from transformers import pipeline

# Public, stable image classifier (generic). You can swap to a deepfake-specific model later.
image_model = pipeline(
    task="image-classification",
    model="google/vit-base-patch16-224",
    device=-1  # CPU
)