MAJORFailure Reference

Kling Prompt Adherence Failure — Pre-Generation Risk Reference

Technical Classification

Compositional Prompt Adherence Failure

Kling v1.5 and v2 share a fixed-length text encoder that silently truncates long, compositional prompts. The most-distant or rarest clause in the prompt is dropped from the latent before generation, producing output that looks correct but ignores specific elements the user mentioned. The user has no signal at submission time that compression happened.

How to identify this failure

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

Real generation examples

Prompt used

"Misty forest at dawn, fox running across path, owl perched on tree branch, golden light"

Failure observed @ 0:00

Owl missing; mist absent; fox present but in mid-day lighting

Prompt used

"1920s ballroom, couples dancing, jazz band in the background"

Failure observed @ 0:00

Ballroom present but jazz band missing; couples reduced to one

Documentation strength

If you need to escalate

HIGH — Kling support treats compositional drops as a documentable failure; refunds granted with the prompt text and a screenshot.

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 Kling refund credits for prompt adherence failures?

Yes. Submit the prompt text, the generation ID, and a description of which named element is missing. Kling support routinely refunds when the dropped element is explicit in the prompt.

Why does Kling drop clauses from my prompt?

The text encoder compresses the prompt to a fixed latent dimension. Content beyond that capacity is silently truncated, with the highest-distance or rarest clause typically dropped first.

How do I improve Kling prompt adherence?

Keep prompts under 30 words. Put the most important element FIRST. Limit to 3 distinct subjects per prompt. AVA pre-flights compositional risk before you spend credits.

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