Skip to main content
Back

AIDC-AI's Pixelle-Video Turns a Single Topic Into a Finished Video

An open-source engine from AIDC-AI turns a single topic prompt into a complete short video with script, visuals, voiceover, and music.

Repo, Full Access
  • Pixelle-Video is an open-source engine from AIDC-AI that generates short videos from a single topic input
  • Pipeline chains an LLM, image/video model, TTS, and compositor with fully swappable components
  • Supports ComfyUI, RunningHub, and direct API calls to DashScope, OpenAI, Kling, and Seedance
  • Runs completely free using Ollama plus local ComfyUI, or cheaply via Qwen APIs
  • Windows one-click package, or uv run streamlit run web/app.py on macOS and Linux
  • Includes digital human, image-to-video, motion transfer, and voice cloning modules

Typing a single topic and getting back a finished short video sounds like a marketing pitch, but Pixelle-Video from AIDC-AI actually ships that pipeline as an open-source project. It has quietly picked up over 25,000 GitHub stars, and the appeal is straightforward: feed it a subject line, and it chains together an LLM, an image or video model, a text-to-speech engine, and a compositor to spit out an MP4.

Pixelle-Video web UI showing configuration panel and video preview

Four stages, four swappable models

The engine breaks generation into four steps that each map to a replaceable model. Input text flows through script generation, image planning, frame-by-frame processing, and finally composition. Every stage exposes model choice, so you can mix a local Ollama LLM with a cloud image API without touching code.

  • Script: supports GPT, Qwen, DeepSeek, Ollama and other LLMs for writing narration from your topic.
  • Visuals: generates matching stills per shot, or dynamic clips using video models like WAN 2.1.
  • Voice: