Highlight detection and face tracking run on the full source file, not just the segment between your start and end parameters. The API applies the start and end window only after that analysis finishes, when it builds the output clip.
Because of this, a long 1080p source can fill up a worker's disk even when the file already passes the API's download and transcription limits. Those limits check file size and duration for download and transcription, not the disk space needed to analyze the full video at 1080p.
Pre-trim the source video before you submit it, so the file only contains the section you care about.
If you must submit a full-length 1080p file, keep it under about 30 minutes.
For longer source material, submit a lower resolution or split the video and send each part as a separate job.
The job status changes to failed.
Any credits charged for the job are refunded automatically. You don't need to contact support to get them back.
Video not found is a separate error from a disk space failure. It means the source URL couldn't be fetched, usually because the hosting server rejected the request or the file was removed before the worker downloaded it. Check that the source URL is public and still live, then retry the job.
