AI & MediaPre-launch
videosynx
Producing a coding tutorial by hand means writing the code, recording the screen, narrating over it and cutting it together. videosynx does the whole chain from a topic: the code is generated, executed for real, and drawn frame by frame into a finished video with narration and a subtitle file.
What it does
- A topic or a pasted script goes in; an LLM chain writes runnable code and the narration that explains it, one sentence anchored to the lines it describes.
- Every snippet is executed in a sandbox with no network and no readable home directory, and its real output is what appears on screen.
- Frames are drawn programmatically and piped to ffmpeg against an integer-frame timing plan, so audio and video cannot drift apart.
- Explainer slides carry diagram specs — grids, cycles, trees, stacks — rendered in code, so nothing is ever screen-captured.
- Narration runs on a local text-to-speech model with dozens of voices, with captions burned in and exported as a matching subtitle file.
Worth calling out
It runs the code, it does not pretend to
The output on screen is the program's actual stdout from a sandboxed run. If isolation cannot be established the worker refuses to start rather than quietly falling back to something less safe.
Deterministic rendering
Because frames are drawn rather than captured, the same lesson renders the same bytes every time — and it renders faster than real time, because nothing has to play out in a recorder.
Have something that needs building?