Seedance (ByteDance) Prompt Adherence Failure — Refund Guide
Technical Classification
Cross-Attention Prompt Drop
Cross-Attention Prompt Drop on Seedance is a known limitation of ByteDance's diffusion model. The model uses a cross-attention text encoder to condition generation on the prompt; under specific prompt structures (long prompts, prompts with multiple subjects, or prompts mixing concrete + abstract concepts) the cross-attention can drop entire prompt clauses. The output then reflects only a subset of the prompt with the rest silently ignored.
How to identify this failure
- ✕Output shows wrong subject (asked for cat, got dog)
- ✕Setting completely different from prompt (asked for beach, got forest)
- ✕Multiple-subject prompts produce only one subject
- ✕Action verb ignored (asked for running, got standing)
Real generation examples
Prompt used
"A red panda and a fox playing chess at a wooden table in a forest"
Failure observed @ 0:00 → 0:05
Output shows a single red panda alone at a table — fox and chess board absent
Documentation strength
If you need to escalate
HIGH — ByteDance Seedance support refunds prompt-adherence failures on paid output with the prompt and output evidence.
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 Cross-Attention Prompt Drop 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 Cross-Attention Prompt Drop 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 "Cross-Attention Prompt Drop". 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 Seedance refund credits for ignored prompts?
Yes — prompt-adherence failures on paid Seedance output are refundable. Cite "Cross-Attention Prompt Drop" with the prompt and the failing clip.
Why does Seedance ignore parts of the prompt?
The cross-attention text encoder can drop entire prompt clauses under long or compositionally complex prompts. There is no signal in the output that this happened.
How do I improve Seedance prompt adherence?
Keep prompts short and single-subject. Put the most important element FIRST. Avoid mixing abstract and concrete concepts in the same 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.
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.
Related failures across models
If you’re seeing this failure, you may also encounter these on other models:
Semantic
Sora output ignores or contradicts explicit prompt instructions — wron…
Text
Runway output ignores the prompt entirely, producing generic motion un…
Camera-Conditioning
Veo output uses a static camera or generic camera motion instead of th…
Text
Luma output partially or completely ignores the prompt — wrong subject…
Prompt-to-Generation
Output ignores key prompt specs — wrong subject, missing actions, wron…
Facial
Asymmetric eye placement, facial morphing across frames, expression dr…
Pick a different tool for Seedance failures
Some prompt shapes will keep failing on Seedance. Routing those shots to a different vendor is the cheapest fix.