📌 Version Notes
My V1 dataset was a bit of a boo-boo — everything was cropped to 512x512, which limited what the model could do. I’m already in the middle of upscaling and regenerating the full dataset properly.
Once that’s done, I’ll dive into finetuning, training fresh LoRAs, merging them back in — the whole ordeal. Since I’m doing this on rented GPUs (because apparently I enjoy hemorrhaging money), V2 should be ready by the end of the week or the weekend… assuming nothing explodes.