BagGEN

Technical Report | Code | Dataset

BagGEN is a fine-tuned IP-Adapter for texture-conditioned bag image generation. Given a texture patch from a reference image and a bag category description, it generates bag images guided by the reference texture and material appearance. BagGEN can optionally be combined with ControlNet, using a contour or scribble as an additional condition to guide the bag's structure.

Model Input Model Output
Texture patch
+ bag category description
+ optional contour or scribble condition via ControlNet
A generated bag image guided by the input texture, category, and optional structure condition
BagGEN sample results

Sample results using texture patches extracted from real-world product images. Insets show the input texture patches and, where applicable, the contour or scribble conditions.

License and Model Provenance

BagGEN adapter weights are released under the Apache License 2.0.

BagGEN was fine-tuned from ip-adapter_sdxl_vit-h.bin provided by IP-Adapter, using SynBag50K. Fine-tuning updates the image projection module and image-conditioned attention parameters while keeping the SDXL backbone and image encoder frozen.

The released checkpoint contains adapter weights only. The SDXL base model, image encoder, VAE, and optional ControlNet must be downloaded separately. Users must comply with the respective licenses of these models, including the CreativeML Open RAIL++-M license for SDXL.

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. 🙋 Ask for provider support