I'm writing on behalf of Astronaut Horse, a group of programmers and artists at the Recurse Center interested in applying for a community GPU grant. Since late 2022 we've been studying Stable Diffusion using fast.ai and Hugging Face resources with the goal of developing practical tools for visual artists.
Our most visible current project, which we wish to expand with the community GPU grant, is an ongoing series of collaborations with visual artists centered on textual inversion trainings using artist-supplied images and text prompts.
This project began with a live demonstration of a Gradio / Hugging Face Space deployment given to the Recurse Center community, after which we began reaching out to individual working studio artists inviting them to collaborate. Using textual inversion, we trained Stable Diffusion on images selected by a given artist and then attached that training to a unique token that was uploaded to the Hugging Face SD concepts library. Surprisingly, the fine-tuned concepts have consistently produced images that avoid rote imitations of the input artworks and instead gravitate towards images that are strongly in-line with the artist’s more nuanced visual tastes. The end result is that the output images seem to have a practical use similar to that of an academic peer critique, highlighting underlying patterns and making visual suggestions about possible next progressive steps in style and content.
With a community GPU grant we could immediately deploy (and debug) the Gradio interface currently in this space that allows an artist (or anyone) to freely load in and prompt any fine-tuned artist’s concept that we have trained. And with increased artist engagement, we could begin soliciting ideas for more artist tools that could include an up-scaler, a prompt transition gif animator, a style transposer and a text-based image editor to name a few that we have been working on implementing.
Thank you in advance for your consideration. We're all so grateful for the work done by the team at Hugging Face and the resources you provide! To see examples of what we've been working on, please visit our Github org page or our static site for artist collaboration images.
Hi @tombetthauser , we have assigned a gpu to this space. Note that GPU Grants are provided temporarily and might be removed after some time if the usage is very low.
To learn more about GPUs in Spaces, please check out https://huggingface.co/docs/hub/spaces-gpus
Thank you so much @akhaliq we're super excited to start making full use of the space!
@akhaliq Not sure who to ping about this but we overhauled our project yesterday to allow flexible SD image dimensions and are having issues.
It was fully debugged and working well for about 30-50 uses then started throwing an error for smaller and smaller dimensions. Seems inconsistent, not sure exactly what's going on but based on running it in colab it seems like just a resource problem?
Assuming it is a resources issue is there a process for requesting a bump up to an A10G or A100?
Thanks in advance!
Moving this to a new discussion!