Hey all! I recently suffered a technopocalypse, and I lost my LoRA for style and my recipe. I've got no choice now but to find a new recipe and train a new LoRA. I can't for the life of me reengineer what I had before, but this is looking promising.
Try as I might, I wasn't getting any luck until I threw in two of my old, failed LoRA from when I first started. They're majorly undercooked so they have very little effect, and I'm using them at a low power, so my best guess is that they're acting a bit like contact lenses. They have little visual effect on their own but they're enough to reduce the model's possible search space and leading to more consistent results.
These are quite pretty! Still not exactly what I want. I'd like something with a bit more vibrancy. Nothing I do seems to work though. Hopefully transfer to NovaFurry, which has a rich and saturated base style, will be the missing ingredient.
I'm out of GPU credits until Wednesday, but I'll start fleshing out full character views then and I should have the new LoRA trained in short order.
Tricky-Style-01 was my very first LoRA, trained on a procedurally generated dataset using this recipe. Tricky-Style-02 was my second LoRA, trained on another procedurally generated dataset using this recipe.
They look okay, eh? My iPad's screen broke - awesome - but once I get it fixed I'm going to start speed-running headshots by hand and see if I can't learn to draw these myself.
They look okay, eh? My iPad's screen broke - awesome - but once I get it fixed I'm going to start sp
