Use "by &lt;artist&gt;" for styles Add artists at the beginning for stronger effectIMPORTANT: If you find the model lacking in some aspect, swap out the CLIP; UNet is flexible Eg: for stronger artist styles/text, use Tofu CLIP for stronger character knowledge, use Urang Diffusion/holodayo/kivotos CLIPPrevious versionsExperimental anime model merge. Model might fall back to default SDXL anime style on some prompts. I encourage you to experiment and would love to see your gens! The FP16 fixed VAE is included.beta2: Updated merge recipe with 2 new models added. Model is now slightly better at composition and prompt understanding. Recommendations: Resolution: Carefully choose an aspect ratio that suits your composition. Mismatched aspect ratio can cause the model to refuse to listen properly to the prompt. Positive prompt: Add "anime art style" at the beginning. Use a hybrid approach with a description of the composition followed by tags specifying the elements. Negative prompt: zombie, comic, sketch Sampler: DPM++ 2M SDE Karras Steps: 20 or more CFG: 8 or higherbeta1: Recommendations: Resolution: Carefully choose an aspect ratio that suits your composition. Mismatched aspect ratio can cause the model to refuse to listen properly to the prompt. Positive prompt: Add "anime art style" at the beginning to reinforce anime style if required. Use a hybrid approach with a description of the composition followed by tags specifying the elements. Colour tags can bleed into other tags, use Cutoff/PromptAttention to fix this. Negative prompt: zombie, comic, sketch Sampler: DPM++ 2M SDE Karras Steps: 20 or more CFG: 8 or higher

LS Vividus XL

Rebuilt from scratch with new recipe

Files with SHA256: 7eb329eff5b1cd061e78ac9074c5f3b5ba94028acbd0297ad9ba26e7b6503435