CRITICALFailure Reference

Luma Dream Machine Face Distortion — Pre-Generation Risk Reference

Technical Classification

Facial Landmark Regression Failure

Luma renders faces well in static frames but struggles to maintain landmark stability across the temporal axis. Eyes drift apart or together, jawlines shift, ears morph in size. This is a Facial Landmark Regression Failure — the model's per-frame inference is internally coherent but globally drifts. For commercial use (any clip with a recognisable subject), this single failure mode makes a clip unusable. Luma support classifies this as a critical generation defect.

How to identify this failure

✕Eye spacing changes visibly across the clip
✕Jawline reshapes mid-motion
✕Ear size or position drifts
✕Skin texture pixelates or smooths inconsistently
✕Facial expression "lerps" through inconsistent intermediate forms

Real generation examples

Prompt used

"Close-up portrait of a young woman smiling, warm light"

Failure observed @ 0:02

Left eye drifts inward by 0:02, jawline narrows by 0:04, expression mismatch between halves

Prompt used

"Man speaking into camera, documentary lighting"

Failure observed @ 0:01 - 0:03

Mouth shape inconsistent with prompted speech; lip-sync drift through 0:01-0:03

Documentation strength

If you need to escalate

VERY HIGH — Luma support treats face distortion as a critical-tier failure; documentation with timestamps approves quickly.

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 Landmark Regression 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 Landmark Regression 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 Landmark Regression 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

Will Luma support response credits for face distortion?

Yes. Use "Facial Landmark Regression Failure" terminology, cite the Asset ID, timestamp the worst drift moment, and attach the PDF audit report. This category has high precedent for support acknowledgement.

Why does Luma distort faces?

Diffusion video models inherit facial-landmark inconsistency from their training distribution. The denoising step is largely per-frame; temporal regularisation tries to smooth across frames but with limited success on small high-detail features like eyes.

Which Luma prompts highest-risk for face distortion?

Close-up portraits, multi-character scenes, fast head motion, side profiles. AVA's pre-flight check flags these patterns.

Catch it before you generate

AVA scores this failure mode against your prompt in real time

Free Chrome extension. Analyzes your prompt as you type, flags failure-prone patterns specific to this model, and tells you what to rewrite — before you commit credits to a generation that will fail.

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