Pixelle-Video

Pixelle-Video is a Python-based automation engine that turns a single topic into a finished short video. It handles script writing, AI image or video generation for each scene, text-to-speech narration, background music, and final video assembly, all through a web interface. It's built for creators, marketers, and hobbyists who want to produce short-form video content without editing skills or a video production pipeline.

The project connects to multiple AI backends: large language models for scripts (Qwen, GPT, DeepSeek, Ollama), ComfyUI or RunningHub workflows for images/video/TTS, and direct API integrations with providers like DashScope, OpenAI, Seedream, Seedance, and Kling. Everything is configured through a Streamlit web UI rather than code, and the whole pipeline (script → storyboard → per-scene media → final composite) is exposed as a visible progress flow while a video renders.

Key features

• Full pipeline automation: enter a topic and get a complete video with script, imagery, narration, and BGM without manual steps.
• AI scriptwriting: an LLM writes the narration based on the topic, or you can supply your own fixed script.
• Per-scene AI media: generates matching illustrations or short video clips (e.g. WAN 2.1) for each line of narration.
• Direct API media models: calls DashScope, OpenAI, Seedream, Seedance, and Kling directly, without requiring a local ComfyUI setup.
• Multiple TTS engines: supports Edge-TTS, Index-TTS, and voice cloning from a reference audio sample.
• Background music: add built-in or custom BGM tracks to set the mood.
• Template system: static, image-based, and video-based HTML templates for different visual styles, in portrait, landscape, or square formats.
• Modular backend swapping: mix and match LLMs, image/video generators, and TTS engines, either via ComfyUI/RunningHub workflows or direct API calls.
• Custom material workflows: upload your own photos/videos and let the AI analyze them to generate a matching script.
• Digital human and motion transfer modules: extended pipelines for talking-head narration, image-to-video, and motion transfer from a reference video.

Ideal use cases

Pixelle-Video fits people who need to produce narrated short videos (explainers, history recaps, novel commentary, personal growth content, knowledge clips) at volume, without touching a video editor. It works well if you're comfortable running a local Python environment or a Windows one-click package, and if you already have (or are willing to get) API keys for an LLM and an image/video/TTS backend.

It's a good fit for content creators experimenting with AI-generated short-form video for platforms like Douyin/TikTok-style feeds, for developers who want to customize ComfyUI workflows or HTML templates, and for anyone wanting a free local pipeline (Ollama + local ComfyUI) versus a paid cloud setup.

It's not a fit if you need frame-accurate manual editing control, complex multi-scene continuity beyond what the storyboard-per-line model supports, or a hosted SaaS product with no setup. It's also not built for long-form video (documentaries, tutorials over several minutes) since the pipeline is designed around short, narration-driven clips.

Installation

The fastest path on Windows is the prebuilt package:

1. Download the latest Windows one-click package from the Releases page.
2. Unzip it and run start.bat.
3. The web UI opens automatically at http://localhost:8501.
4. Configure your LLM API and image generation service in the settings panel.

For macOS/Linux or custom setups, install from source. First install uv (Python package manager) and ffmpeg:

# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt update
sudo apt install ffmpeg

Then clone and run the project:

git clone https://github.com/AIDC-AI/Pixelle-Video.git
cd Pixelle-Video
uv run streamlit run web/app.py

The web UI opens at http://localhost:8501. On first run, expand the system settings panel and fill in:

• LLM config: choose a model (Qwen, GPT, etc.) and enter an API key.
• ComfyUI/RunningHub config: point to a local ComfyUI instance or provide a RunningHub API key for cloud workflows.
• API media model config: enter provider keys (DashScope, OpenAI, ARK, Kling) if using direct API image/video generation instead of ComfyUI.

Save the configuration and start generating videos.