Unload ComfyUI's GPU model cache at start of each @spaces.GPU call (stale pointers after ZeroGPU release)

Pin torch==2.8.0/torchvision==0.23.0/sympy==1.13.3 (sm_120 kernels; pip was downgrading)

Pin starlette<0.38 / fastapi<0.112 (same gradio 4.x + jinja2 LRU dict-key bug as Helper)

Use demo.block_thread() instead of sleep loop

Use demo.launch (triggers ZeroGPU detection); attach custom /magic_quill routes to demo.app post-launch

Move @spaces.GPU to generate_image_handler (the actual Gradio-bound handler)

Lazy-init Client to MagicQuillHelper (avoid ReadTimeout at module load when Helper is cold)

Unpin bitsandbytes (0.43.3 imports triton.ops, removed in triton 3.x)

Point client at multimodalart/MagicQuillHelper for end-to-end test (revert before upstream PR)

Adapt to ZeroGPU: @spaces.GPU on generate; drop triton 2.1 pin

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update MagicQuill/brushnet_nodes.py

Update MagicQuill/scribble_color_edit.py

Update app.py

Update MagicQuill/comfy/cli_args.py

Update MagicQuill/brushnet/powerpaint_utils.py

Update MagicQuill/brushnet_nodes.py

Update MagicQuill/brushnet_nodes.py

Update MagicQuill/brushnet/powerpaint_utils.py

Update MagicQuill/brushnet_nodes.py

Update app.py

Update app.py

Update app.py

Update README.md

Update app.py

Update app.py

Update app.py

Update app.py

update download button

Merge branch 'main' of https://huggingface.co/spaces/AI4Editing/MagicQuill

Update app.py

Fix queueing bug

Update app.py

Fix Queueing Bug

Update README.md

Update app.py

Update emoji

Update app.py

update light theme

update parameters

update link

update notes

remove chinese reminder

remove chinese reminder