I’ve done some research but there’s just too much info out there. I want to get a pretty simple logo out of stable diffusion made of all kinds of materials, so I want to teach it about that logo. I want prompts like “a cloud in the shape of <logo>”.
I think I want to create a LoRA, but I’m not 100% sure about that. I looked into that, and the tutorials want me to use windows tools or online tools, but I want to do everything locally.
Is a LoRA what I need? Which training tools are easy to use on Linux, either CLI or automatic111 webui? Will using a training set of the same logo but in different colors be useful?
I generally run to ControlNet if I need something really specific; either finding an existing vector outline of a shape you want or sketching it in yourself, and then reinforcing it with a good prompt. This is the only way I’ve found to get good output that involves specific letters in logos, for example.
The downside, for me at least, is that ControlNet is always a resource hog and it always takes longer than a standard generation.