NSFW Scanner

The scanner loads a pre-trained MobileNetV2 neural network directly in your browser via TensorFlow.js. When you upload an image, the model resizes it internally and runs inference to produce five confidence scores: Neutral, Drawing, Sexy, Hentai, and Porn. Each score represents the probability that the image belongs to that category. The model achieves approximately 93% accuracy across all categories. Because everything runs client-side, your images are never sent to any server — the entire NSFW detection process happens on your device. For metadata-based image verification, see the Authenticity Checker.

Platform operators use NSFW detection to filter user-generated content before it reaches other users — profile photos, forum uploads, marketplace listings, and chat attachments. Parents and educators check downloaded images or screen content shared with children. HR teams verify that workplace-shared media complies with company policies. Developers integrate client-side content moderation into apps where sending images to a server would violate privacy regulations. Photographers and creators pre-screen portfolios before uploading to platforms with strict content policies. For AI-generated image detection, pair this with the AI Detector.

Neutral covers everyday photos — landscapes, objects, food, animals. Drawing identifies illustrations, cartoons, anime, and digital art that are safe for work. Sexy flags suggestive but not explicit content — revealing clothing, provocative poses, swimwear. Hentai detects explicit illustrated or animated content. Porn identifies explicit photographic content. The scanner outputs a probability for each category so you can set your own threshold — a children's platform might flag anything above 10% Sexy, while an art community might only flag Porn above 50%. For face detection in group photos, try the Face Counter.

No content classifier is perfect. The model may produce false positives on medical imagery, classical art, swimwear catalogs, or close-up skin tones. It may miss explicit content that is heavily cropped, rotated, or embedded in collages. The scanner classifies still images only — it does not analyze video frames, animated GIFs (only the first frame), or text-based content. Treat the results as an automated first pass, not a definitive ruling. For borderline cases, combine the NSFW score with manual review. To check whether an image has been digitally manipulated, use the ELA Scanner for pixel-level forensics.