SupaPic
SupaPic / Blog / What to Do When an AI Video Generation Stops at 90–99 Percent
AI video stuck at 99 percent

What to Do When an AI Video Generation Stops at 90–99 Percent

A progress bar is not the same as a completed video. Here is how to tell whether generation, delivery or your app’s status tracking failed.

What to Do When an AI Video Generation Stops at 90–99 Percent

A video job sitting at 94% is frustrating because it looks so close to finished. But a percentage on the screen is usually an estimate from the application, not a precise measurement of the model's internal work.

There are several stages between “generation started” and “playable MP4”: the provider accepts the task, processes it, writes the output file, returns a result URL, and the app stores or serves that output. A failure in the last step can look like a model failure even when the expensive part actually succeeded.

Check for an output before retrying

If your provider dashboard shows a completed request and an accessible result URL, do not immediately start another paid job. The issue may be in the application's finalization or media delivery path. If the provider itself reports a failed task, inspect the exact error rather than relying on the frontend percentage.

For apps with request IDs, save the ID before refreshing. That identifier is more useful for debugging than a screenshot of “94%”.

Common causes that look similar

  • Input file unavailable: the provider could not retrieve a private or expired media URL.
  • Processing rejection: input or output content did not pass the provider's validation.
  • Polling issue: the app stopped requesting status updates or misread a provider response.
  • Result delivery issue: the output exists but the app cannot download or stream it.

Each category needs a different fix. Replacing an image will not repair a broken media redirect. Increasing a timeout will not remove an unsupported parameter.

Treat retries carefully

A safe app should distinguish “confirmed failure” from “unknown status.” Refunding an active job or starting the same task twice can create double charges and confusing histories. Administrators should check the provider request before changing job state manually.

If you are an end user, wait for the provider-specific timeout, look in the completed-project area, and contact support with the job ID if the result remains inaccessible. Clear diagnostics beat repeated blind retries every time.

Ready to make your own AI video?

Choose a SupaPic scene, add a clear reference photo and follow the generation from one studio.

Create with SupaPic