AIpparel

AIpparel

Holds the weights and dataset of AIpparel, a multimodal foundation model for digital garments that works with sewing patterns. It suits researchers in garment generation and editing.

Made by georgeNakayama on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: AIpparel is a multimodal foundation model for digital garments, presented as a CVPR 2025 Highlight. This repository is its official storage place for the model weights and the dataset, and it is based on the LLaVA 1.5 7B model.

The dataset is called GarmentCodeData-Multimodal, or GCD-MM. It adds captions, keywords and edit instructions to sewing patterns.

Who it suits: Researchers and developers who study garment generation and editing from text suit it best. You work with the files from Python.

Industry | Who uses it | How they use it | Example
Fashion technology | Researchers | Study sewing pattern generation | Read the captions paired with each pattern
Computer graphics | Graphics developers | Build garment editing tools | Pair an original pattern with an edit instruction
Machine learning | ML engineers | Train or fine-tune on garment data | Use the dataset annotations as training text

Anyone who wants a finished design app can skip it, because the repository holds data and weights, not a tool.

Where to use it: The model runs from the official code repository, and this repository supplies the files.

• Windows, macOS or Linux with Python installed
• The project's official code repository on GitHub
• The project website, linked from the model card

Use case | Who | What they get back
Garment dataset study | Researchers | Sewing patterns with captions and keywords
Pattern editing research | Graphics developers | Original and edited patterns with their edit instruction
Weights for experiments | ML engineers | A pretrained checkpoint to build on

How to set it up: You need Python, the hub tools, and the project's code. The checkpoint leaves out part of LLaVA 1.5, so you fetch that separately.

• Python on Windows, macOS or Linux
• The project's official code repository
• The vision encoder and mm-projector weights of LLaVA 1.5

1. Install Python.
2. Install the Hugging Face hub tools with pip.
3. Download the repository files with the hf command.
4. Fetch the LLaVA 1.5 vision encoder and mm-projector weights separately.

pip install -U huggingface_hub
hf download georgeNakayama/AIpparel

How to use it: The card gives no run commands. It lists the files to download, and the official code repository covers running the model.

Tool | What it does
aipparel_pretrained.pth | The pretrained weights, without the LLaVA vision encoder and mm-projector
gcd_mm_captions.zip | Caption annotations for the GCD-MM dataset
gcd_mm_editing.zip | Editing annotations for the GCD-MM dataset
gcd-mm.parquet | A compact version of all the annotation columns

The parquet file has these columns: idx, descriptive_text, speculative_text, keywords, title, original_pattern, edited_pattern and edit_instruction.

Example: You open the parquet file and read one row. It gives you a sewing pattern's title and keywords, a detailed caption, the original pattern as JSON, an edit instruction and the edited pattern as JSON.

For newcomers and experts: A sewing pattern is the flat set of pieces that make up a garment, and this project teaches a model to read and change them. Your first small try is to download the parquet file and look at a few rows.

Under the hood, the captions come in two kinds. Descriptive text describes the pattern in detail, and speculative text guesses at a situation that suits it. The weights file leaves out the LLaVA 1.5 vision encoder and projector, so you combine it with those files yourself.

Download the weights and annotation files
Fetch the LLaVA 1.5 encoder and projector
Follow the official code repository to load them
Work with sewing patterns and captions

Good to know: This repository stores data and weights only, and the weights alone are not a complete model.

• Base model: LLaVA 1.5 7B
• Language: English
• License: mit
• Cost to run: none named
• Data it reaches: the dataset and weight files you download from Hugging Face
• Not for: running on its own, since the checkpoint lacks the LLaVA 1.5 vision encoder and mm-projector

On Hugging Face, 13 November 2026

• Downloads in the last 30 days: 0
• Likes: 11
• License: mit
• Languages: en
• Based on: llava-hf/llava-1.5-7b-hf

Links

• Model: https://huggingface.co/georgeNakayama/AIpparel
• Files: https://huggingface.co/georgeNakayama/AIpparel/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download georgeNakayama/AIpparel

Then follow the usage in its model card. Nothing in this kit runs by itself.