It will not start at all, and you have credits
Different from a failure - nothing runs, and your credits are untouched. The most common cause on a newer account is that your first 25 credits are spent. Every account can spend 25 credits with no card on file; after that, having credits is not enough on its own and generating needs a verified card. The message reads:
You have used your 25 free credits. Add a payment method to keep creating - it is a $0 check and you are not charged now or later unless you choose to buy something. Bundles start at $4.99 and never expire.
It is a $0 check, not a payment. Add a card at /add-payment-method and creating works right away. Your 80 quest credits, streak credits and bundle credits are all still there and all still free. Accounts created before the cutoff are grandfathered and never see this, and it does not apply if you already have a subscription, a saved card, or have bought a credit bundle.
If instead the message mentions a failed subscription payment, your renewal
was declined - update the card at /billing. See
payment methods.
First: did your credits come back?
If a generation fails on our side, credits are refunded automatically. Check your balance before assuming you were charged. If you see the credits back, the system worked as intended - you just hit a transient error.
If credits were not refunded and the generation clearly failed, contact support with the generation ID (it is in the URL of the failed result page).
Common failures and what causes them
"Generation timed out"
The upstream provider (usually KIE.ai, our AI gateway) took too long to respond. This happens more often during peak hours.
Fix: wait 30 seconds and retry. If the same model fails twice in a row, try a different model - there is usually a working alternative with similar quality.
"Model is temporarily unavailable"
Specific models go through upstream outages. Known patterns:
- Grok Imagine - intermittent 404s during high load. Our frontend auto-retries, but if you hit it, try Nano Banana or Flux Pro instead.
- Infinitalk lip sync - currently 100% offline, hidden from the dropdown. For a talking character, use a native-audio engine instead: Grok, or the Seedance 2.x tiers in UGC Studio, which speak the line as part of the render at no extra cost over a silent clip. Kling Avatar remains an option when you are driving a face from an audio file you already have.
- Image-to-Video models (Kontext Pro/Flex, Seedream V4 Edit, Qwen Edit) - hidden because output quality is not reliable yet.
"Content moderation blocked this"
The upstream provider flagged your prompt or reference image. This is a hard stop - no credits spent. Rewrite the prompt or swap the reference.
"NSFW content detected"
Our moderation layer or the upstream model blocked it. Same as above - rewrite and retry.
Image-to-video generation produces garbage
The first frame is the biggest factor. Use a clean, high-resolution input image with a clear subject. Busy backgrounds confuse motion models.
When to retry vs. when to switch models
Retry the same model if:
- The error says "timeout", "rate limited", or "temporary"
- The result came back empty but no explicit error was shown
- You are getting a sporadic "network error" toast
Switch models if:
- Two retries both failed the same way
- The error says "unavailable" or "unsupported"
- Your output quality is consistently poor for this use case
Audio upload failures
If Soul Forge or a Cover generation rejects your audio file, see the audio-upload-issues article.
What to include when reporting a persistent failure
If the failure repeats and you need help, tell us:
- The model name you tried (exact name from the dropdown)
- The generation ID (from the URL or failed result card)
- The time it happened (roughly - "just now" is fine)
- Whether your credits came back
With those four things, we can usually reproduce and fix within a day.