Switch to size='xlarge' GPU pool to route past faulty hardware ECC nodes

Fix ZeroGPU model loading in app.py via lazy-loading initialization

Fix example loading, file upload and ZeroGPU compatibility

redesign

minimal redesign

Update default model to nvidia/LocateAnything-3B and transformers version

Convert to gradio.Server with premium custom HTML/CSS/JS frontend

fix(log): write one JSONL file per record with session_id to prevent overwrite

ui: add resolution note next to header about magi-attention / 1K resize

fix private auth and zerogpu build

demo

initial commit