letmehappycode /auto-video-agent

About auto-video-agent

Give it an SRT subtitle file and Claude Code splits it into 2–6 second shots, writes an illustration prompt for each, draws them with Seedream and stitches the frames into a video. A local web page lets you browse and redo shots. It's built for narrated explainer clips, so the result is still pictures on a timeline, not moving AI video. There is no LICENSE file, so reuse rights are unclear.

Best for: Creators of narrated story or explainer videos who want each line illustrated without drawing by hand.

• Splits an SRT file into shots of 2–6 seconds by meaning, with start and end times
• Writes a structured illustration prompt per shot and checks it against ten rules
• Draws every shot through Volcano Engine's Seedream text-to-image API
• Stitches the images into a silent video that follows the subtitle timing
• Includes a local web UI to browse shots, preview images and edit shot data
• Handles long scripts by cutting them into chunks first, then merging the shots

1. 1
   Clone the repo and copy .env.example to .env, then add your Volcano Engine Seedream key.
2. 2
   Open the folder in Claude Code.
3. 3
   Give Claude the path to your SRT file and the art style you want.
4. 4
   Claude runs the agents and workflows, then scripts/generate_images.py and scripts/compose_video.py.
5. 5
   Open the web UI in web/ to review shots and redo any image.