Troubleshooting clipping
Most clipping issues fall into one of four buckets: upload, transcription, highlight selection, or render. Use this guide to narrow down what went wrong.
Upload fails or stalls
- Check the file type. We accept MP4, MOV, WebM, MKV, MP3, WAV, and M4A. Other formats are rejected before upload starts.
- Refresh and retry. Browser uploads can stall on flaky connections. Reloading the page resumes from scratch — the partial upload is discarded.
- Try a smaller file. If a 2GB+ file consistently fails, re-encode to a lower bitrate (H.264, ~5 Mbps for 1080p is plenty for transcription).
Transcription is missing words or sentences
The highlight model only sees what Deepgram transcribed.
- Background music or noise — clean speech transcribes ~98% accurately; noisy audio drops sharply. Run the source through a noise-reduction tool before uploading.
- Strong accents or technical jargon — both reduce accuracy. Edit the transcript on the affected clip and re-render. See Editing captions and speakers.
- Long silences — Deepgram skips silent regions. If a "missing" sentence was actually a long pause, that's expected.
The highlights aren't good
The highlight selector works best when your source has clear, self-contained moments — a punchline, a strong opinion, a story beat. If your content is one continuous explanation with no obvious peaks, you'll get weaker selections.
Things to try:
- Raise the clip count. More clips means the model picks deeper into the ranking, but it also gives you more material to choose from.
- Pre-split the source. If you have a 3-hour episode, splitting it into 30-minute segments per topic usually produces better clips than running the whole thing at once.
- Edit the highlight. Delete weak clips, keep the ones worth posting.
A clip failed to render
Open the clip — the error message indicates the stage that failed.
- Lambda timeout — usually fixes itself on retry. Click Re-render.
- Source not available — the upload may have been deleted. You'll need to re-upload the source and start a new project.
- Face detection failed — face-detect is optional; the clip falls back to center-crop automatically. No action needed.
If a clip stays in "failed" after a second retry, contact support with the project ID — we can inspect the underlying Trigger.dev run.
Credits were deducted but the project never finished
Cost is estimated up front and refunded when the project finalizes. If a project stalls (no clips after 30+ minutes), the credits are held but not lost. Contact support — we'll either complete the run or refund the hold.