Pixelle Video is an open-source engine that turns one topic into a narrated short, with an AI picture for every sentence. You install it yourself. Here are its demos, the download and setup steps, and an online alternative with nothing to install.
The online studio runs on the MoneyPrinterTurbo engine with real stock footage, not on Pixelle Video code. Not affiliated with the Pixelle-Video authors.
Pixelle Video (written Pixelle-Video on GitHub) is an open-source “AI fully automated short video engine”. You enter a topic and Pixelle Video writes the script, generates an AI image or AI clip for every sentence, adds a voiceover and background music, and assembles the video. It runs on your own computer.
The Pixelle Video GitHub repo belongs to ATH-MaaS, the Model-as-a-Service team inside Alibaba Token Hub. It first appeared under AIDC-AI, and that old address still redirects.
These four shorts were made with Pixelle Video and published by its authors. Each sentence gets its own AI-generated picture, the look Pixelle Video is known for. The narration is in Chinese.
Videos and diagram from the Pixelle-Video README, Apache-2.0. They show Pixelle Video output, not videos made on this site.
To download Pixelle Video on Windows, get the all-in-one package from the GitHub releases page; it needs no Python or ffmpeg. On macOS and Linux you install from source with uv and ffmpeg. Either way you then connect a language model and something to draw the pictures.
Open the latest release and download Pixelle-Video-v0.1.15-win64.zip, about 380 MB, then extract it and double-click start.bat. Your browser opens http://localhost:8501.
Needs Python 3.11 or later, the uv package manager and ffmpeg.
git clone https://github.com/ATH-MaaS/Pixelle-Video.git
cd Pixelle-Video
uv run streamlit run web/app.pyAdd an LLM key (Qwen, GPT or DeepSeek, or a local Ollama model). Then set up the visuals: a local ComfyUI, a RunningHub API key, or an image API such as OpenAI, DashScope or Kling.
The docs say a 5-scene video takes about 2 to 5 minutes, depending on the services you connected.
Pixelle Video is in use, with about 150 open issues, but its last code push was June 14, 2026. These come up most.
The download is v0.1.15 from January 2026, while the Pixelle Video source is at 0.2.0. A request for a new package is still open.
Several open issues report Pixelle Video sending work to a cloud provider even with a local ComfyUI configured.
The default Edge-TTS can fail on an unstable network. The FAQ suggests switching to a ComfyUI TTS workflow.
Most issue threads are in Chinese, and the README’s documentation link returned a 404 when we checked.
If you want the result and not the setup, the online studio does the same job from one line of text. One real difference: it uses stock footage where Pixelle Video draws AI images. Real users made these eight. Tap one to play it with sound.
Videos by MoneyPrinterTurbo Online users. Footage from Pexels.
Open the studio, type a topic, pick a voice, a size and a caption style, then generate and download the MP4. It takes a few minutes, with no ComfyUI and no API keys. The full walkthrough is in how to use MoneyPrinterTurbo.
Sign in with Google or GitHub and describe the video in one sentence, the way you would start in Pixelle Video. Or paste your own script.
Choose from 17 languages, a voice, 9:16, 16:9 or 1:1, a caption style and music. Pixelle Video asks for an image prompt here; the studio matches stock footage instead.
The button shows the credit cost first. The video renders on our servers, then you download the MP4.
Not included online: AI-drawn images, AI video clips (Wan, Seedance, Kling), the digital human, image-to-video and motion transfer pipelines, voice cloning and custom templates. If those are why you searched for Pixelle Video, install Pixelle Video itself.
Pixelle Video is free software under Apache-2.0, but the models behind it are not: cloud APIs charge per call, and a local ComfyUI needs a capable graphics card. Online, a new account gets 60 credits, about one one-minute video, and paid plans start at $9.90 a month.
Every plan includes
Purchase additional credits to use our services
No. As of November 11, 2026, Pixelle Video ships as a Windows package and as source code on GitHub. Its README and docs mention no hosted website or web demo. The studio on this page is independent and is not run by the Pixelle-Video authors.
The Pixelle Video GitHub repo is github.com/ATH-MaaS/Pixelle-Video, with about 30,000 stars. The older address under AIDC-AI redirects there. The latest Pixelle Video release, v0.1.15 from January 2026, is a Windows all-in-one download of about 380 MB.
Not strictly. Pixelle Video can generate its visuals three ways: a local ComfyUI, RunningHub cloud workflows, or direct API calls to image and video models. You need at least one of them, plus an LLM key, before Pixelle Video can make its first video.
Yes, if the pictures are generated in the cloud through RunningHub or an image API. A GPU only matters for a local ComfyUI, where the docs recommend an NVIDIA card with 6 GB of VRAM or more. Text-only templates need no image generation at all.
For no install at all, the online studio on this page. For other open-source projects, MoneyPrinterTurbo and MoneyPrinterV2 do the same topic-to-video job with stock footage or AI images. The open-source AI video generator comparison on this site lists 15 options.
Type one topic, pick a voice and download a finished short. The first video is on us.