CRITICALFailure Reference

OpenAI Sora Anatomy Artifact — Refund Guide

Technical Classification

Skeletal Prior Collapse

Skeletal Prior Collapse on Sora occurs when the model fails to maintain joint count and skeletal topology across frames. Sora learns body structure statistically from training video; under specific conditions (fast motion, multiple people, close-up of hands) the prior degenerates and produces output with 6 fingers, doubled arms, or limbs joined at impossible angles. Common in promotional clips of athletes, dancers, and crowd scenes.

How to identify this failure

  • Six or seven fingers on a clearly framed hand
  • Two left arms or two right legs on the same body
  • Limbs joining torso at impossible angles (e.g. arm exits from waist)
  • Joint flickering — knee or elbow appears and disappears between frames
  • Crowd members merging into single multi-headed figure

Real generation examples

Prompt used

"Pianist playing a Steinway grand piano, close-up of hands on keys"

Failure observed @ 0:00 → 0:05

Right hand has 6 fingers throughout the clip, visible on the keys

Prompt used

"Three friends laughing at a cafe table, candid shot"

Failure observed @ 0:02

Centre friend has two left arms — one on the table, one gesturing

Documentation strength

If you need to escalate

HIGH — Anatomy artifacts on paid Sora output are visually undeniable and refund tickets are honoured by OpenAI support when timestamp evidence is provided.

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

  1. 01

    Score your prompt before you generate

    Run your prompt through AVA's pre-flight scoring against the Skeletal Prior Collapse pattern. Green light = generate. Yellow/red = rewrite using the suggested fix before you commit credits.

  2. 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 Skeletal Prior Collapse first appears (e.g. "failure first visible at 1.2s"). Timestamped evidence is significantly stronger than a general complaint.

  3. 03

    Use the correct technical term in your support ticket

    Describe this failure as "Skeletal Prior Collapse". This term maps to a recognised internal workflow in the support system and routes the ticket to the right team.

  4. 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 credits for anatomy artifacts?

Yes — submit a support response ticket citing "Skeletal Prior Collapse" with the generation ID and a still frame showing the artifact. Credit usually issues within 48 hours.

Why does Sora generate bodies with extra fingers or limbs?

Sora has no explicit skeletal model — body structure is learned statistically through diffusion. Under fast motion or hand close-ups, per-finger or per-joint detail is undertrained, so the prior produces a topologically wrong but locally plausible result.

How do I reduce anatomy artifacts on Sora?

Avoid extreme close-ups of hands and feet. Reduce motion magnitude. Specify visible body part counts ("two hands, five fingers each"). AVA flags hand-heavy and crowd prompts as anatomy-risk 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.

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

Related failures across models

If you’re seeing this failure, you may also encounter these on other models:

Pick a different tool for Sora failures

Some prompt shapes will keep failing on Sora. Routing those shots to a different vendor is the cheapest fix.