Vidu Face Distortion — Pre-Generation Risk Reference
Technical Classification
Facial Identity Coherence Failure
Facial Identity Coherence Failure on Vidu occurs when the model fails to maintain the same facial identity across the clip's temporal axis. The face slowly drifts toward a different person, features (eyes, nose, mouth) misalign frame-to-frame, or the entire face structure collapses during expression changes. Vidu's Reference-to-Video mode is meant to prevent this — but on prompts with multiple subjects or rapid head turns, the reference identity gets averaged with hallucinated alternatives.
How to identify this failure
- ✕Subject's face becomes a different person by mid-clip
- ✕Eyes misaligned, nose shifts position frame-to-frame
- ✕Reference-to-Video identity drifts despite a clean reference image
- ✕Face structure collapses during smile or speech
- ✕Multiple subjects swap faces between cuts
Real generation examples
Prompt used
"Portrait of a woman smiling, soft window light, cinematic"
Failure observed @ 2.7s
Face structure changed at 2.7s — eye spacing widened, nose flattened, became a different person
Prompt used
"Two friends laughing at a coffee shop, handheld feel"
Failure observed @ 3.4s
Face swap occurred between subjects at 3.4s, both identities lost reference fidelity
Documentation strength
If you need to escalate
HIGH — Vidu support recognizes face-identity drift as a documented failure mode, especially for paid Reference-to-Video credits.
AVA is a pre-purchase prevention tool, not a post-purchase recovery tool. Platforms generally do not guarantee credit refunds for output-quality failures; goodwill credits are at each platform's discretion. The strength rating reflects how well-formed your support ticket can be, not a promised outcome.
Prevention + documentation steps
- 01
Score your prompt before you generate
Run your prompt through AVA's pre-flight scoring against the Facial Identity Coherence Failure pattern. Green light = generate. Yellow/red = rewrite using the suggested fix before you commit credits.
- 02
Capture Generation ID + timestamp if it failed anyway
Find the Generation ID in the URL or share link. Note the exact time when the Facial Identity Coherence Failure first appears (e.g. "failure first visible at 1.2s"). Timestamped evidence is significantly stronger than a general complaint.
- 03
Use the correct technical term in your support ticket
Describe this failure as "Facial Identity Coherence Failure". This term maps to a recognised internal workflow in the support system and routes the ticket to the right team.
- 04
Submit via the correct support channel
Runway has no direct email intake. Pro+ plan: open the in-app AI Assistant (help widget bottom-right of app.runwayml.com), describe the failure with the technical term, attach evidence. Free/Standard plan: human support isn't available — your channel is Discord #community-help with @On Call - Moderators.
Frequently asked questions
Why does Vidu's face drift even with a reference image?
The Reference-to-Video pipeline conditions on the reference, but the temporal coherence module can drift when the subject turns >45° or speaks. The reference signal weakens as the clip progresses.
Which Vidu prompts are highest risk for face distortion?
Long clips (>5s) with talking-head shots, profile turns, multi-subject scenes, and rapid expression changes. AVA pre-flight flags these.
Does Vidu refund face drift on Reference-to-Video?
Yes — and Reference-to-Video has stronger escalation precedent because the user is explicitly paying for identity preservation. Document with the Generation ID and reference-vs-output comparison.
Score your prompt
Score your prompt against this failure mode in 30 seconds
Paste your prompt and the platform you intend to use. AVA returns a red/yellow/green score against this specific failure mode plus a concrete rewrite if the risk is high.
AVA Pro · founders' round
$50 for 6 months of unlimited scoring across all failure modes + personal failure-history dashboard. Locks in $13/mo grandfathered after.
Related failures across models
If you’re seeing this failure, you may also encounter these on other models:
Facial
Asymmetric eye placement, morphological drift across frames, non-Eucli…
Facial
Asymmetric eye placement, jaw drift between frames, non-Euclidean faci…
Facial
Sora output shows face morphing, identity inconsistency, or feature di…
Identity
Google Veo output shows facial feature distortion mid-clip or identity…
Identity
Face morphing, identity drift, asymmetric distortion, features melting…
Facial
Asymmetric eye placement, facial morphing across frames, expression dr…
Pick a different tool for Vidu failures
Some prompt shapes will keep failing on Vidu. Routing those shots to a different vendor is the cheapest fix.
Alternatives
Vidu alternatives
Ranked substitutes by shot type — character, motion, lighting, audio, brand product.
Head-to-head
Vidu vs Luma
Vidu 2.0 (ShengShu) · Luma Dream Machine Ray-2
Head-to-head
Vidu vs Runway
Vidu 2.0 (ShengShu) · Runway Gen-4
Head-to-head
Vidu vs Veo
Vidu 2.0 (ShengShu) · Google Veo 3