Zelda

249 Pictures + 4 vids

Zelda has a lot of different designs, which is less than ideal for consistency. I mostly overcame this by generating tons of images and deleting the ones that merged styles. There are some tastefully done in-betweens left in because they're genuinely good images.

We have videos now. They are... well... basically animated gifs. I did a lot of experiments with video generation for not a lot of gain. I saw some of the generated animations PuffyArt made and realized how much further I have to go. The problem is, I don't know how to get there. I follow the guides, I experiment, I try things in between approaches. I ask ChatGPT and Claude for help. I dig through Hugging Face comment sections hunting for fixes in Python code. I read dev blogs and download patches. I watch patch notes for my video generation software being read out by one guy to his webcam. I see that developer's Patreon and he makes about 70 dollars a month. I die a little inside. "The works."

I talked about wanting a tighter set of better images previously. I don't know if I've achieved that or even where to go from here. I could delete more images I think are the worst from the set, but to what end? The only thing that accomplishes is preventing someone from seeing an image they might have liked when I didn't.

How I thought the process would go: I would choose to improve fewer images to meet my expectations threshold, then improve them in even more detail. What actually happened: I used a new model for cleaning, and the number of images already hitting my "good enough" threshold was much higher. The manual cleaning and editing that really made images feel special to me was reduced to "just making these parts look less weird" (with minor exceptions). I have no idea if that comes through in the final work.

When I look at the images, all I see is a lack of variety. The images that generate most easily meet the "good enough" threshold more often and don't get deleted. I don't know how to compensate for that without just deleting more of them out of fairness to the other image types.

I've lost my train of thought on the points I was trying to make. I thought about making a book with AI, then remembered I don't read books, so I wouldn't know literary art if it hit me in the face and sodomized my dinner.

Despite all of this sounding like whining, it's okay. One day (maybe a month from now, maybe ten years from now), you'll unsubscribe from this Patreon because you don't need it anymore. This will be a good thing. You won't need me anymore to see the content you want.

One day, you'll be able to download a program that takes no setup and you can just ask it whatever you want to see and it'll make it. You'll be able to ask it to change anything you want. No models, no trigger words, no GitHub. You'll just be able to see whatever you want, however you want to see it. There are already LLMs made to run alongside image generation software to take your words and make the prompt for you (locally, without copy and paste). This seems to be the future we're heading toward.

Sorry for rambling. I don't have anyone to talk to and I'm desperate for human connection, even if it's just me screaming into ✨the void✨.