Pixelle-Video — короткие видео из темы

★ 28.5K

Pixelle-Video is an open-source engine that turns a single topic into a short video automatically: it writes the script, produces AI illustrations or video clips, adds voiceover and background music, and assembles the clip. Reach for it when you want a steady flow of vertical or horizontal videos without editing skills, when you want to turn a finished text into a video and skip the writing step (the fixed-script mode), or when you need a digital-human presenter, an image-to-video clip or motion transfer from a reference video. Inside: a Streamlit web UI; text through GPT, Qwen, DeepSeek or Ollama; images and video through ComfyUI, RunningHub or direct APIs for DashScope, OpenAI, Seedream, Seedance and Kling; voiceover with Edge-TTS or Index-TTS including voice cloning from a reference recording. Windows users get a ready package that needs no Python, uv or ffmpeg install (`start.bat`, UI at localhost:8501); on macOS and Linux you run from source with `uv run streamlit run web/app.py` and need uv and ffmpeg. API keys are needed only for the cloud services you choose: Ollama plus a local ComfyUI is described as a free setup. The focus is creating video from scratch, not editing or transcribing existing footage.