much improved version of DiffusionDemo Dave Touretzky (29 Mar 2025 04:28 EDT)
Re: [AI4K12] much improved version of DiffusionDemo Ken Kahn (03 Apr 2025 07:23 EDT)
Re: [AI4K12] much improved version of DiffusionDemo Dave Touretzky (03 Apr 2025 18:34 EDT)

much improved version of DiffusionDemo Dave Touretzky 29 Mar 2025 04:28 EDT

I'm pleased to announce a much improved version of our Stable Diffusion
demo by my student, Adithya Kameswara Rao.

The improvements are:

1. No more need to run Google Colab.  The demo is now hosted at the
Indiana Supercomputing Center's JetStream 2 facility.  This allows us to
operate it as an always-on application that starts instantly, so you
don't have to wait 2-3 minutes for the application to load as you did in
Colab.

2. Easy to remember URL:  https://diffusiondemo.org

3. Much better image quality.  The move to JetStream 2's more powerful
GPU cluster allowed us to switch to a larger diffusion model without
harming response time.  The demo is still very fast, so you can do many
experiments in a short amount of time.

4. Minor improvements to the user interface.

This demo nicely complements our "Cats in Latent Space" video, which
can be viewed at https://www.youtube.com/watch?v=hb-KT66rCT8

Using the various tabs in the demo, you can directly experiment with
phenomena shown in the video, such as iterative noise removal and
exploring nearby points on the latent image manifold.

This work was funded by grants from NEOM Company and National Science
Foundation award IIS-2112633.

-- Dave Touretzky