Given a person and a garment image, virtual try-on (VTO) aims to synthesize a realistic image of the person wearing the garment while preserving their original pose and identity. Although recent VTO methods excel at visualizing garment appearance, they largely overlook a crucial aspect of the try-on experience: the accuracy of garment fit -- for example, depicting how an extra-large shirt looks on an extra-small person. A key obstacle is the absence of datasets that provide precise garment and body size information, particularly for "ill-fit” cases, where garments are significantly too large or too small. Consequently, current VTO methods default to generating well-fitted results regardless of the garment or person size.
In this paper, we take the first steps towards solving this open problem. We introduce FIT (Fit-Inclusive Try-on), a large-scale VTO dataset comprising over 1.13M try-on image triplets accompanied by precise body and garment measurements. We overcome the challenges of data collection via a scalable synthetic strategy: (1) We programmatically generate 3D garments using GarmentCode and drape them via physics simulation to capture realistic garment fit. (2) We employ a novel re-texturing framework to transform synthetic renderings into photorealistic images while strictly preserving geometry. (3) We introduce person identity preservation into our re-texturing model to generate paired person images (same person, different garments) for supervised training. Finally, we leverage our FIT dataset to train a baseline fit-aware virtual try-on model. Our data and results set the new state-of-the-art for fit-aware virtual try-on, as well as offer a robust benchmark for future research. We will make all data and code publicly available.
FIT Dataset
The FIT dataset consists of 1,137,282 training and 1000 test samples, each consisting of $(I_{\text{try-on}}, I_{\text{p}}, I_g, m_p, m_g)$. FIT covers 168 distinct body shapes (82 men's, 86 women's) in sizes XS-3XL, 528 body poses, as well as 158,483 unique top and garment designs. The garments come in a diverse range of fits, from loose to tight. Refer to our paper for additional dataset statistics.
Select a data sample to view the try-on image, paired-person image, layflat garment image, and measurements. 📏
We compare FIT to related datasets. FIT is the first large-scale dataset for virtual try-on that provides photorealistic images, ill-fit examples, precise measurement information, and ground-truth paired-person images. For scale, we report the number of training images.
| Dataset | Realism | Ill-Fit | Size Info | Triplet | Scale |
|---|---|---|---|---|---|
| SV-VTO | ✓ | ✓ | ✓ | ✓ | 1,524 |
| SIZER | ✓ | ✓ | ✓ | ✗ | 2,000 |
| DeepFashion3D | ✗ | ✗ | ✗ | ✗ | 2,078 |
| ViTON-HD | ✓ | ✗ | ✗ | ✗ | 11,647 |
| LAION-Garment | ✓ | ✗ | ✗ | ✓ | 60K |
| SewFactory | ✓ | ✗ | ✓ | ✗ | 1M |
| GCD | ✗ | ✗ | ✓ | ✗ | 115K |
| Ours | ✓ | ✓ | ✓ | ✓ | 1.13M |
Try on different sizes of the same garment!
Garment Size
S Person, XL Garment
Prior virtual try-on works excel at garment appearance transfer and generating aesthetically-pleasing images. However, past methods lack precise measurement conditioning and instead hallucinate garment fit. In contrast, Fit-VTO generates high-quality virtual try-on results, while also being measurement-conditioned. Below, we show how Fit-VTO excels on synthetic FIT images with measurement conditioning, as well as real-world VITON-HD images (without measurements).
Inputs
Any2AnyTryOn
Nano Banana Pro
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2AnyTryOn
Nano Banana Pro
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2AnyTryOn
Nano Banana Pro
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2AnyTryOn
Nano Banana Pro
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2AnyTryOn
Nano Banana Pro
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
Inputs
Any2Any
Nano Banana
COTTON
IDM-VTON
Ours
Ground Truth
(a) Overall workflow: We start by simulating a 3D garment on a target body via GarmentCode to render a synthetic image $I_s$. We generate a text prompt $p$ (via VLM) and a composite normal map $I_n$ (stitching estimated normals with realistic head/feet details). These condition our re-texturing model $f_\text{texture}$ to produce the try-on image $I_{\text{try-on}}$. Finally, we use $f_\text{paired}$ to generate a paired person image $I_p$, and a VLM to synthesize a layflat garment $I_g$. (b) GarmentCode simulation: Given a sampled design template, we compute sewing patterns for a specific body size. Then, we cross-drape these patterns onto a different target body, using box-mesh realignment to prevent simulation failures, and extract ground-truth measurements. (c) Using source and target garments draped on the same body, we derive an identity map $I_\text{id}$ by masking the garment in $I_{\text{try-on}}$. Conditioned on $I_\text{id}$, a paired normal map $I_n'$, and a paired prompt $p'$, $f_\text{paired}$ generates the paired person image $I_p$.
Our architecture is a flow-based model based on Flux.1-dev MMDiT architecture and finetuned with LoRA. FiT-VTO generates a try-on image $I_{\text{try-on}}$ given a layflat garment image $I_g$, paired person image $I_p$, and person-garment measurements $m = [m_p, m_g]$. First, image inputs $I_g$ and $I_p$ are encoded into latents separately through a pre-trained VAE encoder. We replace the text embeddings in Flux.1-dev with custom measurement embeddings $m_{\text{embed}}$ computed from $m$. Person latents are channel-concatenated with the noisy target latents, while layflat latents and $m_{\text{embed}}$ are sequence-wise concatenated with $z_t$. After processing through the diffusion transformer, clean latents are decoded by the VAE decoder.
@article{fitvto2026,
author = {Karras, Johanna and Wang, Yuanhao and Li, Yingwei and Kemelmacher-Shlizerman, Ira},
title = {FIT: A Large-Scale Dataset for Fit-Aware Virtual Try-On},
journal = {SIGGRAPH},
year = {2026},
}