New SOTA Virtual Try-On Model by FASHN live on fal

Team Features & Labels

• 2 min read
New SOTA Virtual Try-On Model by FASHN live on fal

fal is partnering with fashn.ai to offer the first commercially available virtual try-on model.

To play around with the new model now, check out the playground page here on fal.

Introducing FASHN’s Virtual Try-On Solution

FASHN’s virtual try-on solution is a pipeline composed of seven models, each specifically adapted for this task.

At the core of the pipeline is a 1.2B-parameter flow-matching model, trained from scratch on a dataset of 4 million images that were carefully curated and labeled to support virtual try-on functionality. The model’s architecture is designed to preserve garment details, ensuring even the finest text and patterns are transferred accurately at 576x864 resolution.

The solution supports a wide range of clothing categories, including tops, bottoms, jackets, dresses, and more. However, intimate apparel and swimwear were excluded from the training dataset to reduce NSFW biases.

FASHN’s virtual try-on model is designed to be flexible, accommodating variations in starting outfits and poses. That said, it performs best when the model and garment images are well-aligned, requiring minimal warping.

FASHN has also developed a web app interface that runs the model behind the scenes, enabling AI models generation with FLUX, try-on generations and more.

For additional details, we recommend exploring their FAQ, Roadmap, and Usage Guides.

Supports front, back and three-quarter poses with great attention to details.

Where the model shines

FASHN VTON stands out as one of the most accurate virtual try-on models, with the following features:

  • Pixel-Perfect: Excels with graphic prints, ensuring precise preservation of text, textures, and shapes.
  • Fast: Produces 576x864 output resolution in approximately 15 seconds.
  • Customizable: Extended model and garment image controls (background restoration, long garments support, and more) and sampling controls for tailored adjustments.
  • Versatile: Supports both flat-lay and on-model garment inputs, performing well with warping techniques and complex poses.
High warping capabilities and text preservation.

We are excited for you to try out this state of the art model on fal. Stay tuned for new model releases and join our Discord here: discord.gg/fal-ai