Tencent’s HY WU turns any outfit into a reusable LoRA wardrobe

HY WU is a Tencent-built image editor optimized for clothes swapping and style transfer at high fidelity. Users feed an initial photo plus one or more reference outfits, and the model redraws the subject.

The chosen model wears new clothes while preserving identity and garment details across multiple examples. Under the hood, HY WU converts text and reference images into a compact “code,” instantly training a tiny LoRA.

It injects it into the base editor for each request. In benchmarks, HY WU beats open tools like Qwen ImageEdit, LongCat ImageEdit, and Flux 2.
It even tops some closed systems such as Cream 4.5 and GPT-Image 1.5. Still, models like NanoBanana 2 and its Pro variant still win overall.

The code is open-source with full setup instructions, but the base model is heavy, needing 8x40GB or 4x80GB VRAM. Of course, this is out of reach for most hobbyists. A distilled checkpoint is planned to bring those requirements down.