Leffa (open-weight VTON)

Leffa (CVPR 2025) is a dedicated person-image try-on model: it learns flow fields in attention so garment logos and fine texture stay aligned. OpenTryOn runs the official inference stack locally (extra="local").

Registry id	leffa
Weights	franciszzj/Leffa
Code	github.com/franciszzj/Leffa (MIT)
Paper	arXiv:2412.08486
VRAM	A100 ~6s/image at fp16; 12GB+ recommended
License	Code MIT; confirm the HF weight card before commercial D2C

This is not Qwen-Image / Muse composition try-on. It is a garment-fit specialist (VITON-HD + DressCode checkpoints, plus pose transfer).

Install

pip install opentryon[local]

CUDA GPU required. First run:

1. Downloads the Leffa GitHub source (or uses LEFFA_HOME if you already cloned it).
2. snapshot_downloads franciszzj/Leffa (try-on pth + demo preprocess ckpts).

Environment

# Optional
# LEFFA_HOME=/path/to/clone/of/franciszzj/Leffa
# LEFFA_CKPT=/path/to/hf/snapshot
# HF_TOKEN=hf_...   # only if the Hub repo is gated for your account

CLI

opentryon vton --model leffa \
  --person-image person.jpg --garment-image top.jpeg \
  --garment-type upper_body --vt-model-type viton_hd --dry-run

opentryon vton --model leffa \
  --person-image person.jpg --garment-image top.jpeg \
  --mask-image agnostic.png --steps 30 --seed 42

--mask-image (white = clothing region to replace) is strongly recommended. Without it, OpenTryOn tries Leffa's AutoMasker/DensePose extras; if those deps are missing, it falls back to a geometric mask and a blank DensePose map (lower quality).

--control-type pose_transfer uses the DeepFashion pose-transfer checkpoint (src = target pose person, garment / ref = appearance person).

Python

from tryon.models import LeffaAdapter

adapter = LeffaAdapter()  # lazy-loads weights on first call
images = adapter.generate_and_decode("person.jpg", "garment.jpg")
images[0].save("tryon.png")

MCP

Tool: vton_leffa. Needs pip install opentryon[local] + GPU. Restart MCP after upgrading OpenTryOn so TryOn Studio lists it.

See also

• CatVTON — lighter <8GB concatenation try-on (CC BY-NC-SA)
• Qwen-Image local — composition I2I, not a VTON specialist
• Hosted dedicated VTON: OutfitAnyone-Plus, Photoroom, Google Vertex, FLUX VTO