Anima workflows
One page, one version per workflow. Same clean design everywhere: numbered groups, group toggles, readable node titles.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
V3.2 # all detailer (main, most complete)
FACE + NSFW + EYES + HANDS detailers, RTX 2x, optional USDU 2x.
~80-140s on RTX 4070 8GB.
★ Kirazuri V4 × USD
For the Kirazuri Anima v4.0 model. FACE + EYES detailers, 4x JaNai upscale, plus an HD refine pass (Ultimate SD Upscale 2x).
Two outputs per run: fast upscale and HD.
V3.2 × SeedVR
Lighter: FACE detailer + SeedVR2 VAE pass + color match.
Good starting point if you want something simple.
Older versions (v2, v1…) are kept for reference.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
GOOD TO KNOW
━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- All versions are built and tested on an RTX 4070
- Turn whole steps on/off with the Group toggles node (no rewiring needed).
- Optional / heavy steps are bypassed by default and marked in the node titles.
- Each version lists its models and custom nodes in its own notes.
- Common custom nodes: Impact Pack, Impact Subpack, rgthree-comfy, ComfyUI_UltimateSDUpscale.
Looking for Krea2? →
Description
If you don't like it, please move along.
Flow: generate -> FACE detailer -> EYES detailer -> 4x upscale (JaNai) -> save
-> HD refine (USDU 2x) -> save HD
Two outputs per run: the fast 4x upscale and the HD refine. Turn off the one you don't need with the Group toggles.
Native generation: the model was trained up to 1536², no hi-res pass needed.
Settings: er_sde / simple, 30 steps, CFG 4, 1872x1248 (3:2)
Face detailer: denoise 0.3, 25 steps, euler_a
Eyes detailer: denoise 0.35, 20 steps, euler + eyes wildcard prompt
HD refine (USDU): 2x, denoise 0.25, 20 steps, tile 1024 (lower to 768 if OOM)
Tips:
- Start the prompt with: masterpiece, best quality, very aesthetic, absurdres
- Then danbooru tags (+ optional natural-language sentence). Artists: @name
- Escape parentheses: \( \)
- Keep the negative short
- Optional style LoRA: Reverse 1999 (bypass the node to turn it off)
Models:
- kirazuriAnima_v40AnimaBase1_full_bf16.safetensors (UNET): civitai.com/models/2495369
- qwen_3_06b_base.safetensors (CLIP)
- qwen_image_vae.safetensors (VAE)
- 4x_IllustrationJaNai_V1_ESRGAN_135k.pth (upscaler)
- sam_vit_b_01ec64.pth (SAM)
- face_yolov8m.pt, Eyeful_v2-Paired.pt (bbox detectors)
- reverse1999style_anima_10.safetensors (optional LoRA)
Custom nodes:
- Impact Pack: https://github.com/ltdrdata/ComfyUI-Impact-Pack
- Impact Subpack: https://github.com/ltdrdata/ComfyUI-Impact-Subpack
- rgthree-comfy: https://github.com/rgthree/rgthree-comfy
- ComfyUI_UltimateSDUpscale: https://github.com/ssitu/ComfyUI_UltimateSDUpscale
- Resolution Master: https://github.com/Azornes/Comfyui-Resolution-Master