Optimize ZeroGPU duration to 25s

Set duration=50s to fit within remaining free ZeroGPU quota

Set ZeroGPU duration to 120s for initial 12B model VRAM load

Add Gemma4UnifiedProcessor dependencies: soundfile, opencv, torchvision, sentencepiece

Pre-cache model weights to disk during startup to prevent ZeroGPU timeout

Support direct base64 image strings and detailed error handling

Upgrade to Gradio 6 and filepath Image

Bind Gradio to 0.0.0.0:7860

Fix: use Textbox with json.dumps to bypass Gradio 5 JSON schema bug

Fix: ensure spaces is imported before any other CUDA packages

Fix: upgrade to Gradio 5 for modern huggingface_hub compatibility

Deploy OmniSnap engine v0

initial commit