Generation Task Troubleshooting Guide

When a generation task fails or gets blocked, this guide tells you what to do about it. Start with the quick reference table, then jump to the relevant section for details.

Applies to: all generation tasks (Image, Video, Audio, etc.) Last updated: 2026-05-25

---

Quick Reference

---

Part 1 — Things you can fix Yourself

This category covers the majority of failed tasks. Most are quick fixes.

1. Insufficient_credits

What it means: Your balance, subscription quota, or free tier doesn't cover this task.

What to do:

• Open the Account / Billing page to check your balance and usage.
• Top up or upgrade to a higher tier.
• Need to save credits: pick a lower spec (lower resolution, shorter duration, cheaper model).

---

2. billing_error

What it means: A billing-related issue other than insufficient credits. Most cases are transient.

What to do:

• Retry the task once — many billing_error cases clear up on retry.
• If neither helps, contact support with your task ID.

---

3. invalid_parameters

What it means: One of your parameters is outside what the model accepts. Common cases:

• Prompt length — your prompt exceeds the model's character limit (e.g. Kling caps prompts at 2500 characters). Fix: shorten the prompt.
• duration — value not in the model's allowed enum (e.g. the model supports only 5s and 10s, but you sent 7s). Fix: use an allowed value.
• resolution / size — unsupported dimensions for this model. Fix: match the model's spec.
• reference_assets count — too many reference assets. Fix: reduce to the allowed limit.
• Other model-specific fields — value out of range. Fix: check the model detail page for supported values. What to do: The error message usually names the offending field. The model detail page lists every supported parameter range.

---

4. image_too_large

What it means: Your input image's resolution or file size exceeds the model's limit.

What to do:

• Compress the image (keep quality while shrinking file size).
• Downscale to a supported resolution.
• Switch to a model that accepts larger inputs.

---

5. media_duration_exceeded

What it means: Your reference video or audio exceeds the model's duration limit. The limit varies by model — check the model's documentation for the specific cap.

What to do:

• Trim the asset in any editor.
• Use just the key clip as your reference.

---

6. unsupported_media_format

What it means: The file format of your input isn't supported by this model.

What to do:

• Convert the file to a supported format. Check the model's documentation for the full list.
• Safe defaults that most models accept: JPEG / PNG for images, MP4 (H.264) for video, MP3 / WAV for audio.

---

7. media_load_error

What it means: The platform couldn't load your asset. The file may be corrupted, the URL unreachable, or the upload incomplete.

What to do:

• Re-upload the asset.
• If you provided a URL, verify the link is publicly accessible (no auth required).
• Make sure the file opens normally on your end before uploading.

---

8. reference_aspect_ratio_invalid

What it means: The aspect ratio of your reference asset doesn't match what the model or target output requires.

What to do:

• Crop the reference to match the target ratio.
• Or change the requested output aspect ratio to match the reference.
• Check the model's documentation for required ratios (commonly 1:1, 16:9, 9:16).

---

9. content_policy_violation

What it means: A content safety policy was triggered. Two things can cause this:

1. Your input (prompt or reference asset) contains something sensitive.
2. The model's generated output contains something sensitive — even if your prompt looked fine, the result may have come out as a recognizable celebrity face, a copyrighted character, or otherwise infringing content. Common sensitive areas:

• Sexual or suggestive content
• Violence, gore, self-harm
• Inappropriate content involving minors
• Celebrity likeness / IP infringement
• Politically sensitive figures or events
• Hate speech, discrimination What to do:

If your input is the problem:

• Rewrite the prompt — remove sensitive terms, use neutral descriptions.
• Swap reference images — avoid real celebrity photos, photos of children, violent or graphic imagery. If the output is the problem (input looked fine but result got blocked): This usually means the model drifted toward a recognizable IP or real person. Guide the model toward original content:
• Describe traits and style instead of naming someone or something protected. Bad: "a cartoon mouse like Mickey." Good: "an original cartoon mouse with large round ears, red shorts, and white gloves, in a 1930s rubber-hose animation style."
• Add modifiers like "original character," "fictional," or "generic" to push the model away from known IP.
• For celebrity styles, describe the look ("vintage Hollywood portrait," "1990s grunge aesthetic") rather than the person.
• Anchor the output with specific, unique details (a fictional name, an unusual color palette, distinctive features) so the model produces something new instead of a known likeness. If neither approach works after a few tries: that creative direction isn't supported on the platform — try a different angle.

This isn't a system bug. It's the platform's content safety line. Retrying the same input will give you the same result.

---

10. no_output

What it means: The model ran but produced nothing usable. Usually the prompt was too complex, contained conflicting constraints, or the reference was too far from the target.

What to do:

• Simplify the prompt — drop secondary modifiers, keep the core intent.
• Resolve conflicts — e.g. don't ask for "minimal" and "highly detailed" at once.
• Try a closer reference image.
• Rephrase and try again.

---

Part 2 — Temporary failures, just Retry

11. Temporary failures (retry)

These error codes all fall in this bucket:

• timeout
• generation_failed
• service_unavailable
• capacity_exceeded
• input_asset_preparation_timeout
• queue_wait_timeout What it means: Temporary congestion, a transient provider glitch, or a network hiccup. Has nothing to do with your input.

What to do:

• Single task: wait 20–60 seconds and retry.
• Batch jobs: use exponential backoff (30s, then 1min, then 2min...). Don't hammer retries back-to-back.
• If the same model keeps failing, it may be having a partial outage — switch to a comparable model to get unblocked.

---

Part 3 — Permission and quota

12. forbidden

What it means: Could be one of several things:

• Your account doesn't have access to this model.
• The model provider's quota is temporarily exhausted (platform-wide).
• The model isn't available in your region. What to do:
• Try a comparable alternative model first.
• Retry later (provider quota issues usually clear up).
• If it persists across multiple models, contact support.

---

Part 4 — When you need us

13. When to contact support

These errors are not something you can fix. Reach out to support with your task ID:

• internal_error — platform internal error
• service_configuration_error — service misconfiguration
• scheduler_error — scheduling problem
• queue_error — queue problem
• Persistent forbidden that doesn't go away when you switch models Include in your support request:

1. Task ID (copy from the task detail page)
2. Time of the error (to the minute)
3. Model used
4. Screenshot of the full error message

---

Pre-submission Checklist

Most failures can be avoided by a 30-second check before clicking "Generate":

• Is your balance enough for the estimated cost of this task?
• Does the prompt touch sensitive content (sexual / violent / minors / celebrities / politics)?
• Are reference images within the model's size and resolution limits?
• Is the reference video within the model's stated duration limit?
• Is the file format supported (JPEG/PNG, MP4, MP3/WAV)?
• Are duration, size, and aspect_ratio set to values the model supports?
• Is the number of reference assets within the model's cap?
• Does the reference image's aspect ratio match the target output?

---

If you have worked through this guide and still can't resolve the issue, reach out to support with your task ID — we can usually pinpoint the cause quickly.