ComfyUI Rating Checker

ComfyUI Rating Checker

[日本語]

This is a custom node for ComfyUI that classifies images as NSFW (Not Safe For Work).

Feature

This node is designed especially for classifying NSFW content in illustrated images.

While existing similar nodes perform well for real human photos, they tend to classify anime-style illustrations as NSFW even when there is minimal nudity. Additionally, distinguishing between R15 and R18 categories is often difficult.

To address these issues, Rating Checker (NudeNet) combines an object detection model (NudeNet) with an NSFW classification model to categorize images into three labels: SFW, NSFW (R15), and NSFW (R18).

Installation

ComfyUI Manager

Search for ComfyUI Rating Checker.

Manual

Clone the repository into your custom_nodes directory:

git clone https://github.com/tighug/comfyui-rating-checker.git

Usage

This package includes three nodes for NSFW rating. The primary node is the NudeNet version, but the others (created during evaluation) are also included.

Rating Checker (NudeNet)

Classifies images into the following three labels based on these conditions:

• nsfw_r18: At least one of the following body parts is detected with detect_[body part] = True:

  • armpits
  • female_breast
  • male_breast
  • female_genitalia
  • male_genitalia
  • belly
  • buttocks
  • anus
  • feet

• nsfw_r15: Not nsfw_r18, but nsfw_score > threshold_nsfw

• sfw: Does not meet any of the above conditions

Models used:

• notAI-tech/NudeNet
• GantMan / nsfw_model

Rating Checker (GantMan)

Classifies images into the following five labels:

• drawings: Illustrations
• hentai: Anime or manga-style content
• neutral: General-purpose images
• porn: Real-world explicit content
• sexy: Images with a sexual vibe

Useful for distinguishing between real images and illustrations and for general NSFW classification.

Model used:

• GantMan / nsfw_model

Rating Checker (Marqo)

Calculates an NSFW score from the image and outputs it as scores. It also performs binary classification (sfw / nsfw) using threshold_nsfw, and outputs the result as ratings.

Model used:

• Marqo/nsfw-image-detection-384

License

MIT