feat: dual-model ensemble for better accuracy on both real and fake

fix: reduce header padding and title size to prevent text clipping

chore: bust Docker cache to rebuild frontend with logo + fixes

feat: add DeepGuard AI logo, favicon, fix image cropping

feat: add file size validation with user-facing error messages

fix: switch to CommunityForensics-DeepfakeDet-ViT (668K downloads)

fix: wrap Grad-CAM in try/except to prevent 500 errors on Swin model

fix: switch to umm-maybe/AI-image-detector (Swin) for better fake detection

Fix: use pipeline scoring with dima806 deepfake model for correct label mapping

fix: switch to dima806 deepfake model, use full image for inference

fix: use Node 20 in Dockerfile for tailwindcss/oxide compatibility

fix: add HF Spaces config, root Dockerfile, and SPA serving

Switch to pretrained ViT deepfake detection model

feat: aggressive GAN detection — FFT sensitivity, skin texture, calibration bias

fix: HF Spaces YAML color, relative API URL for production

feat: enhanced frequency analysis, face landmark consistency, adaptive ensemble

chore: switch HF Spaces config to CPU Basic (free tier)

feat: full-stack deepfake detector with EfficientNet-B4 + FFT ensemble