QuanjianSong

QuanjianSong / FashionChameleon

Public

Official Pytorch Code of the Paper "FashionChameleon: Towards Real-Time and Interactive Human-Garment Video Customization"

quanjiansong.github.ioprojectsFashionChameleon fashion garment-switch interactive real-time video-customization
36
0
80% credibility
Found May 18, 2026 at 60 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

FashionChameleon is a research project from academic researchers that enables real-time video customization of clothing on people. The system can swap garments on a person in streaming video while maintaining natural movement, achieving speeds of about 24 frames per second on a single graphics card. This technology could be used for virtual try-on experiences in shopping, creating fashion content for social media, or visualizing clothing designs. The project is still in active development with plans to release the code and trained models publicly. It includes a published research paper explaining the approach and a project website with demonstration videos.

How It Works

1
🔍 Discover FashionChameleon

You come across a new video technology that can change clothes on a person in real-time videos.

2
👀 See It In Action

You watch demo videos showing how a person's outfit can be swapped instantly while they move naturally.

3
⚡ Experience Real-Time Speed

The system generates fashion videos at 24 frames per second—fast enough to feel instant and responsive.

4
🛍️ Imagine the Possibilities

You think about how this could let shoppers try on clothes virtually, or help creators make fashion content quickly.

5
Choose Your Path
📄
Read the Research Paper

Dive into the academic paper to understand how the garment customization technology works.

🎥
Browse the Demo Gallery

Watch example videos showing different fashion customization scenarios.

6
🔮 Look Forward to the Code

The researchers promise to release their tools publicly so others can try it themselves.

✨ Fashion Videos Reimagined

You've discovered a glimpse of how AI could transform how we shop for clothes and create fashion content.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 60 to 36 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is FashionChameleon?

FashionChameleon is a PyTorch-based research project for real-time video customization that lets users interactively swap garments on a person mid-generation. It uses streaming autoregressive video generation with a teacher-student distillation approach to achieve 23.8 FPS on a single H200 GPU, which is 30-180x faster than existing customization methods. The framework supports garment switching without requiring multi-garment video data during training.

Why is it gaining traction?

The hook is real-time performance with interactive control. Unlike static image-to-video customization pipelines that require retraining per garment, this system allows live garment switching during generation using a training-free KV cache rescheduling technique. The gradient-reweighted distribution matching also addresses a known problem in self-rolled video generation where later frames degrade in quality.

Who should use this?

Researchers in video generation and fashion tech who want to build e-commerce try-on experiences, virtual fitting rooms, or content creation tools. If you're working on interactive video applications where latency matters, this is worth watching. Early adopters comfortable with cutting-edge research code will get the most value.

Verdict

This is promising research that is not yet ready for production use. The project has only 36 stars, most code components remain unreleased (checkpoints, pretraining, post-training code, and benchmarks are all marked TODO), and only the paper and project page are currently available. Given the 0.800000011920929% credibility score, wait for the actual code release before evaluating further.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.