ImgDecom

Official LoRA weights for Cycle-Consistent Tuning for Layered Image Decomposition (CVPR 2026). Built on FLUX.1-Fill-dev, the model separates a logo from its object into an isolated logo and a logo-free object. Cycle-consistent tuning trains the decomposition and recomposition tasks together.

This repository contains the LoRA adapter; the FLUX.1-Fill-dev base model is required. See the code and inference instructions and test dataset.

Paper ยท Project page

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for picana/ImgDecom

Adapter
(203)
this model

Paper for picana/ImgDecom