Binary Classification for Explicit Content Detection (98% Accuracy)
Analyzes images and classifies them as safe ("normal") or explicit ("adult"). Uses Vision Transformer base model trained on 80,000 labeled images. Fast inference for content moderation platforms.
What it does
NSFW (Not Safe For Work) Image Detection analyzes images and predicts whether they contain adult/explicit content. Binary classification model that outputs either "normal" (safe for work) or "adult" (not safe for work) with confidence scores. Works on any image format and resolution.
Problem it solves
- Automated content filtering for user-generated upload platforms
- Screen images before publication or distribution
- First-pass safety filtering (with human review for edge cases)
- Bulk content safety checks and batch moderation
- Reduce manual moderation workload
Input/Output
- Input: Any RGB image
- Output: Classification ("normal" or "adult")
Accuracy & Speed
- Performance:
- Accuracy: 98% on test dataset
- Model: Binary classification (normal vs. adult)
- Training data: 80,000 proprietary labeled images
Technical Details
- Architecture:
- Base: Vision Transformer (ViT-base)
- Pre-training: ImageNet-21k supervised learning
- Model checkpoint: google/vit-base-patch16-224-in21k
- Parameters: ~85.8M
- Head: Binary classification layer
- Training:
- Dataset: 80,000 proprietary images (binary labeled)
- Labels: "normal" (safe) or "adult" (explicit)
- Pre-training: ImageNet-21k
- Fine-tuning: Binary classification head
Compliance & Provenance
Provider | Open-source (HF) |
Provider type | Specialized |
License | Apache 2.0 |
EU AI Act risk class | Limited Risk |
Art. 50 transparency | Not applicable |
Region availability | Available globally |
Training data summary | Pending — provider has not yet published per Art. 53(d) |
For more on how we classify models and mark outputs, see our AI Policy.
Model Source
- HuggingFace: https://huggingface.co/Falconsai/nsfw_image_detection
- Paper: https://arxiv.org/pdf/2010.11929
- License: apache-2.0