Pixelle-Video
ATH-MaaS/Pixelle-VideoReviewed by AI
Checked by RepoPicks Radar on
What is Pixelle-Video?
Pixelle-Video makes vertical or horizontal short videos on its own. You type a topic in a page that opens in your browser; it writes the narration with an AI text model, creates one image or clip per sentence, reads the text out loud with a text-to-speech voice, adds background music and stitches everything together. You can also paste your own script, upload your own photos or a voice sample to clone. Windows users unzip a ready bundle and double-click start.bat; on Mac and Linux you install uv and ffmpeg and start it from the terminal.
How easy is it to install Pixelle-Video?
B · installer plus one simple step · Windows
- Windows B
- macOS C
- Linux C
Windows is the easiest and sets the grade: the release page has a 380 MB zip, and the README says it needs no Python, uv or ffmpeg — you unzip, double-click start.bat and the browser opens at http://localhost:8501, where the one remaining step is pasting your model keys into the settings panel. Mac and Linux have no bundle: you install uv and ffmpeg, clone the project and run "uv run streamlit run web/app.py" before the same key setup. Paying nothing for the AI part means also installing ComfyUI and Ollama yourself, which is a much longer job on any of the three.
Pixelle-Video on GitHub
- Stars
- 29,954on Nov 10, 2026
- Downloads
- 87,147891 this week
- Growth
- +193stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Nothing extra
- To build from source
- Python, Docker, FFmpeg
- Disk space
- 380 MB–1.1 GB · from the installer
- Accounts or keys
- An AI text model: Qwen/Tongyi, OpenAI, DeepSeek and similar, or Ollama on your own computer, Image and video models: DashScope/Wan, OpenAI GPT Image, Volcengine Seedream/Seedance, Kling (not needed if you run ComfyUI yourself), RunningHub, a cloud ComfyUI used instead of a local one
- Cost
- Free, API costs extra · The program is free (Apache-2.0), but it only works once you plug in model services: an AI text model plus image, video or voice generation. The README's own FAQ says a completely free route exists — "LLM 使用 Ollama(本地运行)+ ComfyUI 本地部署 = 0 元" (Ollama and ComfyUI on your own machine, which needs a graphics card) — and otherwise you pay per video to providers such as Qwen, OpenAI, Kling, Seedance or RunningHub.
- License
- Apache-2.0 (Permissive)
- Last change on GitHub
- Jun 14, 2026
Pixelle-Video risk signals (5)
- Infowindows bundle older than the code
The newest Windows zip is Pixelle-Video-v0.1.15-win64.zip from 2026-01-27, while pyproject.toml already says version 0.2.0 and the last commit here is 2026-06-14. So the ready bundle does not include the later changes the README lists, such as the 2026-06-01 direct API image and video model settings.
AI review
- Inforepository owner differs from readme
This copy sits under ATH-MaaS, but every link in README.md points somewhere else: the badges, the issues page, the documentation at aidc-ai.github.io and the clone command "git clone https://github.com/AIDC-AI/Pixelle-Video.git". GitHub does not mark it as a fork, so the account was probably renamed and the texts were not updated.
AI review
- Infoyour keys sit in a plain config file
The keys you type in the browser are saved to config.yaml, which docker-compose.yml mounts read-write into both containers ("./config.yaml:/app/config.yaml"). The README also offers "打印模型请求参数" (print model request parameters) for debugging, which writes the prompt, model name and input file paths to the terminal.
AI review
- Inforetries around provider content checks
README.md, API video generation: "支持网络下载重试与内容审核失败后的提示词中性化重试" — when a provider's content check rejects a scene, the program rewrites the prompt in a more neutral way and tries again. Nothing in the tool checks the finished video before you publish it.
AI review
- Infowhole video is made without a human in between
The feature list promises "输入主题,自动生成完整视频" (type a topic, get a complete video), with the script written by a language model and no fact-checking step. The videos still carry whatever each provider's terms say about AI content and about the music in bgm/.
AI review
Evidence is shown in its original language.
What Radar could not check about Pixelle-Video
- Whether the Windows zip really runs with no extra installs, and what is inside those 380 MB
- What a finished video costs with paid image or video models, since each scene is a separate call
- How long one video takes, and whether a computer without a graphics card can use the free Ollama plus ComfyUI route
- Whether the older Windows bundle still talks to the current provider APIs
- Real disk space in the output/ folder after a few videos
- Whether the voice cloning from an uploaded sample works outside Chinese