CRITICALFailure Reference

OpenAI Sora Face Distortion — Refund Guide

Technical Classification

Facial Identity Drift

Facial Identity Drift on Sora occurs when the diffusion process fails to preserve the subject's facial geometry across the temporal axis. The model encodes identity through a low-dimensional latent that is supposed to remain stable across all frames of a single generation — but on close-up or rotating-head shots, this latent can drift, causing the subject to appear as multiple slightly-different people through the duration of the clip. Distinct from anatomy artifact: the face is anatomically valid in each frame, but the identity changes.

How to identify this failure

✕Subject's eye color shifts mid-clip
✕Nose or jawline geometry morphs between frames
✕Subject appears to age or de-age within a single clip
✕Identity drifts into a different person on profile turns
✕Skin tone or facial proportions change without prompt direction

Real generation examples

Prompt used

"Close-up of a woman smiling, slowly turning her head left to right, cinematic"

Failure observed @ 0:03

Identity drifts into a visibly different person at 0:03 during the head turn

Prompt used

"Portrait of an elderly man speaking to camera, warm lighting"

Failure observed @ 0:02 → 0:04

Eye color shifts from brown to green at 0:02, jawline morphs at 0:04

Documentation strength

If you need to escalate

HIGH — Sora is marketed on character consistency. Identity drift on close-up paid output is a clear feature defect.

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 Drift 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 Drift 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 Drift". 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

Does OpenAI refund Sora identity drift failures?

Yes — submit the clip with timestamps where identity shifts. Cite "Facial Identity Drift" and reference the inconsistency between specific frames. Refunds for close-up character work on Sora are routinely approved.

Why does Sora change a subject's face mid-clip?

Sora's temporal attention has a finite receptive field; over long durations or fast head motion, the identity latent can drift outside its stable basin and resample to a nearby person in the model's training distribution. This is most likely on profile rotations and close-ups.

How do I prevent Sora face drift?

Keep clips under 4 seconds for close-up character work. Avoid fast head turns past 45°. Use reference-image grounding when available. AVA flags high-drift-risk prompt structures before generation.

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.

Install free All 105 failure modes

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.

Claim $50 founders