MAJORFailure Reference

Pika Labs Prompt Adherence Failure — Pre-Generation Risk Reference

Technical Classification

Compositional Prompt Adherence Failure

Pika 1.5's text encoder silently truncates compositionally complex prompts before generation. Distant or rare clauses are dropped from the latent — the output looks fine on inspection but ignores specific elements the user asked for. The failure is reproducible on prompts that exceed ~25 words or include 3+ distinct visual entities.

How to identify this failure

✕Last clause of prompt has no visible effect
✕Named objects missing entirely
✕Setting reverted to a generic version of the request
✕Multi-subject prompts losing one or more subjects
✕Style modifiers ignored

Real generation examples

Prompt used

"Snowy mountain village, candles in windows, smoke from chimneys, fox running across the road"

Failure observed @ 0:00

Village correct but candles missing, smoke absent, fox replaced with a deer

Prompt used

"Vintage diner, jukebox playing, two customers at the counter, waitress refilling coffee"

Failure observed @ 0:00

Diner present but jukebox missing; only one customer rendered; waitress absent

Documentation strength

If you need to escalate

HIGH — Pika support refunds compositional drops when the missing element is explicit 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 Pika support response credits for prompt adherence failures?

Yes — when the dropped element is explicit in the prompt and demonstrably absent in the output. Submit prompt text + generation ID + screenshot.

Why does Pika drop clauses from my prompt?

The text encoder compresses the prompt to a fixed latent dimension. Content beyond that capacity is silently truncated — typically the most-distant or rarest clause first.

How do I improve Pika prompt adherence?

Keep prompts under 25 words. Put the most important element first. Limit to 3 subjects per prompt. AVA pre-flights compositional risk.

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