OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Image & Video AI

Multimodal Garment Designer

Multimodal Garment Designer is a Python research project implementing human-centric latent diffusion for fashion image editing. It guides image generation with multimodal inputs including text, human body poses and garment sketches, and provides an evaluation workflow for Dress Code Multimodal and VITON-HD Multimodal datasets.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Multimodal Garment Designer — OpenViST project profile in Image & Video AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you want a research codebase and evaluation workflow for multimodal-conditioned fashion image editing, rather than a hosted design interface.

Who it is for

  • Computer-vision researchers studying multimodal fashion image generation and editing
  • Developers exploring latent-diffusion methods for fashion imagery

Documented uses

  • Generate human-centered fashion images guided by text, body poses and garment sketches.
  • Evaluate the approach on Dress Code Multimodal or VITON-HD Multimodal data in paired or unpaired test settings.
  • Load the pretrained MGD denoising UNet through Torch Hub or integrate it with the documented custom Diffusers pipeline.
Before you choose

Requirements & limits

  • The documented workflow is Python-based and recommends Conda; the setup instructions specifically describe Linux.
  • The repository provides sample data for Dress Code and VITON-HD, while full-dataset runs require downloaded and prepared data.
  • Inference is exposed through a command-line evaluation script, with output paths, dataset selection, test order and conditioning parameters supplied by the user.
  • The repository documents access to pretrained model checkpoints through Torch Hub and a custom Diffusers pipeline.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up the MGD Python environment

Clone the repository and create its documented Conda environment before proceeding to inference.

Before you start

  • Linux system, as described in the repository setup guidance
  • Anaconda or Conda
  • Git

Steps

  1. Clone https://github.com/aimagelab/multimodal-garment-designer.

  2. From the repository, create the environment with conda env create -n mgd -f environment.yml.

  3. Activate it with conda activate mgd.

Expected outcome

The repository’s `mgd` Conda environment is created and activated; inference is the next documented stage.

Guide limitations

  • This setup recipe does not run inference or confirm compatibility on a particular machine.
  • The README also gives an alternative manual environment using Python 3.9 and listed package versions.
Documentation 1 ↗
Source-based practical guide

Run the documented inference evaluation

Run the evaluation script on one of the repository’s included sample datasets and save generated images.

Before you start

  • A configured MGD Python environment
  • Repository sample data under assets/data
  • An output directory path and a save-folder name

Steps

  1. Choose dresscode with dataset path ../assets/data/dresscode, or vitonhd with ../assets/data/vitonhd.

  2. Run the command below with --dataset_path, --dataset, --output_dir and --save_name, along with the documented batch, precision, worker, conditioning-rate and test-order options.

    Documented command
    python src/eval.py

  3. Select paired or unpaired for --test_order; the README’s example uses --mixed_precision fp16, --sketch_cond_rate 0.2 and --start_cond_rate 0.0.

Expected outcome

The evaluation script is instructed to save generated images in the selected output directory and named subfolder.

Guide limitations

  • The recipe covers the included sample data; whole-dataset inference requires downloaded and prepared Dress Code Multimodal or VITON-HD Multimodal data.
  • The documented command uses placeholders for paths and integer or string values, which must be supplied for the intended run.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
aimagelab/multimodal-garment-designer · Official homepage title · retained in its source language
How we review sourcesSuggest a correction