🌟 A Curated List of Awesome Virtual Try-Off (VTOFF) Research 🌟

Demos

• [🤗 Space] finetuned Stable Diffusion-v1.4 for VTOFF and introduced TryOffDiff.
• [🤗 Space] uses FLUX.1-dev and the LoRA "prithivMLmods/Canopus-Clothing-Flux-LoRA" to generate clothes from text.
• [🤗 Space] trained a new model "cat-try-off-flux" based on FLUX.1-dev and CatVTON.
• [🤗 Space] uses FLUX.1-dev-Redux + FLUX.1-dev-Depth. Set mask as 'structure' and model image as 'style' for targeting VTOFF.
• [fal.ai] trained a LoRA with FLUX.2-klein-9B for vtoff task. They also generate 360-video of the garment with a workflow.

Datasets

• 2021, VITON-HD
• 2022, Dress Code
• 2024, IGPair
• 2025, LAION-Garment
• 2025, Dress Code-MR
• 2026, MV-Fashion
• 2026, Garments2Look

Papers _{^{(do not explicitly target VTOFF but evaluate or incorporate it)}}

^[2025-06-17] FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space
^{[paper] experimented with task, see Figure 6.}

^[2026-03-10] InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editing
^{[paper] experimented with task, see Figure 16.}

^[2026-04-09] FIT: A Large-Scale Dataset for Fit-Aware Virtual Try-On
^{[paper] used Nano Banana Pro as VTOFF method to generate paired dataset.}

^[2026-06-11] A VLM-based framework for evaluating garment consistency in AI-generated images based on DeLong’s theory
^{[paper] developed a garment-specific evaluation framework, surpassing traditional eval metrics.}

Papers _{^{(directly related to VTOFF)}}

^[2020-05-08] TileGAN: category-oriented attention-based high-quality tiled clothes generation from dressed person
^{[paper] targeted vtoff task with a two-stage method.}

^[2021-08-08] ViTon-GUN: Person-to-Person Virtual Try-on via Garment Unwrapping
^{[paper] targeted p2p-vton task: run vtoff first, then vton.}

^[2022-08-11] ARMANI: Part-level Garment-Text Alignment for Unified Cross-Modal Fashion Design
^{[paper] text-to-garment generation.}

^[2023-08-15] SGDiff: A Style Guided Diffusion Model for Fashion Synthesis
^{[paper] finetuned GLIDE to generate garment images from "text" (garment attributes) + "style image".}

^[2023-08-22] DiffCloth: Diffusion Based Garment Synthesis and Manipulation via Structural Cross-modal Semantic Alignment
^{[paper] finetuned StableDiffusion for targeting garment synthesis and manipulation.}

^[2024-01-29] DressCode: Autoregressively Sewing and Generating Garments from Text Guidance
^{[paper] [project] [code] text-to-3dGarment with 2 branches: text-to-sewing-patterns with SewingGPT (3d garment) and text-to-texture with finetuned StableDiffusion.}

^[2024-04-22] FLDM-VTON: Faithful Latent Diffusion Model for Virtual Try-on
^{[paper] [code] vtoff task is included in the loss function for training a vton model, vtoff was not the focus of the paper, nor was a stand-alone task introduced.}

^[2024-04-26] FashionSD-X: Multimodal Fashion Garment Synthesis using Latent Diffusion
^{[paper] finetuned SD-v1.5 with ControlNet to generate garments from text + sketch image.}

^[2024-11-25] Controllable Human Image Generation with Personalized Multi-Garments
^{[paper] [project] [code] generated garment images from segmented garments—extracted from reference images—to construct a dataset for training a multi-garment VTON model.}

^[2024-11-27] TryOffDiff: Virtual-Try-Off via High-Fidelity Garment Reconstruction using Diffusion Models
^{[paper] [project] [code] coined the term Virtual Try-Off (VTOFF), formally introduced the task and introduced the baselines.}

^[2024-11-29] RAGDiffusion: Faithful Cloth Generation via External Knowledge Assimilation
^[paper]

^[2024-12-11] TryOffAnyone: Tiled Cloth Generation from a Dressed Person
^{[paper] [code] finetuned Stable Diffusion-v1.5-inpainting for vtoff in CatVTON-style.}

^[2024-12-16] IGR: Improving Diffusion Model for Garment Restoration from Person Image
^{[paper] finetuned Stable Diffusion-v1.5 for vtoff using 2 UNets (Reference and Denoiser).}

^[2025-01-08] Enhancing Virtual Try-On with Synthetic Pairs and Error-Aware Noise Scheduling
^{[paper] used VTOFF to generate synthetic pairs to enrich VTON training dataset. Trained a U-Net model for warping latent features.}

^[2025-01-27] Any2AnyTryon: Leveraging Adaptive Position Embeddings for Versatile Virtual Clothing Tasks
^{[paper] [project] [code] 1st unified model targeting both VTON and VTOFF.}

^[2025-03-07] DiffDesign: A diffusion model using garment Knowledge-Enhanced for Fashion Design Synthesis
^{[paper] trained a WURSTCHEN-based model for text-to-garment task.}

^[2025-04-17] MGT: Extending Virtual Try-Off to Multi-Garment Scenarios
^{[paper] [project] [code] 1st VTOFF model supporting multiple garment reconstruction. Finetuned StableDiffusion-v1.4 (similar to TryOffDiff) and trained a class embedding for multi-garment support.}

^[2025-04-17] IMAGGarment-1: Fine-Grained Garment Generation for Controllable Fashion Design
^{[paper] [project] [code] garment synthesis with precise control over silhouette, color, and logo placement (3 inputs). Incorporates a tower architecure with SD-1.5 for coarse generation followed by a fine-grained approach utilizing SD-1.5-inpaint.}

^[2025-05-26] ImgEdit: A Unified Image Editing Dataset and Benchmark
^{[paper] [code] trained an image editing model using Vision-Language Model to process a reference image and editing prompt, targeting multiple tasks including VTOFF.}

^[2025-05-27] Inverse Virtual Try-On: Generating Multi-Category Product-Style Images from Clothed Individuals
^{[paper] [project] [code] 1st dual-DiT model for VTOFF with multi-garment support, built on a finetuned StableDiffusion-v3 with modified attention. It accepts images, text, or masks as input, enabling multi-category garment handling.}

^[2025-08-06] Two-Way Garment Transfer: Unified Diffusion Framework for Dressing and Undressing Synthesis
^{[paper] unified framework (TWGTM) targeting both VTON & VTOFF.}

^[2025-08-06] One Model For All: Partial Diffusion for Unified Try-On and Try-Off in Any Pose
^{[paper] [project] unified framework (OMFA) targeting both VTON & VTOFF.}

^[2025-08-06] Voost: A Unified and Scalable Diffusion Transformer for Bidirectional Virtual Try-On and Try-Off
^{[paper] [project] unified framework (Voost) targeting both VTON & VTOFF.}

^[2025-08-25] JCo-MVTON: Jointly Controllable Multi-Modal Diffusion Transformer for Mask-Free Virtual Try-on
^{[paper] trained a VTOFF model for cyclic data generation pipeline.}

^[2025-11-19] UniFit: Towards Universal Virtual Try-on with MLLM-Guided Semantic Alignment
^{[paper] [code] a universal framework driven by Multimodal Large Language Model.}

^[2026-01-05] AlignVTOFF: Texture-Spatial Feature Alignment for High-Fidelity Virtual Try-Off
^{[paper] dual-U-Net pipeline trained with a loss combining regular diffusion loss + LPIPS.}

^[2026-03-10] BridgeDiff: Bridging Human Observations and Flat-Garment Synthesis for Virtual Try-Off
^{[paper] dual-U-Net framework based on Stable-Diffusion-v1.5.}

^[2026-03-23] Dress-ED: Instruction-Guided Editing for Virtual Try-On and Try-Off
^{[paper] introduced a synthetic benchmark Dress Editing Dataset (Dress-ED) and Dress Editing Model (Dress-EM) targeting both VTON and VTOFF.}

^[2026-03-24] OmniDiT: Extending Diffusion Transformer to Omni-VTON Framework
^{[paper] introduced a unified Omni-VTON framework based on the Diffusion Transformer (DiT) that combines model-based VTON, model-free VTON, and VTOFF tasks in a single mask-free model, trained on a self-curated Omni-TryOn dataset of over 380k image pairs.}

^[2026-04-09] What Matters in Virtual Try-Off? Dual-UNet Diffusion Model For Garment Reconstruction
^{[paper] introduced a Dual-UNet Diffusion Model for VTOFF, thoroughly ablating design choices in generation backbones, conditioning (masks, inputs, semantics), and training strategies/losses to reconstruct canonical garments from draped images.}

^[2026-07-09] MMTryOff: multi-category virtual try-off with mask-free inference via diffusion transformer
^{[paper] LoRA training with FLUX.1-dev, on newly proposed VITOFF-HD dataset, incorporating frequency loss and mask loss during training. CatVTON-style training. No code, no dataset available.}

^[2026-08-29] RAGDiffusion++: From Macro-Retrieval to Micro-Fidelity Alignment for Garment Generation
^{[paper] follow-up to RAGDiffusion: Dual-Image-Stream FLUX + AR-GRPO RL post-training with Garment-RM to recover high-frequency textures (weaves, logos) after RAG’s macro-constraints. Introduces STGarment-Plus.}