nicegen-ai/nsfw-filter-for-portrait

Portrait NSFW Classification & Safety Filter

Public
14 runs
Playground API Examples README Versions

Portrait NSFW Classification Model

Detects and classifies NSFW (Not Safe For Work) content in human portrait images.

This model identifies people in an image, analyzes the relevant regions, and assigns a risk level based on the presence of potentially inappropriate or explicit content.

🚀 Inputs

input type default description
image file required Input image.
expand float 0.2 Person bounding box expansion ratio to include more contextual information around detected individuals.
conf float 0.5 Confidence threshold for person detection using YOLO.

📤 Output

The model returns a risk_level indicating the NSFW risk classification of the input image.

Possible risk_level values:

risk_level Meaning
normal Normal or low-risk content.
suggestive Revealing clothing, swimwear (e.g., bikini), or sexually suggestive content without explicit nudity.
explicit Full nudity, visible sexual organs, or explicit sexual content.

Notes

  • The final risk_level is determined based on detected persons in the image
  • If multiple people are present, the highest risk level is returned
  • Results are probabilistic and may not be 100% accurate

🧠 How It Works

  1. Detects human subjects in the image using a YOLO-based model
  2. Expands bounding boxes using the expand parameter to capture additional context
  3. Crops detected regions and feeds them into an NSFW classification model
  4. Aggregates predictions across all detected persons
  5. Outputs the highest risk level as the final result

⚙️ Parameters Guide

  • expand

  • Higher values include more surrounding context (useful for borderline cases)

  • Too large values may introduce background noise

  • conf

  • Lower values detect more people but may increase false positives

  • Higher values are stricter but may miss small or partially visible persons

📌 Use Cases

  • Content moderation systems
  • Social media safety filtering
  • Image dataset cleaning
  • Automated compliance pipelines

⚠️ Limitations

  • May produce false positives or false negatives
  • Performance depends on quality of person detection
  • Not suitable for images without human subjects
  • Cultural/contextual sensitivity may vary

⚠️ Disclaimer

This model is intended as an assistive tool for content moderation. It should not be used as the sole basis for making critical or legal decisions.

Model created
Model updated