OpenSDI-SDXL-SigLIP2

OpenSDI-SDXL-SigLIP2

OpenSDI-SDXL-SigLIP2 is a vision-language encoder model fine-tuned from google/siglip2-base-patch16-224 for binary image classification. It is trained to detect whether an image is a real photograph or generated using Stable Diffusion XL (SDXL), utilizing the SiglipForImageClassification architecture.

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

[!note] OpenSDI: Spotting Diffusion-Generated Images in the Open World https://arxiv.org/pdf/2503.19653, OpenSDI SDXL SigLIP2 works best with crisp and high-quality images. Noisy images are not recommended for validation.

[!warning] If the task is based on image content moderation or AI-generated image vs. real image classification, it is recommended to use the OpenSDI-Flux.1-SigLIP2 model.

Classification Report:
                precision    recall  f1-score   support

    Real_Image     0.8632    0.8757    0.8694     10000
SDXL_Generated     0.8739    0.8612    0.8675     10000

      accuracy                         0.8685     20000
     macro avg     0.8685    0.8684    0.8684     20000
  weighted avg     0.8685    0.8685    0.8684     20000

---

Label Space: 2 Classes

The model classifies an image as either:

Class 0: Real_Image  
Class 1: SDXL_Generated

---

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/OpenSDI-SDXL-SigLIP2"  # Replace with your model path
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

# Label mapping
id2label = {
    "0": "Real_Image",
    "1": "SDXL_Generated"
}

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="SDXL Image Detection"),
    title="OpenSDI-SDXL-SigLIP2",
    description="Upload an image to determine whether it is a real photograph or generated by Stable Diffusion XL (SDXL)."
)

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

---

Intended Use

OpenSDI-SDXL-SigLIP2 is intended for the following scenarios:

• Generative Content Detection – Accurately identify images generated using SDXL.
• Dataset Integrity – Screen datasets to ensure they contain only authentic photographic content.
• Trust and Safety – Flag AI-generated media in user-generated content pipelines.
• Digital Media Forensics – Support authenticity verification in investigative workflows.