Model
Blue-Eye is an image classifier for content moderation. Given an image, it predicts whether the content is safe, suggestive or explicit, together with a probability for each class. On a benchmark of 3,000 real photographs it reaches 88.9% accuracy, ahead of AWS Rekognition, Gemini 3.1 Pro, Google Cloud Vision and the open-source NSFW detectors it was compared with. It handles both photographs and anime/illustration.
Blue-Eye is a DINOv3 ViT-L/16 vision transformer fine-tuned end to end for three-class content classification. The model takes a 512x512 RGB image and returns three class probabilities.
safe (0), suggestive (1), explicit (2)Blue-Eye is built for moderating sexual content in images:
The class definitions follow a nudity and sexual-content rubric:
The default prediction is the highest-probability class. Platforms with a stricter or looser policy can
set their own threshold on p(explicit) or p(suggestive) + p(explicit), tuned on their own data. The
model can also be fine-tuned further on a platform's own labels.
Blue-Eye covers sexual content only; violence, gore and other policy areas are out of scope. It does not estimate age and is not a tool for detecting child sexual abuse material, for which dedicated hash-matching services should be used. It should not be used to monitor or make decisions about individual people.
The boundary between suggestive and its neighbouring classes is the most subjective part of the task, for people and models alike, and it is where most disagreements occur. Performance across demographic groups was not part of this evaluation.
Validate the model on images representative of your own platform before deployment, and keep a human review step for actions that affect user accounts.
pip install torch "transformers>=4.56" safetensors pillow numpy huggingface_hub
import sys
from huggingface_hub import snapshot_download
path = snapshot_download("pp1618/Blue-Eye")
sys.path.insert(0, path)
from inference import classify
for result in classify(["photo.jpg", "drawing.png"], model=path):
print(result["label"], result["probabilities"])
From the command line:
python inference.py photo.jpg folder_of_images/ --model pp1618/Blue-Eye --device cuda
On GPUs with bfloat16 support, add --precision bf16 (or precision="bf16" in Python) for faster
inference with practically identical predictions.
About 2.6 million web images covering real photographs and anime/illustration, labelled into the three classes using a commercial content-moderation service, source content ratings and model-assisted relabelling. Evaluation images were removed from the training data.
Training ran in three progressive fine-tuning stages starting from the DINOv3 ViT-L/16 checkpoint:
Training regime: 2 epochs per stage at 512x512, AdamW with a one-cycle schedule, label smoothing 0.05, random resized crops, horizontal flips and colour jitter, float32 master weights with bfloat16 mixed precision.
The Blue-Eye benchmark contains 3,000 real photographs (1,480 safe, 513 suggestive, 1,007 explicit),
including 525 non-sexual, skin-heavy images such as swimwear, fitness and medical photos. Every system
below was evaluated on the same images with the same three-class labels; commercial services were
queried in August 2026 and their outputs mapped to the three classes. Google Cloud Vision counts an
image as explicit when adult is VERY_LIKELY and as suggestive when racy is VERY_LIKELY; AWS
Rekognition uses its default 50% confidence. The metric is three-class accuracy.
| System | Type | Accuracy |
|---|---|---|
| Blue-Eye | open weights | 88.9% |
| AWS Rekognition | commercial API | 87.3% |
| Gemini 3.1 Pro | commercial model | 86.4% |
| Gemini 3.7 Flash | commercial model | 84.1% |
| Google Cloud Vision SafeSearch | commercial API | 83.5% |
| TostAI/nsfw-image-detection-large | open weights | 79.2% |
| Marqo/nsfw-image-detection-384 | open weights | 73.8% |
| Falconsai/nsfw_image_detection | open weights | 70.9% |
| NudeNet | open weights | 70.4% |
| Freepik/nsfw_image_detector | open weights | 67.6% |
| AdamCodd/vit-base-nsfw-detector | open weights | 62.1% |
Many open-source detectors are binary, so they were also compared on the two binary tasks:
| System | Safe vs not safe | Explicit vs rest |
|---|---|---|
| Blue-Eye | 92.2% | 95.1% |
| Marqo/nsfw-image-detection-384 | 86.4% | 78.3% |
| TostAI/nsfw-image-detection-large | 85.7% | 90.1% |
| Falconsai/nsfw_image_detection | 83.3% | 75.6% |
| Freepik/nsfw_image_detector | 82.6% | 71.8% |
| AdamCodd/vit-base-nsfw-detector | 78.6% | 62.7% |
| NudeNet | 76.9% | 87.5% |
Blue-Eye results across domains:
| Evaluation set | Images | Accuracy |
|---|---|---|
| Real photographs | 4,000 | 91.4% |
| Anime / illustration | 2,000 | 87.2% |
| All | 6,000 | 90.0% |
Per-class recall on the 3,000 real photographs: safe 89.5%, suggestive 75.4%, explicit 94.8%.
Blue-Eye is a derivative of DINOv3 and is released under the
DINOv3 License. The full text is in
LICENSE. Commercial use is permitted under its terms.
BibTeX
@misc{patel2026blueeye,
title = {Blue-Eye: a content-safety image classifier},
author = {Patel, Pranshu},
year = {2026},
url = {https://huggingface.co/pp1618/Blue-Eye}
}