Guard _install_cuda_packages() startup call against quota exhaustion

Reduce startup GPU quota burn: skip CUDA pkg compile if already importable

Remove 'spaces' from requirements: pre-installed by HF ZeroGPU runtime

Remove open3d: conflicts with numpy>=2 (breaks Docker build); TexturePipeline bypassed

Bump cache-bust to force fresh Docker build (fix stale BUILD_ERROR)

Remove xatlas from requirements: not needed (fallback now just copies raw mesh)

Skip CPU texture bake fallback: raw mesh passes to rigging instead

Increase apply_texture GPU duration to 600s; use 512px UV in CPU fallback

Add CPU texture-bake fallback: xatlas UV-unwrap + numpy projection

Fix nvdiffrast: use CameraProjection context_type='cuda' (GL context fails on ZeroGPU)

Bypass TexturePipeline: use CameraProjection directly for UV baking

Add open3d to requirements: needed by mvadapter pipeline_texture (Python 3.10 has wheels)

Replace bake_texture with TexturePipeline (mvadapter API change)

Remove hyperswap_1a_256.onnx from _ensure_ckpts: file not publicly hosted

Fix get_plucker_embeds_from_cameras_ortho: width is positional not keyword

Update apply_texture to MV-Adapter new API

Remove unused get_ipadapter_image import (removed from mvadapter.utils)

Patch autoencoder_kl_triposg: fix embedder fp32 upcast in _decode

Fix dtype mismatch in flash_extract_geometry: force fp16 for VAE decode

Fix TripoSG NaN cascade and remove stale autocast wrapper

Skip DiffDMC (CUDA arch mismatch on A10G), use marching cubes with adaptive iso-level and bg_fill

Fix NaN handling in flash_extract_geometry: preserve NaN for DiffDMC, use -1.0 for MC fallback

Pre-convert grid_logits to CPU before DiffDMC attempt (prevents CUDA ctx corruption in fallback)

Add explicit error logging + last-resort MC fallback in flash_extract_geometry

Patch flash_extract_geometry: marching cubes fallback when DiffDMC unavailable (fixes NoneType on mesh output)

Patch TripoSG flash_extract_geometry: guard empty next_points (IndexError on min())

Patch PSHuman attn: use kwargs in Attention.__init__ (diffusers added kv_heads param, breaking positional args)

Patch PSHuman unet: _load_state_dict_into_model moved to model_loading_utils in diffusers>=0.28

Fix PSHuman CLIPFeatureExtractor removed in transformers>=5.0

Add api_name to run_full_pipeline and pshuman_face for Gradio API access

Replace runtime TripoSG string-patching with pre-patched files in patches/triposg/

Fix pymeshlab patch: always write file after import removal

Integrate PSHuman locally — multi-view diffusion in-process, no remote calls

Replace pymeshlab with trimesh in inference_triposg.py (no py3.13 wheels)

Move CUDA compilation into @spaces.GPU function — nvcc unavailable at APP_STARTING

Fix exec()+locals() for all 4 xinntao packages (basicsr/realesrgan/gfpgan/facexlib)

Detect CUDA home dynamically instead of hardcoded /cuda-image/... path

Fix basicsr setup.py patch: keep _ns inside with-block so f is still open

Pre-install meson-python+ninja before --no-build-isolation step

Fix basicsr Python 3.13 exec()+locals() crash at runtime startup

Fix basicsr/realesrgan/gfpgan build: move to runtime --no-build-isolation install (setup.py imports torch)

requirements: remove open3d/pymeshlab/pyrender (no py3.13 wheels), drop unused motion training deps

Fix packages.txt: replace libgl1-mesa-glx with libgl1 (removed in Debian Bookworm)

Port MeshForge features to ZeroGPU Space: FireRed, PSHuman, Motion Search

Patch all-zero alpha: fallback to full-foreground instead of raising ValueError

Fix empty_contours_guard patch: use gray_image (correct param name) not alpha

Patch find_bounding_box: guard empty contours (blank alpha fallback to full image bbox)

Fix queries dtype cast in inference_utils.py; remove autocast (caused NaN/no isosurface)

Wrap run_triposg in torch.autocast float16 to fix Float/Half dtype mismatch

Fix RMBG loading with torch.device(cpu) context; fix None fallback alpha shape [1,1,H,W]