Story videos by committee: one theme, many models

What it does You give it a theme—say, a rabbit and a fox—and it splits the idea into segments, drafts a script, generates one image per segment via your chosen providers, and assembles a narrated short video with subtitles. It is fundamentally an orchestration layer: FastAPI backend, React frontend, and a lot of patience while remote models return images and copy.

The interesting bit The project treats video generation as a supply-chain optimization problem. It supports five text providers and three image providers, letting you mix and match based on quota, latency, or whichever service is having a good day.

Key highlights

• Mix-and-match backends: text from OpenAI, DeepSeek, Ollama, Aliyun, or SiliconFlow; images from OpenAI, Aliyun, or SiliconFlow.
• One image per story segment; segment count directly dials video length and generation time.
• Produces synthesized narration and burned-in subtitles automatically.

Caveats

• DeepSeek and Ollama integrations are text-only; image generation still requires a cloud provider such as Aliyun or OpenAI.
• Ollama text generation demands a hefty local model—the README suggests 14B parameters or larger, and smaller weights are explicitly unsupported.
• SiliconFlow support is narrow: text models must be OpenAI-compatible, and the image pipeline has only been validated against black-forest-labs/FLUX.1-dev.

Verdict Worth a look if you want a prototype for AI story videos without writing your own prompt chains. Skip it if you need motion graphics or anything beyond static image slideshows with narration.