Fashn Vton 1.5

FASHN VTON v1.5 is an open weights virtual try on model from FASHN AI. Given an image of a person and an image of a garment, it generates a photorealistic result showing the garment worn, working directly on RGB pixels rather than in a compressed latent space and without requiring a segmentation mask of the body. Both choices are deliberate: latent warping distorts colors, logos, patterns and textures, while masking erases body specific characteristics and limits how much the shape of a garment can change. The architecture is a multimodal diffusion transformer of 972 million parameters, with eight double stream blocks handling attention between person and garment tokens, sixteen single stream blocks over the combined sequence, and four patch mixer blocks ahead of the main transformer. The only spatial reduction is a patch embedding of twelve by twelve. Pose keypoints are supplied as a grayscale channel and category embeddings cover tops, bottoms and one pieces. Training ran in two phases, first on 18 million masked pairs and then on an even mix of those pairs and 4 million synthetic triplets produced by the first checkpoint, which is what allows try on from an unmasked input. Inference takes roughly five seconds on an H100 class GPU, fast enough for interactive consumer use. Code and weights are released under Apache 2.0 for research and commercial use, with an online demo and a paper listed as forthcoming. The model is free; the company's commercial products are not. The creative app publishes Basic at $19 per month with 200 monthly credits, Pro at $49 per month with 750 monthly plus 50 daily credits, and Agency at $99 per month with 1,500 monthly plus 100 daily credits, with annual billing described as roughly two months free and credit top ups at $0.10 per credit in minimum lots of 100. Developer API credits start at $7.50 with no subscription, and commitment plans are quoted.