Spaces:
Running
on
Zero
Apply for community grant: Academic project (gpu)
The project will be a demonstration of the novel ID-Booth framework introduced in our newest scientific paper titled: "ID-Booth: Identity-consistent Face Generation with Diffusion Models", which was accepted for presentation at the 19th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2025). The method builds on the existing DreamBooth fine-tuning approach and entails the use of a novel triplet identity loss to improve the identity consistency of generated images, while enabling better image diversity than methods such as PortraitBooth. Details of the proposed approach can be found on the project site https://dariant.github.io/publications/ID-Booth or in the arXiv paper https://arxiv.org/abs/2504.07392.
As part of this project, we plan to upload a few LoRA models of Stable Diffusion 2.1, which we trained on consenting subjects, to enable the generation of synthetic in-the-wild images of these subjects in predefined environments (e.g. outdoors, at the office, on the bus, ...).
Hi
@DarianT
, congratulations on the release and ZeroGPU grant! It would be wonderful if you could include some examples in the app. This would enhance the user experience and broaden the app's outreach alongside the project.
Gradio docs for examples: https://www.gradio.app/docs/gradio/examples