fix(near): defer renderer/tone-mapper init to first GPU ensure

- app: unify ensure_near/geometry with CPU preload helpers; fix generate_mesh; restore preload in _near_launch
- pipeline: comment out from_pretrained renderer+tone_mapper setup (initialized in ensure_near_on_cuda)

Made-with: Cursor

app.py

@@ -14,7 +14,6 @@ import os
 import shutil
 import sys
 import threading
-import time
 from pathlib import Path
 from typing import Any, Dict, Optional
 
@@ -143,23 +142,30 @@ def run_model_cpu_preload_blocking() -> None:
 
 
 def ensure_near_on_cuda() -> None:
+    """Load NeAR if needed (respects CPU preload), move to CUDA once, init renderer / tone mapper."""
     global _NEAR_ON_CUDA
     with _MODEL_LOCK:
         _ensure_near_loaded_on_cpu_locked()
+        assert PIPELINE is not None
         if torch.cuda.is_available() and not _NEAR_ON_CUDA:
-            assert PIPELINE is not None
             PIPELINE.to("cuda")
             _NEAR_ON_CUDA = True
+        if torch.cuda.is_available():
+            if PIPELINE.renderer is None:
+                PIPELINE.setup_renderer()
+            if PIPELINE.tone_mapper is None:
+                PIPELINE.setup_tone_mapper("AgX")
 
 
 def ensure_geometry_on_cuda() -> None:
     global _GEOMETRY_ON_CUDA
     with _MODEL_LOCK:
         _ensure_geometry_loaded_on_cpu_locked()
+        assert GEOMETRY_PIPELINE is not None
         if torch.cuda.is_available() and not _GEOMETRY_ON_CUDA:
-            assert GEOMETRY_PIPELINE is not None
             GEOMETRY_PIPELINE.to("cuda")
             _GEOMETRY_ON_CUDA = True
+            print("[NeAR] Hunyuan geometry pipeline on CUDA.", flush=True)
 
 
 def _try_release_cuda_memory() -> None:

trellis/pipelines/near_image_to_relightable_3d.py

@@ -116,12 +116,12 @@ class NeARImageToRelightable3DPipeline(Pipeline):
         t_stage = time.time()
         new_pipeline.hdri_processor = HDRI_Preprocessor(envmap_h=512, envmap_w=1024)
         print(f"[NeAR] timing near.from_pretrained.hdri_processor: {time.time() - t_stage:.1f}s", flush=True)
-        t_stage = time.time()
-        new_pipeline.setup_renderer()
-        print(f"[NeAR] timing near.from_pretrained.renderer: {time.time() - t_stage:.1f}s", flush=True)
-        t_stage = time.time()
-        new_pipeline.setup_tone_mapper("AgX")
-        print(f"[NeAR] timing near.from_pretrained.tone_mapper: {time.time() - t_stage:.1f}s", flush=True)
+        # t_stage = time.time()
+        # new_pipeline.setup_renderer()
+        # print(f"[NeAR] timing near.from_pretrained.renderer: {time.time() - t_stage:.1f}s", flush=True)
+        # t_stage = time.time()
+        # new_pipeline.setup_tone_mapper("AgX")
+        # print(f"[NeAR] timing near.from_pretrained.tone_mapper: {time.time() - t_stage:.1f}s", flush=True)
         print(f"[NeAR] timing near.from_pretrained.total: {time.time() - t0:.1f}s", flush=True)
         return new_pipeline