Spaces:

longlian

/

llm-grounded-diffusion

like 37

Runtime error

LLM-grounded Diffusion is a personal project that enables LLM grounding to diffusion models (https://llm-grounded-diffusion.github.io/).

LLM-grounded Diffusion allows enhanced prompt understanding for text-to-image generation models such as stable diffusion. Our method significantly improve the prompt understanding compared to baseline SD and allow multi-round dialog-based text-to-image generation and generation with prompts in languages that SD doesn't support.

We already have a working local version of the space that includes the full pipeline (both text-to-layout and layout-to-image generation). However, due to a limited personal budget (as I'm a PhD student) and the fact that image generation is computationally intensive, I'm not able to make the layout-to-image part work without a GPU accelerator.

In order to allow people in the community to try our model and compare with baseline image generation methods, I'm requesting an Ampere-series GPU accelerator to this space (for its support of FlashAttention). I hope the hf team could kindly understand the request.

Thanks for the support to this project and the hf space!