Trigger rebuild to pick up Space secrets

Uncomment spaces import to match original working Space

Remove version pins to match original working Space

Update accelerate and transformers for newer diffusers compatibility

Fix diffusers version for flux fill pipeline

Add torchvision and torchaudio to requirements

Fix: patch torch cpp_extension.py on disk before detectron2 pip install

Fix detectron2 install: use sitecustomize.py to bypass CUDA version check in pip subprocess

Minimal fix: patch CUDA version check for detectron2 build

Python 3.13: install cu124 PyTorch + CUDA runtime libs via pip

Try Python 3.10.13

Use original app.py + Python 3.12.12 + pinned deps

Fix: --no-deps + specific cu124 torchvision/torchaudio versions

Fix: --no-deps on PyTorch reinstall + pin exact diffusers/transformers versions

Re-add PyTorch CUDA 12.4 install + pin diffusers/transformers for Python 3.13

Remove .python-version - Python 3.10 not available in ZeroGPU image

Remove PyTorch force-reinstall - rely on Docker image PyTorch (Python 3.10)

Pin Python 3.10 to avoid 3.13 incompatibility

Switch to CUDA 12.4 (cu124) which has torchvision wheels for cp313

Fix: let pip resolve torchvision version automatically

Remove PyTorch from requirements.txt - handle in app.py instead

Fix requirements.txt: use --index-url for PyTorch CUDA 12.1

Fix: use --force-reinstall for PyTorch CUDA 12.1

Pin PyTorch 2.5.1 CUDA 12.1 to fix CUDA version mismatch

Revert to source-build detectron2 (GPU compatible on a10g)

Fix: CPU detectron2 wheels to bypass CUDA mismatch

Switch to zero-a10g to fix CUDA compatibility

Fix: packages.txt + RGB conversion in submit_function returns

Full codebase, fixed RGBA→RGB, no demo images, no LFS

Fix RGBA→RGB for API serialization

initial commit