fix: pasar HF_TOKEN como auth header al llamar OOTDiffusion (cuota ZeroGPU)

fix: parse gr.Image output from OOTDiffusion (filepath directo, no gallery)

fix: point tryon to our own OOTDiffusion fork (eduardo4547/OOTDiffusion)

revert: use OOTDiffusion Space for tryon, keep Zero123++ multiview improvements

feat: implement CatVTON directly for proper virtual try-on

fix: correct Zero123++ grid crop coordinates for human views

feat: load tryon model directly using SD Inpainting + IP-Adapter

fix: switch tryon from OOTDiffusion to Kolors-Virtual-Try-On

debug: enable show_error=True to expose Space errors

fix: revert to Zero123++ with proper preprocessing for human multiview

feat: replace Zero123++ with Wonder3D for human multiview generation

fix: switch try-on from IDM-VTON to OOTDiffusion (simpler API, more reliable)

fix: add trust_remote_code=True for Zero123++ custom pipeline

fix: use gradio_client to call IDM-VTON Space instead of loading model directly

fix: remove fp16 variant from IDM-VTON

add IDM-VTON + Zero123++ space

initial commit