Skip to content

EthenEthenEthen

LoginPlatformTry Ethen

Open Source Model Profile · Ateeqq

ai-vs-human-image-detector

ai-vs-human-image-detector is a 92.9M-parameter Siglip classifier from Ateeqq. According to the model card, it was fine-tuned on 60,000 AI-generated and 60,000 human images.

Publisher
Ateeqq
Task
image-classification
Model type
siglip
License
apache-2.0
Library
transformers
Publication status
Accepted · not indexed

Model overview

ai-vs-human-image-detector is published by Ateeqq for image classification. The captured configuration identifies SiglipForImageClassification with model type siglip, and Safetensors metadata reports 92,885,762 parameters with Apache-2.0 licensing. According to the model card, it is a fine-tuned model trained on 60,000 AI-generated and 60,000 human images.

Recorded capabilities

120,000-image fine-tune

According to the model card, the model was fine-tuned on 60,000 AI-generated and 60,000 human images.

Publisher-reported 0.9923 accuracy

According to the model card, the fine-tuned model reports 0.9923 test accuracy with matching 0.9923 macro and weighted F1 scores at 5.0 epochs.

Captured 92.9M Siglip weights

Safetensors metadata reports 92,885,762 parameters, and the hub record lists Transformers support.

Publisher-noted overfitting reports

According to the model card, some users reported overfitting issues, so the headline metrics should be read with that caveat.

Documented Transformers inference path

According to the model card, inference runs through AutoImageProcessor with SiglipForImageClassification, including a worked device, preprocessing, and logit-to-label example.

Use cases in the source record

  • AI-versus-human image screening experiments using the publisher's documented Siglip inference workflow.
  • Accuracy-benchmark comparison work grounded in the publisher-reported 0.9923 test accuracy and F1 figures, read alongside the overfitting caveat.

Limitations and unknowns

  • Accuracy and F1 figures are publisher-reported test values and were not independently verified by Ethen.
  • According to the model card, some users reported overfitting issues.
  • No context-window value was extracted from this record.
  • Provider state is historical snapshot data and should be refreshed before being presented as current.

Source and provenance

Source: Ateeqq/ai-vs-human-image-detector

Captured: Unknown. Processed: 2026-09-07T19:35:03.057634+00:00.

AI and Human Image Classification Model - v1 A fine-tuned model trained on 60,000 AI-generated and 60,000 human images. Detailed Training Code is available here: blog/ai/fine-tuning-siglip2 Evaluation Metrics 🏋️‍♂️ Train Metrics Epoch: 5.0 Total FLOPs: 51,652,280,821 GF Train Loss: 0.0799 Train Runtime: 2:39:49.46 Train Samples/Sec: 69.053 Train Steps/Sec: 4.316 📊 Evaluation Metrics (Fine-Tuned Model on Test Set) Epoch: 5.0 Eval Accuracy: 0.9923 Eval Loss: 0.0551 Eval Runtime: 0:02:35.78 Eval Samples/Sec: 212.533 Eval Steps/Sec: 6.644 NOTE: Users report Some users reported overfitting issues 🔦 Prediction Metrics (on test set): {…

F001F002F003F004F005F006F007F008F010F011F012F013F014F015F016