A local toolkit that narrates your text, builds the visuals, burns in captions and delivers a 1080p MP4. No subscription, no cloud, no rendering queue. It runs on your own machine.
Checkout is currently in test mode - no real payment is taken.
$ ./toolkit build examples/deep-ocean.json
Script: Why the Deep Ocean Is Still a Mystery (3 scenes, 1920x1080)
Voice: af_heart
backgrounds: 3 generated (1920x1080)
scene 01: 7.1s
scene 02: 7.8s
scene 03: 9.9s
Joining scenes ...
join round 1: 2 clips
join round 2: 1 clips
Normalizing loudness ...
Done: deep-ocean.mp4
23.8s, 1.73 MB, rendered in 0m 56sReal output from a 6-core desktop, no GPU.
Kokoro-82M runs on your CPU. Eight voices, American and British. Nothing is sent to a server, and no API key is involved.
Backgrounds are drawn procedurally with a slow zoom. No stock library, so there is no licence to check before you publish.
Two-pass EBU R128 normalisation lands every video at -16 LUFS. YouTube stops turning your audio down.
Scenes render separately and join pairwise. A 60-scene script uses no more RAM than a 3-scene one.
| Component | Requirement |
|---|---|
| Python | 3.10 - 3.13 |
| FFmpeg | 5.0 or newer |
| Disk | 2 GB free |
| CPU | Any 64-bit. No GPU. |
| Internet | Only for the one-time voice download |
Windows, macOS and Linux. The installer creates a private virtual environment inside the folder and installs nothing system-wide.
No music library. No stock footage. No auto-upload. No subtitle files. Each of those drags in either a licence risk or an account dependency, and this tool does one thing well: script in, finished video out.
The toolkit is MIT. Your videos are entirely yours - sell them, monetise them, no attribution. Every third-party dependency is documented in the package, with the exact terms and what changes if you redistribute.