In 2025, as the wave of AI‑generated content (AIGC) swept the short‑video field, the open‑source project Pixelle‑Video from AIDC‑AI stood out. As an AI fully automated short‑video engine, it shifts video creation from complex editing to a "one‑click generation" mode. By entering a single topic, the system can automatically complete script writing, AI image generation, speech synthesis, background music addition and video composition, making the entire process zero‑threshold and requiring no professional skills. This not only lowers content production costs but also democratizes short‑video production, promoting broad use from individual creators to enterprise marketing.
Pixelle‑Video was open‑sourced on GitHub in November 2025 under the Apache 2.0 license and supports commercial use. The project draws inspiration from Stable Diffusion, ComfyUI and TTS toolchains and integrates large language models (LLMs) such as GPT or Tongyi Qianwen to achieve end‑to‑end automation. Compared with closed‑source tools like HeyGen or Synthesia, Pixelle‑Video emphasizes open‑source flexibility and workflow customizability, making it suitable for education, marketing and social media. The project goal is to "make video creation a one‑sentence task," aiming to secure a position in the AI short‑video market (projected at $50 billion in 2025). Its web preview interface is concise, supports Chinese and English, and marks the rise of Chinese open‑source AI in multimodal generation.
Highlights: End‑to‑End Automation from Text to Video
The core capability of Pixelle‑Video is fully automatic short‑video generation that covers every stage of content creation. Users input a topic (for example, "Why cultivate a reading habit"), and the system launches a pipeline: first, an LLM generates the script (supporting GPT, Tongyi Qianwen or Ollama); next, the image plan is created and ComfyUI is called to generate AI illustrations (one image per sentence, supporting style prefixes like "black‑and‑white line drawing"); then the voiceover is synthesized (compatible with Edge‑TTS, Index‑TTS, etc., and supports voice cloning); finally, BGM is added and the video is composed with one click.
Specific features include:
- AI Script Writing: Automatically creates narration based on the topic, with the option to supply a fixed script to skip generation. Script length is controllable and suitable for 15–60 second short videos.
- AI Image Generation: Produces high‑quality illustrations for each shot, supporting local ComfyUI or RunningHub cloud workflows. Flexible dimensions (1024x1024 default, supports vertical/landscape).
- Speech Synthesis: Integrates multiple TTS solutions and allows uploading reference audio for voice cloning. A preview function enables iterative adjustment.
- Background Music: Built‑in or custom BGM, with a no‑music option. Preview buttons facilitate selection.
- Video Templates: Multiple layouts available (e.g., vertical short‑video styles), with support for HTML customization.
- Composable Atomic Capabilities: Based on the ComfyUI architecture, users can swap base image models (e.g., from SDXL to FLUX) or TTS providers (e.g., ChatTTS) to achieve personalized extensions.
The video generation pipeline is clear: script generation → image planning → frame‑by‑frame processing → video composition. Example videos demonstrate realistic narration for historical commentary (such as "Lessons from Zizhi Tongjian"), showing potential for educational content. Community feedback indicates generation times of a few minutes, suitable for batch production.
Technical Characteristics: Modular Design and Open‑Source Flexibility
Pixelle‑Video's technical strengths lie in modularity and openness, using the Streamlit web framework to ensure user‑friendliness. The core is the ComfyUI workflow system, and users can customize JSON workflow files in the workflows folder to perform atomic‑level replacements (for example, switching image‑generation models from SDXL to FLUX). LLM configuration supports multiple models (Ollama local free option), image generation is compatible with local or cloud services, and the voice workflow auto‑detects and loads available options.
Notable points include:
- Balance of Automation and Customization: Preset templates (e.g., default.mp3 BGM) simplify operation while Prompt Prefix controls allow stylistic variations (such as "Minimalist black‑and‑white"), producing diverse outputs.
- Cost Optimization: A fully free setup (Ollama + local ComfyUI) has zero provider cost; cloud configurations (e.g., Tongyi Qianwen + RunningHub) offer cost‑effective options well below services like ElevenLabs.
- Multilingual Support: Seamless Chinese and English support with potential extension to Japanese and Korean, suitable for global users.
- Safety and Ethics: Built‑in watermarking to prevent misuse; TTS supports cloning but requires users to upload lawful reference audio.
- Extensibility: Following examples like Midjourney and Stable Diffusion, future integration of multimodal inputs (such as video‑to‑video generation) is possible. GitHub star history shows rapid growth, indicating community recognition.
Compared with similar projects (like HeyGen), Pixelle‑Video's open‑source nature makes it easy to fork and improve; a drawback is that newcomers need to configure API keys.
Installation and Usage Guide: Practical Steps from Zero to One
Pixelle‑Video is easy to install and supports a Windows one‑click package and source‑code installation.
Windows One‑Click Package: Download and extract, then double‑click start.bat to launch the web interface (http://localhost:8501). No need to install Python/uv/ffmpeg—works out of the box. Configure LLM (API Key/Base URL/Model) and image settings (ComfyUI address/RunningHub Key), save and start generating.
Source Installation (macOS/Linux): Install uv and ffmpeg first (via brew/apt). git clone the repository, cd into it, and run uv run streamlit run web/app.py. Configuration is the same as above.
Usage workflow:
- System Configuration: Expand the panel, select preset models (e.g., Tongyi Qianwen) and fill in keys. Choose local or cloud image services.
- Content Input: Choose AI‑generated or fixed script and add BGM (built‑in or from a custom folder).
- Voice Settings: Select a TTS workflow and upload reference audio for preview.
- Visual Settings: Choose image workflow, dimensions and Prompt Prefix for preview.
- Generate Video: Click the button and monitor real‑time progress. Outputs are saved in the output folder and support preview.
FAQ: If results are unsatisfactory, swap models; recommended cost‑free setup is Ollama + local ComfyUI. Report issues via GitHub Issues.
Conclusion: The Future Potential of Pixelle‑Video
As an open‑source AI short‑video engine, Pixelle‑Video lowers the barrier to creation and promotes AIGC adoption. The author acknowledges inspiration and support from several open‑source projects such as MoneyPrinterTurbo ‑ an excellent video generation tool. Future extensions may include more multimodal inputs to support the content economy. In the wave of a potential $5 trillion AI market, it represents Chinese open‑source capability and deserves attention from developers.