xiaden-comfyui-imgutils/README.md

ImgUtils Nodes for ComfyUI

ComfyUI custom nodes wrapping deepghs/imgutils — a comprehensive toolbox for anime image analysis, tagging, detection, pose estimation, and segmentation.

All nodes use the ComfyUI V2 io.ComfyNode API.

Nodes (26)

🏷 Tagging (6)

Node	Category	Knobs
Imgutils WD14 Tagger	imgutils/tagging	general_threshold, character_threshold, drop_overlap
Imgutils DeepDanbooru Tagger	imgutils/tagging	general_threshold, character_threshold, drop_overlap
Imgutils DeepGelbooru Tagger	imgutils/tagging	general_threshold, character_threshold, drop_overlap
Imgutils MLDanbooru Tagger	imgutils/tagging	threshold, drop_overlap
Imgutils Camie Tagger	imgutils/tagging	mode (balanced/high_precision/…), drop_overlap
Imgutils PixAI Tagger	imgutils/tagging	threshold
Imgutils Tags	imgutils/tagging	Tag string processing — filter, sort, format

🔎 Judge (3)

Node	Category	Knobs
Imgutils Classify	imgutils/judge	11 classification ops — safety, NSFW, rating, style, character
Imgutils Check	imgutils/judge	5 boolean checks — AI-created, monochrome, greyscale, truncation, completeness
Imgutils Metric	imgutils/judge	2 numeric metrics — sharpness (Laplacian), monochrome score

📝 Describe (1)

Node	Category	Knobs
Imgutils OCR	imgutils/describe	heat_threshold, box_threshold — PaddleOCR text extraction

🔍 Detect (1)

Node	Category	Knobs
Imgutils Detect	imgutils/detect	confidence — faces, hands, heads, eyes, persons, half-bodies, text regions, nudenet, censors, Booru YOLO

✏️ Edge (3)

Node	Category	Knobs
Imgutils Edge (Canny)	imgutils/edge	low_threshold, high_threshold
Imgutils Edge (Lineart)	imgutils/edge	coarse, detect_resolution
Imgutils Edge (Lineart Anime)	imgutils/edge	detect_resolution

🛠 Restore (3)

Node	Category	Knobs
Imgutils Restore (SCUNet)	imgutils/restore	model (GAN/PSNR), tile_size
Imgutils Restore (NAFNet)	imgutils/restore	model (REDS/GoPro/SIDD), tile_size
Imgutils Restore (Adversarial)	imgutils/restore	Denoise adversarial noise — no ML model needed

🎨 Transform (4)

Node	Category	Knobs
Imgutils Upscale (CDC)	imgutils/transform	tile_size — high quality anime upscaling
Imgutils Censor	imgutils/transform	method (pixelate/blur/color), body-part toggles, conf_threshold
Imgutils Align	imgutils/transform	max_size — resize longest side, preserve aspect ratio
Imgutils Squeeze	imgutils/transform	threshold, median_filter — auto-crop to visible content

📊 Compare (2)

Node	Category	Knobs
Imgutils Compare (CCIP)	imgutils/compare	Character identity similarity + same-character check
Imgutils Compare (LPIPS)	imgutils/compare	Perceptual similarity distance

🦴 Pose (1)

Node	Category	Knobs
Imgutils Pose	imgutils/pose	visualization preset, auto_detect — DWPose keypoints

✂️ Segment (1)

Node	Category	Knobs
Imgutils Segment	imgutils/segment	scale — isnetis character segmentation (mask + RGBA)

Installation

cd ComfyUI/custom_nodes/
git clone https://github.com/xiaden/comfyui-imgutils.git
cd comfyui-imgutils
pip install -r requirements.txt

Or install via ComfyUI-Manager (once published to the Registry).

Requirements

• Python >= 3.10
• dghs-imgutils >= 0.19.0 — the underlying analysis library (installed via requirements.txt)
• onnxruntime — for ONNX model inference. ComfyUI already ships both onnxruntime (CPU) and onnxruntime-gpu (CUDA) in its venv. The dghs-imgutils[gpu] extra also pulls onnxruntime-gpu on install, so there is nothing to manually install or replace.

Models are downloaded automatically from HuggingFace Hub on first use and cached to ~/.cache/huggingface/hub/. No manual model downloads needed.

Structure

imgutils_nodes/
├── __init__.py           # NODE_CLASS_MAPPINGS + ComfyExtension entrypoint
├── node_*.py             # Root nodes (OCR, Detect, Pose, Segment, Tags, CDC, Censor, Align, Squeeze)
├── tagging/              # 6 tagger nodes — one per model
├── edge/                 # 3 edge detection nodes
├── restore/              # 3 restoration/denoising nodes
├── compare/              # 2 image comparison nodes
├── judge/                # 3 image analysis nodes
└── utils.py              # PIL ↔ ComfyUI tensor conversion

License

MIT — see LICENSE.

deepghs/imgutils is also MIT-licensed.