Nekofantasia Alpha (workflow and links) — v0.1

Use this ComfyUI workflow to generate anime images with Nekofantasia 0.1 from positive and negative text prompts.

Use comma-separated tags without underscores; the release recommends longer, detailed prompts for version 0.1.

Suggestion · not verified

Place Nekofantasia-01.safetensors in ComfyUI/models/checkpoints, and t5xxl_fp16.safetensors, clip_l.safetensors, and clip_g.safetensors in ComfyUI/models/text_encoders.

Suggestion · not verified

Install the ComfyUI-RK-Sampler node pack.

Suggestion · not verified

The listed model files are Nekofantasia-01.safetensors, t5xxl_fp16.safetensors, clip_l.safetensors, and clip_g.safetensors.

The required node pack is ComfyUI-RK-Sampler.

The release recommends Dopri5 or Bosh3 with CFG Scale 3–7; Dopri5 adjusts its step count and size adaptively.

Suggestion · not verified

The release recommends aiming for about 1 MP, with each side divisible by 64; it reports that portrait images tend to produce better results than landscape.

Suggestion · not verified

The prompt guidance suggests adding highres and absurdres tags.

Suggestion · not verified

The release describes version 0.1 as an early preview that has not completed training; it reports significant issues with fingers and fine details, and says character tags currently have almost no effect.

The release says uncommon or missing booru tags can rarely cause photorealistic style bleed-through, and NSFW generation is currently almost impossible because of base SD 3.5 safety filters.