Model Info:
griboedo4ka art style for Z-Image Turbo
Trained on 57 images in 1080p.
Triggered by "art by griboedo4ka, griboedo4ka style". Strengh 1.0
Presented in 2 versions - 1400 and 2500. I add identical preview images with identical seed for both version, so you can compare and choose. As usual:
Step 1400 is more reliable and stable, sufficiently conveying the style of drawing, but not the small nuances. Overall more accurate drawing in cost of style similarity.
Step 2500 is more accurate and full style reassembling, but can fuse fingers or small details more often than previous one. Overall more full style similarity in cost of drawing accuracy.
v0.1: Due to the lack of a full model, training was performed on distillation, resulting in occasional anatomical distortions. Will be retrained after full model release.
Usage Recomendation:
IMPORTANT: always include prompts like Anime-style drawing or something similar. Z-Image Turbo STRONGLY leans towards photorealism, and unless explicitly instructed to create drawn graphics, it will produce photos or 3D CGI!
Author notes:
1) Please use with respect and support original artist because it's a very talented person.
2) If you want to support me - like, upload your pictures or generate with site generator, cause it give me both motivation and buzz for new Lora.