Generates anime and illustration images in ComfyUI using the ANIMA 1.0 text-to-image model.
npx skills add https://github.com/artokun/comfyui-mcp --skill anima-base
Please help me install this Agent Skill. Skill: anima-base Source: https://github.com/artokun/comfyui-mcp/tree/main/plugin/skills/anima-base Command: npx skills add https://github.com/artokun/comfyui-mcp --skill anima-base
Setting up the ANIMA 1.0 anime text-to-image model in ComfyUI requires assembling split model files, custom nodes, LoRAs, and LLLite ControlNets with exact node IDs and settings; this Skill provides verified workflows and configurations so the graph runs correctly the first time.
Use the anima-base skill to build a ComfyUI workflow that generates an anime illustration of a character with the turbo LoRA at 896x1152.
High-intent search queries and answers about installing and using this skill.
Load anima-base-v1.0.safetensors with UNETLoader, qwen_3_06b_base.safetensors with CLIPLoader set to type stable_diffusion, and qwen_image_vae.safetensors with VAELoader, then sample with er_sde and the simple scheduler. Add the turbo LoRA for 12-step generation at CFG 1.
The anima-turbo-lora-v0.1 LoRA runs at 12 steps, CFG 1.0, er_sde sampler, simple scheduler, and denoise 1.0. Without it, the base model needs 30-50 steps at CFG 4-5 for full quality.
Yes, ANIMA is a roughly 2B-parameter model that generates and trains on under 6GB VRAM, running anywhere SDXL or Illustrious runs. A GGUF quantized build also exists for even lower memory, though it requires a separate GGUF loader node.
The node comes from the ComfyUI-Anima-LLLite custom node pack, which must be installed separately. Do not use the core AnimaLLLiteApply node, since ComfyUI now owns that ID with a different input signature.
Load the masked image, encode it with VAEEncode, apply SetLatentNoiseMask, and patch the model with AnimaLLLiteApply_sdscripts using anima-lllite-inpainting-v1.safetensors fed the same image and mask. The masked region regenerates from your prompt while the rest is preserved.
ANIMA accepts Danbooru tags mixed with natural language: start with quality tags like masterpiece and score_7, add subject tags, an optional @artist name, style tags, then 2-4 descriptive sentences. Use a negative prompt with low-quality and anatomy tags since CFG above 1 is supported.
jackspaceGuide ComfyUI workflow creation, node configuration, and optimization.
ZurybrSegment cells in images using MobileSAM and SAM, then convert masks to polygons.
fabioc-alohaTrain custom FLUX LoRA fine-tunes on Replicate from curated image datasets.
NVlabsAutomate RL post-training for Alpamayo VLM action models with Cosmos-RL and GRPO.
