hotfix: force zeroGPU duration clamp to 45s to fit inside remaining 73s quota

hotfix: force zeroGPU duration clamp to 120s to bypass 270s rejection bug

hotfix: force zeroGPU duration clamp to 180s to bypass 279s rejection bug

Fix comfyui import error by correctly detecting Hugging Face Spaces on non-ZeroGPU hardware (T4)

Fix 50G storage limit by forcing lowvram and removing uncompressed models from preload

Pivot architecture to LTX-2.3 Studio (All-In-One)

Switch to ghost_3_256 model to improve source identity likeness

Switch to restoreformer++ and reduce blend to preserve identity

Fix pixel boost resolution mismatch for 512x512 model

Upgrade to Dedicated GPU: high fidelity 512x512 model and video api endpoint

Temporarily fallback to CPU to test ZeroGPU CUDA incompatibility

Increase ZeroGPU duration to 120s

Pre-fetch models before launching Gradio

Fix syntax and add traceback

Add traceback to process_swap

Use onnxruntime-gpu for CUDA support

Replace FaceFusion UI with Headless ZeroGPU API wrapper

Set Python version to 3.12 for onnxruntime 1.24

Fix gradio version conflict

FaceFusion Initial Push with fixed metadata