Please help, I am loosing my mind RuntimeError: Ninja is required to load C++ extensions

[gerroon]

Hi

I have been trying to get this working for like 24 hours, I just lost it. I need help desperately. How do I pass this error? The pac seems to install fine but next run I have these issues. Ninja installed under the python env. 5090 seems to require c128 I have already lost two nights of sleep :(

[INFO] Python Version: 3.12.3
[INFO] Platform: Windows-11-10.0.26100-SP0
   > PyTorch: 2.7.0+cu128
   > CUDA:    12.8
   > GPU:     NVIDIA GeForce RTX 5090
[OK] Environment matches RTX 5090 requirements.
------------------------------------------------------------
[OK] SUCCESS: spconv
      Location: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\spconv\pytorch\__init__.py
[OK] SUCCESS: nvdiffrast
      Location: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\nvdiffrast\torch\__init__.py
[OK] SUCCESS: simple_knn
      Location: None
[OK] SUCCESS: diff_gaussian_rasterization
      Location: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\diff_gaussian_rasterization\__init__.py
[OK] SUCCESS: kiui
      Location: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\kiui\__init__.py
[OK] SUCCESS: pytorch3d
      Location: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\pytorch3d\__init__.py

ice: cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync
Using async weight offloading with 2 streams
Enabled pinned memory 29210.0
WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.
Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
Found comfy_kitchen backend cuda: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']}
Using sage attention
Python version: 3.12.3 (tags/v3.12.3:f6650f9, Apr  9 2024, 14:05:25) [MSC v.1938 64 bit (AMD64)]
ComfyUI version: 0.10.0
ComfyUI frontend version: 1.36.14
[Prompt Server] web root: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\comfyui_frontend_package\static
c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\kiui\nn\__init__.py:31: FutureWarning: `torch.cuda.amp.custom_fwd(args...)` is deprecated. Please use `torch.amp.custom_fwd(args..., device_type='cuda')` instead.
  @torch.cuda.amp.custom_fwd(cast_inputs=torch.float32)

c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\kiui\nn\__init__.py:37: FutureWarning: `torch.cuda.amp.custom_bwd(args...)` is deprecated. Please use `torch.amp.custom_bwd(args..., device_type='cuda')` instead.
  @torch.cuda.amp.custom_bwd

Warn!: xFormers is available (Attention)
Warn!: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\slangtorch\slangtorch.py:3: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
  import pkg_resources

Warn!: c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\setuptools\_distutils\_msvccompiler.py:12: UserWarning: _get_vc_env is private; find an alternative (pypa/distutils#340)
  warnings.warn(

Warn!: Traceback (most recent call last):
  File "c:\ComfyUI_windows_portable_nightly_pytorch\ComfyUI\nodes.py", line 2158, in load_custom_node
    module_spec.loader.exec_module(module)
  File "<frozen importlib._bootstrap_external>", line 995, in exec_module
  File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
  File "c:\ComfyUI_windows_portable_nightly_pytorch\ComfyUI\custom_nodes\ComfyUI-3D-Pack\__init__.py", line 54, in <module>
    module = importlib.import_module(f".{nodes_filename}", package=__name__)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "importlib\__init__.py", line 90, in import_module
  File "<frozen importlib._bootstrap>", line 1387, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1360, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1331, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 935, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 995, in exec_module
  File "<frozen importlib._bootstrap>", line 488, in _call_with_frames_removed
  File "c:\ComfyUI_windows_portable_nightly_pytorch\ComfyUI\custom_nodes\ComfyUI-3D-Pack\nodes.py", line 85, in <module>
    from Unique3D.scripts.project_mesh import multiview_color_projection, multiview_color_projection_texture, get_cameras_list, get_orbit_cameras_list
  File "c:\ComfyUI_windows_portable_nightly_pytorch\ComfyUI\custom_nodes\ComfyUI-3D-Pack\Gen_3D_Modules\Unique3D\scripts\project_mesh.py", line 99, in <module>
    pix2faces_renderer = Pix2FacesRenderer()
                         ^^^^^^^^^^^^^^^^^^^
  File "c:\ComfyUI_windows_portable_nightly_pytorch\ComfyUI\custom_nodes\ComfyUI-3D-Pack\Gen_3D_Modules\Unique3D\scripts\project_mesh.py", line 71, in __init__
    self._glctx = dr.RasterizeCudaContext(device=device)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\nvdiffrast\torch\ops.py", line 184, in __init__
    self.cpp_wrapper = _get_plugin().RasterizeCRStateWrapper(cuda_device_idx)
                       ^^^^^^^^^^^^^
  File "c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\nvdiffrast\torch\ops.py", line 125, in _get_plugin
    torch.utils.cpp_extension.load(name=plugin_name, sources=source_paths, extra_cflags=common_opts+cc_opts, extra_cuda_cflags=common_opts+['-lineinfo'], extra_ldflags=ldflags, with_cuda=True, verbose=False)
  File "c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\torch\utils\cpp_extension.py", line 1623, in load
    return _jit_compile(
           ^^^^^^^^^^^^^
  File "c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\torch\utils\cpp_extension.py", line 2076, in _jit_compile
    _write_ninja_file_and_build_library(
  File "c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\torch\utils\cpp_extension.py", line 2180, in _write_ninja_file_and_build_library
    verify_ninja_availability()
  File "c:\ComfyUI_windows_portable_nightly_pytorch\python_embeded\Lib\site-packages\torch\utils\cpp_extension.py", line 2241, in verify_ninja_availability
    raise RuntimeError("Ninja is required to load C++ extensions")
RuntimeError: Ninja is required to load C++ extensions

Warn!: Cannot import c:\ComfyUI_windows_portable_nightly_pytorch\ComfyUI\custom_nodes\ComfyUI-3D-Pack module for custom nodes: Ninja is required to load C++ extensions
[ComfyUI-Easy-Use] server: v1.3.4 Loaded
[ComfyUI-Easy-Use] web root: c:\ComfyUI_windows_portable_nightly_pytorch\ComfyUI\custom_nodes\comfyui-easy-use\web_version/v2 Loaded
[GeomPack] CGAL Python package found - CGAL Isotropic Remesh node available

[DenisKochetov]

@gerroon

Hi, this error is almost never about installing ninja — it’s about which ninja ComfyUI actually sees at runtime.

From similar cases, this is usually caused by PATH / MSVC issues, especially on portable + RTX 5090 setups.

Please check the following exactly in this order:

1. Verify which ninja is being used

Run in cmd.exe (not PowerShell):

where ninja
ninja --version

It must point to:

<ComfyUI>\python_embeded\Scripts\ninja.exe

If it points elsewhere (conda, system, CUDA), PyTorch will fail to detect it.

---

2. Reinstall ninja into ComfyUI embedded Python

cd <ComfyUI>\python_embeded
python.exe -m pip install --force-reinstall ninja

Verify Python sees it:

python.exe -c "import shutil; print(shutil.which('ninja'))"

---

3. Fix PATH order (critical)

Add these to System PATH and move them to the top:

<ComfyUI>\python_embeded
<ComfyUI>\python_embeded\Scripts

Restart the system after this.

---

4. Ensure MSVC is available

Check:

where cl

If nothing is returned, install Visual Studio 2022 Build Tools with:

• MSVC v143
• Windows 10/11 SDK

Guide:
https://github.com/bycloudai/InstallVSBuildToolsWindows

---

5. (RTX 5090) Explicitly set CUDA arch

Some JIT builds fail without this:

set TORCH_CUDA_ARCH_LIST=12.0

(or set it in Python before importing torch)

---

If ninja is found correctly and MSVC is available, this error disappears.
If it still fails, manually running ninja in the torch_extensions build directory will reveal the real linker error.

Hope this helps.

[gerroon]

HI @DenisKochetov Thank you so much that was a lot of help, so I devised this script and reinstalled all the stuff yo u mentioned, I already had them but maybe COmfy python wasnt seeing them. Now I want to make sure comfy3d is installed properly, it seems it installed in the comfyu manager. Is there a way to rerun the insrtaller sfely or test? There are so many nodes that come with it it would be hard to test them all for testing purposes manually

Thanks

@echo off
setlocal

:: 1. SET DYNAMIC ROOT
:: "%~dp0" automatically expands to the folder this script is inside.
cd /d "%~dp0"
set "ROOT=%CD%"
echo [INIT] Running from: %ROOT%

:: 2. FIND GIT (Critical for Manager)
:: We capture your existing Git path before cleaning the environment.
set "GIT_PATH="
for /f "delims=" %%i in ('where git') do set "GIT_FULL=%%i" & goto :found_git
:found_git
if defined GIT_FULL (
    echo [INIT] Git found at: %GIT_FULL%
    for %%F in ("%GIT_FULL%") do set "GIT_PATH=%%~dpF"
) else (
    echo [WARNING] Git not found! ComfyUI-Manager may fail.
)

:: 3. BUILD CLEAN PATH (The "Isolation" Fix)
:: We wipe the PATH and rebuild it with ONLY:
::   1. Portable Python (First priority)
::   2. Windows System Defaults
::   3. Git (for Manager)
:: This GUARANTEES the system Ninja is invisible.
set "PATH=%ROOT%\python_embeded;%ROOT%\python_embeded\Scripts;C:\Windows\System32;C:\Windows;C:\Windows\System32\Wbem;%GIT_PATH%"

:: 4. INITIALIZE MSVC (Issue #558 Fix)
:: Adds Visual Studio 2022 Community tools to our clean path.
call "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat"

:: 5. SET 5090 BLACKWELL FLAGS (Issue #5090 Fix)
echo.
echo [INIT] Setting RTX 5090 Flags (sm_120)...
set "TORCH_CUDA_ARCH_LIST=12.0"
set "DISTUTILS_USE_SDK=1"
set "FORCE_CUDA=1"

:: 6. LAUNCH
echo.
echo [LAUNCH] Starting ComfyUI...
echo ----------------------------------------------------------------
.\python_embeded\python.exe -s ComfyUI\main.py ^
    --windows-standalone-build ^
    --listen ^
    --port 8188 ^
    --use-sage-attention ^
    --disable-auto-launch ^
    --disable-xformers

pause

[DenisKochetov]

@gerroon There are many nodes available, but for SOTA workflows I recommend Hunyuan 2 Turbo (mesh) and Hunyuan 2.1 (texturing).
There is also an open-source alternative that might perform better for mesh generation, though I haven’t tested it yet:
https://github.com/jtydhr88/ComfyUI-UltraShape1

[gerroon]

Thanks man you are very helpful. I hjave another question if you dont mind. Would it be possible tooo make these work together? https://github.com/visualbruno/ComfyUI-Trellis2, or do we know if there are plans to include trellis2 in the pack?

[DenisKochetov]

@gerroon You could try Trellis2 in our environment first — it’ll probably just need a few extra packages.
What do you feel is missing in the Trellis2 node that you’re hoping to get from ours?

[gerroon]

I like to have all in one type of solution for 3d stuff. Trellis2 is really good. I have a separate Comfy for Trellis and another for 3DPack. The last time I tried I couldnot get them agree on the packages etc but I didn't have the working 3dPack either. I was trying to install 3DPack inthe Trellis2 Comfy.