Stable Diffusion training not going great so far. I think a specific, unchanging physical item like a nostr:npub17tyke9lkgxd98ruyeul6wt3pj3s9uxzgp9hxu5tsenjmweue6sqq4y3mgl is harder for it to learn than it is to teach it a new face since it already has a super-deep training set on what a face is.
Could be this LORA approach is more well-suited to learning new variations of a known concept rather than trying to develop a completely new SeedSigner-ness concept.
Still tons of different settings to try.
