← Home

Week 1156

I took 3 days off from work this week to learn more about image generation with stable diffusion. I installed this tool called ComfyUI which breaks down image generation workflows into easily customizable steps. It’s a nice abstraction on top of pytorch code where you can customize your models, parameters, and other workflow options without needing to write code yourself. I’ve found it helpful for learning the different parts that go into image generation:

It’s amazing that this is an open source project that anyone can try. I would really like to contribute to the codebase at some point and give back to this tool and it’s community.

My friend Justin and I created ghiblify.com as a placeholder for some product experiments that we want to run together. Our first one is a tool to re-style images as if they were drawn in the style of Studio Ghibli. He made a tiktok to promote the site which currently has around 36K views and generated some inbound asking us to do contract image editing work.

The subtitle says we are out of GPU capacity but in reality I just haven’t had much time to figure out the infrastructure side of things to deploy our local workflows on a cloud GPU. I got pretty busy with Retool work and got a bit distracted by other miscellaneous stuff.

--------

Some miscellaneous stuff:

--------

Reading: