AI Stuff

Tuesday February 27, 2024

I last praised the M1 Mac - it’s not that great for more complicated workloads unfortunately. I (mistakenly) have a strong PC for these situations!

In addition to A111’s StableDiffusion Web UI, I really think ComfyUI is an even better way to understand things under the hood better (making the option to transition to more theoretical stuff easier for me).

Though I’m still learning the weeds of things, I think it’s a fun entry way to get excited about some of the recent developments that are happening in this space.

I think with respect to ethics of AI (in addition to the biases, I think another huge thing is not using copyrighted work without permission) this can get a little tricky too so I think it’s important to keep that in consideration while generating fun stuff.

Today I’m pretty proud that I can say that now I actually know what I am doing without guessing. I wanted to make myself into an 80s style anime character (think Hayao Miyazaki or Toshihiro Kawamoto style).

I present to you, anime Shah:

For more information on what I did to achieve these results, click me

So how did I do this? I first used a prompt (positive and negative) to just generate a regular picture using an anime style checkpoint. You can get checkpoints from HuggingFace. Started with a blank image (i.e txt2img) then continued using previous outputs to generate what I want.

After some configuration I got this:



OK cool. How can I make this look more like me?

I used https://github.com/Gourieff/sd-webui-reactor which was pretty hilarious but, helped to achieve my results:



After this, I applied a denoise of values < 0.5, so that it would weigh this source image and slowly changed sampling steps until I got the result I shared above.

This was pretty fun. No, I am not taking commissions 😛