Narration on your CPU
Kokoro-82M reads each scene. Eight voices, American and British. Nothing leaves your machine and there is no per-word billing.
Local video toolkit
Write the script.
Write the script.
Get the video.
Narrates your text, builds the visuals, burns in captions and returns a finished 1080p MP4. Runs on your own machine. No subscription, no queue, no API keys.
Buy the toolkit
€29one-time payment
One file, no account, no subscription. It runs offline, so there is nothing to cancel later.
$ ./toolkit build examples/deep-ocean.json Script: Why the Deep Ocean Is a Mystery scenes: 3 · 1920×1080 · af_heart background 1/3 ... ok background 2/3 ... ok background 3/3 ... ok scene 01 7.1s scene 02 7.8s scene 03 9.9s Joining scenes (pairwise) ... round 1: 2 clips round 2: 1 clip Normalising loudness (EBU R128) ... Done deep-ocean.mp4 23.8s · 1.73 MB · rendered in 56s
What it makes
Three scenes, one voice, captions burned in. This is the entire output — nothing was edited afterwards and no other software was used.
23.8 s · 1080p
What you get
Four things that usually take a weekend of glue code.
Visuals without a licence audit
Backgrounds are drawn procedurally and given a slow push. No stock library, no attribution file, nothing to check before you publish.
Captions burned in
Set caption on a scene. It lands on screen in a soft box, wrapped to stay readable on a phone.
Constant memory use
Scenes are joined pairwise, two at a time. A 60-scene script peaks at about the same RAM as a 3-scene one.
Requirements
If FFmpeg runs on your machine, this will too.
| Python | 3.10 to 3.13 |
|---|---|
| FFmpeg | 5.0 or newer, on your PATH |
| Disk | 2 GB for the voice model and working files |
| CPU | Any 64-bit processor. No GPU. |
| Internet | Only for the one-time voice download |
| Platforms | Linux, macOS, Windows |
What it does not do
Each of these adds either a licence risk or an account dependency, so they are left out on purpose.
- No music library. Add your own track in any editor; licensed music is the fastest way to get a claim.
- No stock footage. Backgrounds are generated, so nothing needs a per-clip licence check.
- No auto-upload. The toolkit stops at a finished file. Publishing stays your decision.
- No subtitle sidecar. Captions are burned in; a .srt generator is not included.
Licence
The toolkit is MIT licensed. Your videos are entirely yours — sell them, monetise them, no attribution required. Every third-party dependency is listed in the package with its exact terms, including what changes if you repackage the tool itself.
€29, once
Runs offline after the first download. Support is a reply to your receipt email.