ATH-MaaS /Pixelle-Video

About Pixelle-Video

Type a topic and it writes the script, draws the images, adds a voice. You work in a Streamlit web page: pick your models and hit generate. Cloud image and video models need paid API keys, but local Ollama and ComfyUI cost nothing. Most docs are in Chinese.

Best for: First-time editors who want a topic turned into a finished short video.

• Writes the narration from a topic, or uses a script you paste in
• Makes one image or video clip per sentence, then stitches the video
• Runs image, video and TTS steps on local ComfyUI, RunningHub, or APIs
• Clones a voice from a short reference clip with Index-TTS
• Picks HTML templates for vertical, horizontal or square video
• Adds extra pipelines: digital-human presenter, image-to-video, motion transfer

1. 1
   On Windows, download the all-in-one package, unzip it, run start.bat, and open http://localhost:8501
2. 2
   On macOS or Linux, install uv and ffmpeg (brew install ffmpeg, or sudo apt install ffmpeg)
3. 3
   Clone the project: the command below then cd Pixelle-Video

   git clone https://github.com/AIDC-AI/Pixelle-Video.git

   Copy
4. 4
   Start the web page: the command below

   uv run streamlit run web/app.py

   Copy
5. 5
   In ⚙️ System Configuration, add your LLM API key, and optionally a ComfyUI URL or RunningHub key
6. 6
   Enter a topic or a fixed script, choose TTS, image workflow, template and size, then click 🎬 Generate Video

• Pixelle-Video vs MoneyPrinterTurbo