AI video generation stuck? How to diagnose a queued or failed render
If an AI video render appears stuck, save the provider’s request ID and check its status before pressing Generate again. The same symptom can come from a pending render, a rejected request or a finished video that has not downloaded.
Find out where the job stopped
Record the model, shot ID, time submitted, exact error and provider request ID, if available. A shot ID identifies your work in the project. A provider request ID identifies the remote operation; use it in the provider dashboard or API before sending a replacement request. If several shots failed at the same time, compare whether they share a provider, account or input setting.
| What you see | What to do first |
|---|---|
| Queued or running with a request ID | Check the existing request. Keep it as one job while the provider is still processing it. |
| Connection timed out after submission | Look for a matching job in the provider history by ID, timestamp and model. A lost response does not prove the service rejected the request. |
| HTTP 429 or temporary capacity error | Pause new submissions for that provider. Check its quota and service status, then reduce concurrency and space out retries. |
| Invalid input, credentials or billing | Read the error message and fix the cited field, key permission or account problem. A retry with the same inputs is unlikely to help. |
| Provider finished, but the file is missing locally | Inspect the existing result and download it again if available. Check the destination folder, free space and write access before rendering again. |
Status names vary by provider. For example, fal documents that its final COMPLETED state can also mean a job failed or was cancelled; check the result or error fields. fal queue status reference.
Retry only after you know what happened
Suppose shot 12 is running, shot 13 returned a 429 and shot 14 finished at the provider but the MP4 is missing. Leave 12 in progress. Wait before retrying 13 after reducing the pressure on that provider. For 14, recover the existing output if possible. Restarting all three sends new work for shots that may already be running or billed.
For temporary 429 or 503 responses, Google Cloud recommends increasing the delay between attempts and adding variation so retries do not arrive together. Follow the specific provider’s limits and any retry instructions it returns. A depleted balance or a hard quota needs an account or workload change before continuing. Google Cloud guidance for 429 errors.
Resume the batch on Windows
- Keep the current project folder and inspect the queue after reopening it. Check which files are already on disk before replacing them.
- Save the error and request ID for the affected shot. Keep API keys, credentials and private reference files out of screenshots you share.
- After fixing the cause, test with one affected shot. Open the output and check that the duration and format match your request before resuming the remaining work.
Scena Studio stores the batch on disk and supports per-service submission controls. See the batch queue settings before changing retry counts or concurrency for a longer run.