Open video generation that fits on a consumer GPU
Wan2.1 is an open suite of video foundation models built to match closed-source quality without requiring a data-center GPU.

What it does
Wan2.1 is a family of open video foundation models ranging from 1.3B to 14B parameters. It generates video from text, images, or book-end frames, and also handles video editing, text-to-image, and video-to-audio. The project publishes inference code, model weights, and a custom video VAE that processes 1080p video of arbitrary length while preserving temporal information.
The interesting bit
The 1.3B text-to-video model is the concession to reality: it needs only 8.19 GB of VRAM and can produce a 5-second 480p clip on an RTX 4090 in roughly four minutes without quantization tricks. It also claims to be the first video model that can render both Chinese and English text inside generated scenes.
Key highlights
- Consumer-grade option: the 1.3B T2V model runs in ~8.19 GB VRAM and targets 480p generation (720p is possible but the README warns it is less stable).
- Multimodal inputs: supports text-to-video, image-to-video, first-last-frame-to-video, video editing, plus text-to-image and video-to-audio.
- Text in video: explicitly claims to be the first video model capable of generating readable Chinese and English text overlays.
- Wan-VAE: a dedicated video autoencoder that handles 1080p at any length while preserving temporal information.
- Ecosystem integrations: already plugged into ComfyUI and Diffusers, with a growing list of community forks for motion control, virtual try-on, and long-form generation.
Caveats
- The README asserts that Wan2.1 outperforms commercial solutions on multiple benchmarks, but it does not provide the actual benchmark figures or comparisons to back that up.
- Several features on the todo list remain unchecked: Diffusers multi-GPU inference is missing across the board, and both ComfyUI and Diffusers integrations are absent for the first-last-frame and VACE variants.
- The first-last-frame model was trained mainly on Chinese text-video pairs, so the authors recommend Chinese prompts for better results.
Verdict
Worth exploring if you want open weights for video generation and only have a mid-range GPU. Skip it if you need a pipeline where every integration is already wired.
Frequently asked
- What is Wan-Video/Wan2.1?
- Wan2.1 is an open suite of video foundation models built to match closed-source quality without requiring a data-center GPU.
- Is Wan2.1 open source?
- Yes — Wan-Video/Wan2.1 is open source, released under the Apache-2.0 license.
- What language is Wan2.1 written in?
- Wan-Video/Wan2.1 is primarily written in Python.
- How popular is Wan2.1?
- Wan-Video/Wan2.1 has 16.8k stars on GitHub and is currently holding steady.
- Where can I find Wan2.1?
- Wan-Video/Wan2.1 is on GitHub at https://github.com/Wan-Video/Wan2.1.