README.md

metadata

license: apache-2.0
datasets:
  - prithivMLmods/OpenDeepfake-Preview
language:
  - en
base_model:
  - google/siglip2-base-patch16-512
pipeline_tag: image-classification
library_name: transformers
tags:
  - deepfake
  - detection
  - SigLIP2
  - art
  - synthetic

open-deepfake-detection

open-deepfake-detection is a vision-language encoder model fine-tuned from siglip2-base-patch16-512 for binary image classification. It is trained to detect whether an image is fake or real using the OpenDeepfake-Preview dataset. The model uses the SiglipForImageClassification architecture.

SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features https://arxiv.org/pdf/2502.14786

Experimental Model

Classification Report:
              precision    recall  f1-score   support

        Fake     0.9718    0.9155    0.9428     10000
        Real     0.9201    0.9734    0.9460      9999

    accuracy                         0.9444     19999
   macro avg     0.9459    0.9444    0.9444     19999
weighted avg     0.9459    0.9444    0.9444     19999

---

Label Space: 2 Classes

The model classifies an image as either:

Class 0: Fake  
Class 1: Real

---

Install Dependencies

pip install -q transformers torch pillow gradio hf_xet

---

Inference Code

import gradio as gr
from transformers import AutoImageProcessor, SiglipForImageClassification
from PIL import Image
import torch

# Load model and processor
model_name = "prithivMLmods/open-deepfake-detection"  # Updated model name
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

# Updated label mapping
id2label = {
    "0": "Fake",
    "1": "Real"
}

def classify_image(image):
    image = Image.fromarray(image).convert("RGB")
    inputs = processor(images=image, return_tensors="pt")

    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist()

    prediction = {
        id2label[str(i)]: round(probs[i], 3) for i in range(len(probs))
    }

    return prediction

# Gradio Interface
iface = gr.Interface(
    fn=classify_image,
    inputs=gr.Image(type="numpy"),
    outputs=gr.Label(num_top_classes=2, label="Deepfake Detection"),
    title="open-deepfake-detection",
    description="Upload an image to detect whether it is AI-generated (Fake) or a real photograph (Real), using the OpenDeepfake-Preview dataset."
)

if __name__ == "__main__":
    iface.launch()

---

Demo Inference

real

fake

Intended Use

open-deepfake-detection is designed for:

• Deepfake Detection – Identify AI-generated or manipulated images.
• Content Moderation – Flag synthetic or fake visual content.
• Dataset Curation – Remove synthetic samples from mixed datasets.
• Visual Authenticity Verification – Check the integrity of visual media.
• Digital Forensics – Support image source verification and traceability.