 Omni Ref Iteration Review

Gemini Omni dropped · Original H3 · 4×GB200 optimized FastVideo path

Updated 2026-08-30 17:58 PDT · no auto refresh

Audio Repair 5 · No Invented Speech

Five A3/B3 retries with the same reference workloads but an explicit non-verbal contract: mouths at rest, no spoken dialogue, and ambient room tone only.

Generating 1 / 5 · QA 0/1 accepted · unintelligible speech 1
ref2va-sample-2026083600-000000

ref2va-sample-2026083500-000000

Make the woman from the photo perform the hand gestures and upper body movements from the video while keeping her in her own kitchen.
motion_performance_transferdirect ablation10sgeneratedno_speechrejected
<Picture 1>
<Picture 1>subject_identity
image
<Video 1>motion_performance
video · 10.0s
Exact H3 prompt
A 10-second static medium shot of the woman from <Picture 1> standing behind the counter in her modern kitchen setting. She retains her exact visual identity, blonde hairstyle, and black blazer over a colorful top from <Picture 1>, while performing the non-verbal upper-body motion, hand gestures, hand clasping, and forward pointing sequence from <Video 1>. Her mouth remains naturally closed and at rest throughout the performance. The camera remains fixed at eye level matching <Picture 1>. No spoken dialogue; ambient room tone only.
Final delivered output
REJECTED
Motion transfer applied, but generated unauthorized speech violating the no-speech constraint and leaked the source video's background environment.
failure: unexpected_speech, unintelligible_speech, source_environment_leakage
media contract: 10.176s actual / 10.0s target · audio present · pass
primary_transformation: 2source_preservation: 2target_identity: 3source_leakage: 4motion_or_temporal: 3audio: 2artifact: 3
Gemini 3.7 independent output QA · 414.0s generation
ref2va-sample-2026083600-000001

ref2va-sample-2026083500-000005

Animate the man from the picture using the head movements, downward glances, and subtle nods from the video, keeping him seated in his studio setup with no spoken dialogue.
motion_performance_transferdirect ablation5sgeneratedno_speechpending
<Picture 1>
<Picture 1>subject_identity
image
<Video 1>motion_performance
video · 5.0s
Exact H3 prompt
Generate a 5-second realistic video of the man from <Picture 1> seated in his studio setup. Animate him by transferring the head tilts, downward glances, subtle nods, and upper-body posture timing from <Video 1>. Maintain his exact appearance, clothing, headphones, and cluttered studio background from <Picture 1>. His mouth remains naturally at rest with no spoken dialogue; ambient room tone only.
Final delivered output
H3 generation in progress
PENDING
QA pending
Gemini 3.7 independent output QA
ref2va-sample-2026083600-000002

ref2va-sample-2026083500-000011

Generate a video placing the woman from Picture 1 and the man from Picture 2 together in the modern reception lobby from Picture 3, interacting non-verbally.
multi_subject_interactiondirect ablation12sgeneratedno_speechpending
<Picture 1>
<Picture 1>subject_identity
image
<Picture 2>
<Picture 2>subject_identity
image
<Picture 3>
<Picture 3>environment
image
Exact H3 prompt
A 12-second cinematic shot set in the modern reception interior from <Picture 3>. The woman from <Picture 1> and the man from <Picture 2> stand near the white counter. The woman glances over and gestures subtly while the man turns toward her, nodding politely. The camera slowly tracks backward in a smooth medium-wide framing. Natural indoor lighting with ambient room tone and no spoken dialogue.
Final delivered output
H3 generation in progress
PENDING
QA pending
Gemini 3.7 independent output QA
ref2va-sample-2026083600-000004

ref2va-sample-2026083500-000014

Transfer the head movements, subtle tilts, and facial expression timing from the man in the video to the man with long hair in the photo, keeping his dark room background, with no spoken dialogue.
motion_performance_transferdirect ablation8sgeneratedno_speechpending
<Picture 1>
<Picture 1>subject_identity
image
<Video 1>motion_performance
video · 8.0s
Exact H3 prompt
Animate the man with long hair from <Picture 1> in his dark room setting, transferring the subtle head turns, nods, and facial gesture timing from <Video 1> across 8 seconds. Keep the man's mouth naturally at rest with no spoken dialogue; ambient room tone only. Exclude the identity, clothing, background, and speech from <Video 1>.
Final delivered output
H3 generation in progress
PENDING
QA pending
Gemini 3.7 independent output QA
ref2va-sample-2026083600-000005

ref2va-sample-2026083500-000016

Create a 10-second video of the man from <Picture 1> and the woman from <Picture 2> interacting in the reception area from <Picture 3>. The woman stands behind the reception desk while the man approaches the counter, pauses, and gestures politely toward the paperwork on the desk. They interact naturally through body language and nods without spoken dialogue.
multi_subject_interactiondirect ablation10sgeneratedno_speechpending
<Picture 1>
<Picture 1>subject_identity
image
<Picture 2>
<Picture 2>subject_identity
image
<Picture 3>
<Picture 3>environment
image
Exact H3 prompt
A 10-second realistic video set in the reception office from <Picture 3>. The woman from <Picture 2> stands behind the wooden reception desk, holding her black clutch on the counter. The man from <Picture 1> enters from the hallway, approaches the front of the desk, and stops. He places a hand on the counter and gestures politely toward a visitor ledger. The woman nods with a welcoming, professional expression and gestures toward the sign-in sheet. Both characters interact purely through subtle non-verbal gestures and expressions with mouths resting naturally; no spoken dialogue, with gentle indoor office ambient room tone only.
Final delivered output
H3 generation in progress
PENDING
QA pending
Gemini 3.7 independent output QA