Spaces:
Running
on
Zero
Apply for community grant: Academic project (gpu and storage)
Title: InfiniteYou: Flexible Photo Recrafting While Preserving Your Identity
Project Page: https://bytedance.github.io/InfiniteYou/
GitHub: https://github.com/bytedance/InfiniteYou
Model: https://huggingface.co/ByteDance/InfiniteYou
Our paper has been submitted to arXiv and will appear very soon.
Abstract: Achieving flexible and high-fidelity identity-preserved image generation remains formidable, particularly with advanced Diffusion Transformers (DiTs) like FLUX. We introduce InfiniteYou (InfU), one of the earliest robust frameworks leveraging DiTs for this task. InfU addresses significant issues of existing methods, such as insufficient identity similarity, poor text-image alignment, and low generation quality and aesthetics. Central to InfU is InfuseNet, a component that injects identity features into the DiT base model via residual connections, enhancing identity similarity while maintaining generation capabilities. A multi-stage training strategy, including pretraining and supervised fine-tuning (SFT) with synthetic single-person-multiple-sample (SPMS) data, further improves text-image alignment, ameliorates image quality, and alleviates face copy-pasting. Extensive experiments demonstrate that InfU achieves state-of-the-art performance, surpassing existing baselines. In addition, the plug-and-play design of InfU ensures compatibility with various existing methods, offering a valuable contribution to the broader community.
@hysts Could you please grant GPU and storage for our demo in this space? Thank you so much for your support!
@hysts
Here is our arXiv paper link: https://arxiv.org/abs/2503.16418
We may wish our users to try our demo at the earliest convenience. Thanks a lot for your help. :)
Hi @EndlessSora I can't find your grant request in our internal Slack channel, so I believe you opened this request without using the button in the Space Settings (docs). The button sends a notification to our internal slack channel, where we review grant requests. Even HF staff members who have permission to view private Spaces don't receive notifications from private Spaces, so without using the button, we can't notice your request. So, we would appreciate it if you could use the button next time.
@hysts Got it. Thanks!
@hysts We got an error here "NVML_SUCCESS == r INTERNAL ASSERT FAILED at "/pytorch/c10/cuda/CUDACachingAllocator.cpp":1015, please report a bug to PyTorch." when we run our demo. Could you please help us with it? Is it due to OOM or other issues?
@EndlessSora on my end I haven't performed any iteration with any zero space, yet the it's saying that I reached my quota limitation. Maybe it's OOM as you said?
@EndlessSora Not 100% sure, but the error is often raised when CUDA OOM occurred on ZeroGPU Spaces. FYI, the underlying hardware of ZeroGPU is A100 with 40GB, not 80GB.
@hysts Is it possible for us to host our model on GPUs with a slightly larger memory like 48GB? Thanks!