Description
Initial.
The LoRA was trained on 100 images from 2024 through 2026.
The dataset consists of:
year 2026: 51 imagesyear 2025: 35 imagesyear 2024: 14 images
The latter half of the showcase features comparisons between this LoRA, NekosokoStyleAnima.safetensors by Ai_Corn (trigger: @nekoshokostyle), and no LoRA—that is, the artist's style as learned by Anima Turbo-v1.1 itself.
dim32/alpha32, CAME, UNet LR 3e-05, cosine 1,000 steps. 100 images (1024²) 😭