Seedance Studio
Seedance Studio is my local tool for AI video and voice: a command line and a web page over the Seedance video model on kie.ai, Gemini text-to-speech and Munsit's Arabic voices. Every job can run as a dry run first, shows its cost, and refuses anything over a credit cap.
- Python
- JavaScript
- kie.ai API
- Gemini TTS
- Munsit
- ffmpeg
- Whisper
- My part
- All of it, my own
- When
- Sep to Oct 2026
- Status
- Runs on my Mac

Overview
It started as a way to let Claude make videos for me without surprises on the bill. Every command has a JSON mode for scripts and a distinct exit code for over budget, and every video and voice is kept in a local history with its estimated and charged credits. The Python side uses only the standard library, and the whole interface is one HTML page.
My role
My own project. I wrote the command line, the local server, the one-page interface, a mock of both APIs for testing, and a Claude skill that records voice-overs for videos.
What shipped
- Video from text alone, from a first and last frame, or from up to nine reference images, three videos and three audio clips.
- 4 to 15 seconds, 480p or 720p and seven aspect ratios, picked from chips that show the credit rate.
- Gemini voices, alone or as a two-speaker dialogue, with style, pace and accent.
- Native Arabic voices from Munsit in Najdi, Hijazi, Emirati and Fusha, switched off until its key is set.
- A history with search, Arabic included, playback in place, "use again", and estimated against charged credits.
- A mock of both APIs and a smoke test that strip the real keys, so testing costs nothing.
- A voice-over skill: one recorded take, split on silences, each part checked with Whisper, then mixed over the video.
DecisionNothing is generated without its price stated and an explicit yes. The approved amount becomes that job's credit cap.
Worth a closer look
The cost guard is the point of the tool. A dry run uploads nothing and spends nothing, the estimate is shown before every job, and a job over its cap is refused before it reaches the API. For Munsit the estimate is deliberately pessimistic, because its pricing page does not list the voice I use and it offers no balance check.
LimitThe Munsit estimate is an assumption until a real charge confirms it.
Screens

