Image: GitHub
Why it mattersA rendered film can be changed and re-run. A film made by a model has to be re-prompted. The next take is different.
A new agent skill called cinematic-skill has passed 295 stars on GitHub. The project was created on 9 November 2026. The traction is real inside a 7-day window. The author, exploraX, has written it as a Claude Code skill. The README also lists Codex, Cursor, Gemini CLI, OpenCode and other agents through the npx skills CLI. The license is MIT.
What the skill makes
The skill is for short documentaries and explainer films. The author writes the design rule on the README. The project says "Nothing is generated video. Every frame is drawn by code (Skia, in Python)". The film renders the same every time. Changes are cheap. A new voice re-times the whole film by itself. The dependencies are listed in order. Skia for the drawing. Numpy and pillow for the maths and image work. Scipy as an option. FFmpeg on the user's PATH for the final assembly.
What the user does
The user hands the skill a subject. The agents do the research and write the script. The user approves a beat table, picks a visual look, and approves a 30-second draft. The agent then renders the full film from code. The example film in the README runs 92 seconds. It takes about six minutes to render on a 16 GB laptop, from one npx skills call. The voice is a two-stage trick. A free scratch voice times the film first, using macOS say or espeak-ng. The final voice can be any TTS. A human voice works too, recorded one section at a time. The tool splits the recording into timed lines.
Why "no video model" is the news
Model-generated video has one big problem for a business. The output is not deterministic. Two runs with the same prompt give two different films. A small edit is not a small edit. A client who asks to change one line of narration has to agree to a new film. A rendered film owned by code is the other thing. The script and the asset list are the file. Change a word in the script. Run the skill again. The only thing that moves is the two seconds of narration around that word. Every other frame renders the same way. For a team building a product on top of this, the cost of a change is the whole point. The use cases: short explainers, case studies, release films, data stories.
A broader audience will want this before an engineering audience does. A marketing team or a product manager making a launch film cares most about the cost of an edit. They care less about which Python library drew the frame.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.