Guides
Create a job
Choose a source and set clip, transcription and encoding options.
A job takes one video and produces a set of clips. Create one from New job.
Sources
| Source | What happens |
|---|---|
| Upload | The file is copied into ~/.clipforge/uploads. Convenient, but it duplicates the file. |
| Local path | The file is used in place, and nothing is copied. A leading ~/ is expanded. |
| URL | Downloaded with yt-dlp at up to 1080p, with a live progress bar. See URL downloads. |
Large recordings
Browsers upload every byte, so for multi-gigabyte recordings enter the path instead.
Clips
| Field | Default | Meaning |
|---|---|---|
| Count | 8 | Maximum number of clips |
| Min length | 15 s | Shortest allowed clip |
| Target | 30 s | Preferred clip length |
| Max length | 60 s | Longest allowed clip |
| Output folder | ~/Movies/clipforge | Clips go in a subfolder named after the video |
Lengths must satisfy min, then target, then max. The form rejects anything else.
Transcription
Speech recognition is on by default using Whisper. Models:
large-v3-turbo(default): best balance of speed and accuracylarge-v3: slower, slightly more accuratesmall: fast, lower accuracy
Leave Language empty to auto-detect, or enter a code such as en.
Turn speech recognition off for silent gameplay. Captions are then unavailable, and moments are chosen from audio and visual signals only.
Encoding
| Option | Default | Notes |
|---|---|---|
| Codec | H.264 | HEVC gives smaller files |
| Quality | High | High, medium and low map to 12, 8 and 5 Mbit/s on the hardware encoder |
| Parallel renders | 3 | Clips rendered at the same time |
| Software encoding | Off | Forces libx264 or libx265 instead of the media engine |
| Fresh analysis | Off | Ignore cached transcript, signals and picks |
Overlay and layout options have their own guides: Layouts and facecam and Captions and hooks.
Edit on GitHubSpotted a mistake? Suggest a change and open a pull request.