MAJORFailure Reference

Hailuo AI Prompt Adherence Failure — Pre-Generation Risk Reference

Technical Classification

Compositional Prompt Adherence Failure

Hailuo's text encoder collapses long, multi-clause prompts to a fixed-length latent before generation. When the prompt exceeds the encoder's effective compositional capacity, entire clauses are silently dropped — the output looks fine but ignores the rarest or last-mentioned elements. The failure mode is highly reproducible and the user has no signal that compression happened.

How to identify this failure

✕Output ignoring an entire mentioned subject
✕Last clause of the prompt has no visible effect
✕Wrong setting (e.g. asked for "rainy night", got "sunny day")
✕Specific objects mentioned in prompt are missing entirely
✕Output looks generic despite a highly specific prompt

Real generation examples

Prompt used

"Cyberpunk street at midnight, neon signs, rain falling, lone figure under umbrella, dog beside them"

Failure observed @ 0:00

No dog generated; rain absent; umbrella present but figure altered

Prompt used

"Vintage 1970s diner, jukebox playing, three customers at the counter"

Failure observed @ 0:00

Diner correct but only one customer rendered; jukebox missing

Documentation strength

If you need to escalate

HIGH — compositional drops are documentable; Hailuo support refunds when the missing element is named in the prompt.

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 Compositional Prompt Adherence 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 Compositional Prompt Adherence 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 "Compositional Prompt Adherence 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

Does Hailuo refund credits for prompt adherence failures?

Yes — when the dropped element is explicit in the prompt and absent in the output, Hailuo treats the failure as a deliverability defect. Submit the prompt text, the generation ID, and a screenshot.

Why does Hailuo drop prompt clauses?

The text encoder compresses the prompt into a fixed latent dimension. Compositional content beyond the encoder's effective capacity is silently truncated; there is no warning at submission time.

How do I improve Hailuo prompt adherence?

Keep prompts short and single-subject. Put the most important element first. Avoid mixing more than 3 distinct elements per prompt. AVA pre-flights compositional risk before submission.

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