Real-versus-fake image check
The model card says the model checks whether an image is real or fake (AI-generated) as an image-classification task.
Open Source Model Profile · dima806
deepfake_vs_real_image_detection
deepfake_vs_real_image_detection is an 85.8M-parameter ViT classifier from dima806 that checks whether an image is real or AI-generated.
- Publisher
- dima806
- Task
- image-classification
- Model type
- vit
- License
- apache-2.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
deepfake_vs_real_image_detection is published by dima806 as a ViT image-classification model. The captured configuration identifies ViTForImageClassification and Safetensors metadata reports 85,800,194 parameters. The model card says it checks whether an image is real or AI-generated, and card data records apache-2.0.
Recorded capabilities
ViT fine-tune footprint
The captured configuration identifies ViTForImageClassification with about 85.8M parameters, and hub tags record a google/vit-base-patch16-224-in21k base.
Publisher drift warning
The publisher warns of significant concept drift since the training data was collected and urges retraining on current labeled data.
Use cases in the source record
- Screening images as real or AI-generated in research workflows that account for the publisher's age-of-data warning.
- Retraining and evaluation work using newer labeled data, which the publisher explicitly recommends before production use.
Limitations and unknowns
- No evaluation results were extracted from this record.
- According to the model card, production use is discouraged without retraining because the training data is several years old.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
Source and provenance
Source: dima806/deepfake_vs_real_image_detection
Captured: Unknown. Processed: 2026-09-07T19:34:43.560838+00:00.
Checks whether an image is real or fake (AI-generated). Note to users who want to use this model in production Beware that this model is trained on a dataset collected about 3 years ago. Since then, there is a remarkable progress in generating deepfake images with common AI tools, resulting in a significant concept drift. To mitigate that, I urge you to retrain the model using the latest available labeled data. As a quick-fix approach, simple reducing the threshold (say from default 0.5 to 0.1 or even 0.01) of labelling image as a fake may suffice. However, you will do that at your own risk, and retraining the model is the better wa…
F001F002F003F004F005F006F007F009F010F011F012F013F014