ComfyUI Anima Suite

A dedicated custom node suite and ready-to-use workflow designed for Anima 3.8B generation in ComfyUI. It packages two-stage sampling, automatic LoRA block remapping, aspect ratio presets, and built-in offline Danbooru tag, character, and artist browsers into a compact canvas setup.

Everything runs 100% locally with zero external network requests.

Download size is ~4.6 GB because it bundles a complete offline library of 78,000+ local artist and character reference thumbnails.

Quick Setup

1. Copy or extract this folder into ComfyUI/custom_nodes/comfyui-anima-suite/.

2. Restart ComfyUI.

3. Load the bundled workflow: example_workflows/Anima_delta_mix.json.

4. Enter your prompt (or pick from the tag/artist drawers) and queue!

Required Model Files:

• Checkpoint: Anima model (models/checkpoints/)

• Text Encoder: Compatible text encoder (models/text_encoders/)

• VAE: Compatible image VAE (models/vae/)

• LoRAs (Optional): Any Anima-compatible LoRA (models/loras/)

Usage & Recommended Settings

Sampler: euler_ancestral Scheduler: simple

Dual-Pass Setup

• Pass 1 (Base): Steps: 10 – 15 | CFG: 4.0 – 5.0 | Denoise: 1.00

• Pass 2 (Hires Refine): Steps: 10 – 15 | CFG: 2.0 – 4.0 | Denoise: 0.20 – 0.60
  Tip: Lower denoise (0.20 – 0.30) keeps composition identical to Pass 1; higher denoise (0.45 – 0.60) adds micro-details and change some details.

Single-Pass Setup

• Steps: 25 – 35 | CFG: 3.0 – 5.0 | Denoise: 1.00 (disable Pass 2 via the run_pass_2 toggle on the sampler)

Key Features

• Anima Dual Sampler: Unified two-stage KSampler with shared seed coordination, pass toggles (run_pass_1 / run_pass_2), and an option to refine with or without LoRAs applied.

• Anima LoRA Stack: 4-slot stack featuring automatic block remapping across 28, 40, and 52-block Anima architectures to match your active checkpoint.

• Anima Prompt Hub & Resolution: Single-pipe pass-through wiring, collapsible negative drawer, and aspect ratio presets (1:1, 4:5, 3:4, 2:3, 9:16, etc.) with live dimension calculation.

• Built-in Tag & Artist Drawers: Searchable offline browser with ~36,000 characters and ~42,000 artists (with thumbnails), plus a Danbooru tag helper.

• Prompt History & Image Inspector: Drop any generated image onto the canvas to read embedded metadata back into nodes, with automatic prompt version saving.

Tested system:

GPU: NVIDIA GeForce RTX 5070 (12 GB VRAM) --- RAM: 32 GB

Generation Speed (2:3 aspect ratio / 832×1216 base):

Dual-Pass Full High-Res (1.5× upscale): ~35 seconds total

Pass 1 (11 steps): ~8 seconds --- Pass 2 (15 steps): ~25 seconds

Drop a review or showcase your generations below, and let me know in the comments if you run into any issues or want new features added!