Seeker — AI Image Detector

Upload an image to check if it's real or AI-generated, powered by a fine-tuned CLIP ViT-B/32 model.

Limitations: Trained on 32×32 CIFAKE images (upscaled to 224×224) — performance may vary on high-resolution inputs. AI-generated images in the training set are from Stable Diffusion v1.4 only; newer generators (DALL·E 3, Midjourney, Flux) may not be detected as reliably. GradCAM heatmaps are coarse (7×7 grid) due to ViT-B/32's patch size.