The BEST Low VRAM Image Model Has Been Under Our Nose THE WHOLE TIME

If you're really into the AI space, you've probably heard the crazy recent discovery in regard to Wan 2.1, but if you haven't - allow us to break it down!

Wan 2.1, the break through open-source video model released nearly 6 months ago was recently leveraged to generate just 1 frame (a single image) and the results are insane, a break through for low Vram users and our understanding of video models in general.

Using quantized versions of Wan 2.1 along with the attached workflow, we can get some absolutely incredible HD resolution images at Flux Dev level quality without the hardware needs. And when outputs from Wan are paired with our skin refiner and upscaling workflows, you can get some images that are absolutely indistinguishable from reality:

Original

Refined & Upscaled

This workflow generally produces an image in ~2 min on 8GB Vram.

In our video, we also teased a first look at our new platform where we will be hosting our workflows and documentation as well as a first look at our newest skin refiner workflow - RealSkin AI V2.

Instead of using the base Wan 2.1 model, you can also leverage the Wan Fusion X model linked below along with the Lightx2v Lora to get more dynamic characters and results. There are a ton of settings to play around with in such a simple workflow, but generally the results are all great.

You can find all the models and workflow for Wan 2.1 image generation below Sir. Frogus as an image!

As always, feel free to leave any questions or comments below!