A virtual try-on model that generates photorealistic images directly in pixel space without requiring segmentation masks.
FASHN VTON v1.5 is a state-of-the-art virtual try-on model based on the MMDiT (Multimodal Diffusion Transformer) architecture. Given a person image and a garment image, the model generates a photorealistic image of the person wearing the garment. It supports both model-worn garments and flat-lay product shots.
1git clone https://github.com/fashn-AI/fashn-vton-1.5.git
2cd fashn-vton-1.5
3pip install -e .
The human parser weights (~244 MB) are automatically downloaded on first use.
1from fashn_vton import TryOnPipeline
2from PIL import Image
3
4# Initialize pipeline (auto-detects GPU)
5pipeline = TryOnPipeline(weights_dir="./weights")
6
7# Load images
8person = Image.open("person.jpg").convert("RGB")
9garment = Image.open("garment.jpg").convert("RGB")
10
11# Run inference
12result = pipeline(
13 person_image=person,
14 garment_image=garment,
15 category="tops", # "tops" | "bottoms" | "one-pieces"
16)
17
18# Save output
19result.images[0].save("output.png")
1python examples/basic_inference.py \
2 --weights-dir ./weights \
3 --person-image person.jpg \
4 --garment-image garment.jpg \
5 --category tops
FASHN VTON v1.5 was trained from scratch in pixel space using a two-phase approach:
Training optimizations included token dropping up to 75% to reduce computational demands.
1@article{bochman2026fashnvton,
2 title={FASHN VTON v1.5: Efficient Maskless Virtual Try-On in Pixel Space},
3 author={Bochman, Dan and Bochman, Aya},
4 journal={arXiv preprint},
5 year={2026},
6 note={Paper coming soon}
7}
This model is released under the
Apache-2.0 License.
