Add FlowMatchEuler/Heun schedulers to all swappable model dropdowns

Default 512 resolution for -32 model variants (JiT, pMF)

Fix iMF/pMF inference by filtering unsupported pipeline kwargs

Fix scheduler dropdown: include checkpoint in choices before load

Default scheduler to checkpoint config instead of forcing DDIM

Add built-in diffusers scheduler selection in inference config

Fix _gpu_duration arg order to match generate handler

Resolve class labels via id2label comma-separated synonyms

Fix inference type coercion for Gradio float inputs (ADM)

Fix model_loader.py circular import (restore correct file)

Fix app.py

Fix ZeroGPU: keep model load on CPU, move to CUDA only inside @spaces.GPU.

Fix Hub loading: download variant subfolder instead of invalid nested repo id.

Fix ZeroGPU startup by importing spaces before torch.

Add class-conditional BiliSakura diffusers playground UI.

initial commit