配置

硬件

  • i7-11700@2.50GHz
  • 64GB RAM
  • RTX 3060
    • 12GB

软件

1
pip install torch torchvision trochaudio --index-url https://download.pytorch.org/whl/cu128

模型

本地部署使用，每个模型使用独立的虚拟环境运行。

1
2
python -m venv venv
.\venv\Scripts\activate

SF3D

1
2
3
# code
git clone https://github.com/Stability-AI/stable-fast-3d.git
cd stable-fast-3d

依赖安装报错，把 requirement.txt 里的 texture_baker 和 uv_unwrapper 删除，手动安装

1
2
3
4
5
6
7
8
9
10
11
12
# dependencies
cd texture_baker
python setup.py install
cd ../uv_unwrapper
python setup.py install
cd ..
pip install -r requirements.txt
pip install -r requirements-demo.txt
pip install -U gradio gradio_client     # gradio-litmodel3d incompatible 暂时不管

# interface
python gradio_app.py

TripoSR

1
2
3
4
5
6
7
8
9
10
11
# code
git clone https://github.com/VAST-AI-Research/TripoSR.git
cd TripoSR

# dependencies
pip install -r requirements.txt
pip install onnxruntime
pip install -U gradio transformers tokenizers

# interface
python gradio_app.py --listen

Hunyuan3D-2.0

1
2
3
4
5
6
7
8
9
10
11
# code
git clone https://github.com/Tencent-Hunyuan/Hunyuan3D-2.git
cd Hunyuan3D-2

# dependencies
pip install -r requirements.txt
pip install -e .
cd hy3dgen/texgen/custom_rasterizer
python setup.py install
cd ../differentiable_renderer
python setup.py install

直接运行的话路由有问题，需修改 gradio_app.py

1
app = gr.mount_gradio_app(app, demo, path="") # 删除 /

运行

1
2
# image-to-3d
python gradio_app.py --model_path tencent/Hunyuan3D-2 --subfolder hunyuan3d-dit-v2-0 --texgen_model_path tencent/Hunyuan3D-2 --low_vram_mode

启用 text-to-3d 需要加个 patch，在项目根目录新建 fix_lazy_loading.py

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
import sys
import importlib

def patch_t5_tokenizer():
    """Force proper loading of T5 tokenizer"""
    try:
        # Force import of T5 tokenizer module
        import transformers.models.t5.tokenization_t5
        
        # Ensure the module is properly loaded
        if hasattr(transformers.models.t5.tokenization_t5, 'T5Tokenizer'):
            print("T5Tokenizer successfully loaded")
            return True
        else:
            print("T5Tokenizer not found in module")
            return False
    except Exception as e:
        print(f"Failed to patch T5 tokenizer: {e}")
        return False

def patch_transformers():
    """Apply patches to fix lazy loading"""
    # Import transformers first
    import transformers
    
    # Force load tokenizer modules
    from transformers import T5Tokenizer, T5TokenizerFast
    
    # Monkey patch the lazy module if needed
    import transformers.models.t5.tokenization_t5 as t5_tokenizer_module
    
    # Ensure classes are properly exposed
    if not hasattr(t5_tokenizer_module, 'T5Tokenizer'):
        t5_tokenizer_module.T5Tokenizer = T5Tokenizer
    
    print("Transformers patching complete")

if __name__ == "__main__":
    patch_transformers()
    patch_t5_tokenizer()

更新 gradio_app.py，导入 patch 文件

1
2
3
import logger
import fix_lazy_loading
fix_lazy_loading.patch_transformers()

启用 text-to-3d 需要在运行时添加命令行参数 --enable_t23d

1
2
3
4
5
# dependencies
pip install sentencepiece

# text-to-3d
python gradio_app.py --model_path tencent/Hunyuan3D-2 --subfolder hunyuan3d-dit-v2-0 --texgen_model_path tencent/Hunyuan3D-2 --low_vram_mode --enable_t23d

文本转模型相当慢，预计 40min 左右😂

TRELLIS

Windows 上直接跑各种问题（Setting Up Trellis on Windows #3），后面直接找了个现成的 IgorAherne/trellis-stable-projectorz installer，暂不支持 text-to-3d

修改 code/gradio_main.py ，绑定到 0.0.0.0

1
demo.launch(server_name="0.0.0.0") # 添加参数

解压后双击运行 run-gradio-fp16.bat，会自动创建虚拟环境。如果需要登录 huggingface 的话进入虚拟环境操作即可。

InstantMesh

1
2
3
4
5
6
7
8
9
10
11
# code
git clone https://github.com/TencentARC/InstantMesh.git
cd InstantMesh

# dependencies
pip install Ninja onnxruntime
pip install -r requirements.txt
pip install -U gradio tokenizers transformers diffusers

# interface
python app.py

Unique3D

  1. cuDNN 9.12.0：https://developer.nvidia.com/cudnn-downloads
  2. TensorRT 10.13.2 GA: https://developer.nvidia.com/tensorrt/download/10x 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
# code
git clone https://github.com/AiuniAI/Unique3D.git
cd Unique3D

# 下载 ckpt 文件夹
https://huggingface.co/spaces/Wuvin/Unique3D/tree/main

# torch
pip install torch torchvision torchaudio xformers --index-url https://download.pytorch.org/whl/cu128
pip install torch-scatter -f https://data.pyg.org/whl/torch-2.8.0+cu128.html
pip install onnxruntime-gpu --extra-index-url https://pypi.ngc.nvidia.com
# dependencies
pip install diffusers==0.27.2 huggingface_hub==0.34.4 opencv-python-headless==4.8.1.78
pip install ninja rembg peft gradio datasets accelerate
pip install wandb omegaconf transformers huggingface_hub[hf_xet] hf_xet
pip install pygltflib trimesh typeguard fire jaxtyping pymeshlab "triton-windows<3.5"
# pytorch3d
git clone https://github.com/facebookresearch/pytorch3d.git
cd pytorch3d
python setup.py install
cd ..
# nvdiffrast
git clone https://github.com/NVlabs/nvdiffrast.git
cd nvdiffrast
pip install .
cd ..

# interface
python app/gradio_local.py --listen --port 7860

修改 ./app/gradio_local.py 添加调用参数 allowed_paths，创建并添加临时目录（运行的磁盘，可能需要手动创建）

1
2
3
4
5
6
7
demo.queue(default_concurrency_limit=1).launch(
    server_port=None if port == 0 else port,
    share=share,
    root_path=gradio_root if gradio_root != "" else None,  # "/myapp",
    allowed_paths=["D:/tmp/gradio"], # 新增
    **launch_args,
)

报错：ImportError: cannot import name 'cached_download' from 'huggingface_hub'

  • 在 .\venv\Lib\site-packages\diffusers\utils\dynamic_modules_utils.py 的 import 中删除 cached_download 导入

报错：显存不足（暂时没管）

1
2
3
4
5
2025-09-05 11:40:44.0622507 [E:onnxruntime:Default, tensorrt_execution_provider.h:89 onnxruntime::TensorrtLogger::log] [2025-09-05 03:40:44   ERROR] [virtualMemoryBuffer.cpp::nvinfer1::StdVirtualMemoryBufferImpl::resizePhysical::141] Error Code 1: Cuda Driver (In nvinfer1::StdVirtualMemoryBufferImpl::resizePhysical at optimizer/builder/virtualMemoryBuffer.cpp:141)
2025-09-05 11:40:44.0821153 [E:onnxruntime:Default, tensorrt_execution_provider.h:89 onnxruntime::TensorrtLogger::log] [2025-09-05 03:40:44   ERROR] [globWriter.cpp::nvinfer1::builder::HybridGlobWriter::HybridGlobWriter::494] Error Code 2: OutOfMemory (Requested size was 1024 bytes.)
2025-09-05 11:40:44.0911560 [E:onnxruntime:, sequential_executor.cc:572 onnxruntime::ExecuteKernel] Non-zero status code returned while running TRTKernel_graph_main_graph_15095719904311875781_0 node. Name:'TensorrtExecutionProvider_TRTKernel_graph_main_graph_15095719904311875781_0_0' Status Message: TensorRT EP failed to create engine from network.
EP Error: [ONNXRuntimeError] : 11 : EP_FAIL : Non-zero status code returned while running TRTKernel_graph_main_graph_15095719904311875781_0 node. Name:'TensorrtExecutionProvider_TRTKernel_graph_main_graph_15095719904311875781_0_0' Status Message: TensorRT EP failed to create engine from network. using ['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']
Falling back to ['CPUExecutionProvider'] and retrying.

requirements.txt

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
accelerate==1.10.1
aiofiles==23.2.1
aiohappyeyeballs==2.6.1
aiohttp==3.12.15
aiosignal==1.4.0
altair==5.5.0
annotated-types==0.7.0
antlr4-python3-runtime==4.9.3
anyio==4.10.0
attrs==25.3.0
Brotli==1.1.0
certifi==2025.8.3
charset-normalizer==3.4.3
click==8.2.1
colorama==0.4.6
coloredlogs==15.0.1
contourpy==1.3.3
cycler==0.12.1
dataclasses-json==0.6.7
datasets==4.0.0
Deprecated==1.2.18
diffusers==0.27.2
dill==0.3.8
fastapi==0.116.1
ffmpy==0.6.1
filelock==3.13.1
fire==0.7.1
flatbuffers==25.2.10
fonttools==4.59.2
frozenlist==1.7.0
fsspec==2024.2.0
gitdb==4.0.12
GitPython==3.1.45
gradio==5.44.1
gradio_client==1.12.1
groovy==0.1.2
h11==0.16.0
hf-xet==1.1.9
httpcore==1.0.9
httpx==0.28.1
huggingface-hub==0.34.4
humanfriendly==10.0
idna==3.10
imageio==2.37.0
importlib_metadata==8.7.0
importlib_resources==6.5.2
iopath==0.1.10
jaxtyping==0.3.2
Jinja2==3.1.4
jsonschema==4.25.1
jsonschema-specifications==2025.4.1
kiwisolver==1.4.9
lazy_loader==0.4
llvmlite==0.44.0
markdown-it-py==4.0.0
MarkupSafe==2.1.5
marshmallow==3.26.1
matplotlib==3.10.6
mdurl==0.1.2
mpmath==1.3.0
msvc_runtime==14.44.35112
multidict==6.6.4
multiprocess==0.70.16
mypy_extensions==1.1.0
narwhals==2.3.0
networkx==3.3
ninja==1.13.0
numba==0.61.2
numpy==1.26.4
nvdiffrast @ file:///D:/workplace/Unique3D/nvdiffrast
omegaconf==2.3.0
onnxruntime-gpu==1.22.0
opencv-python-headless==4.8.1.78
orjson==3.11.3
ort-nightly-gpu==1.15.0.dev20230502003
packaging==25.0
pandas==2.3.2
peft==0.17.1
pillow==10.4.0
platformdirs==4.4.0
pooch==1.8.2
portalocker==3.2.0
propcache==0.3.2
protobuf==6.32.0
psutil==7.0.0
pyarrow==21.0.0
pyarrow-hotfix==0.7
pydantic==2.6.1
pydantic_core==2.16.2
pydub==0.25.1
pygltflib==1.16.5
Pygments==2.19.2
PyMatting==1.1.14
pymeshlab==2025.7
pyparsing==3.2.3
pyreadline3==3.5.4
python-dateutil==2.9.0.post0
python-multipart==0.0.20
pytorch3d==0.7.8
pytz==2025.2
pywin32==311
PyYAML==6.0.2
referencing==0.36.2
regex==2025.9.1
rembg==2.0.67
requests==2.32.5
rich==14.1.0
rpds-py==0.27.1
ruff==0.12.11
safehttpx==0.1.6
safetensors==0.6.2
scikit-image==0.25.2
scipy==1.16.1
semantic-version==2.10.0
sentry-sdk==2.36.0
shellingham==1.5.4
six==1.17.0
smmap==5.0.2
sniffio==1.3.1
starlette==0.47.3
sympy==1.13.3
tensorrt @ file:///C:/Program%20Files/NVIDIA/TensorRT-10.13.2.6/python/tensorrt-10.13.2.6-cp311-none-win_amd64.whl#sha256=bed88a7ce09338faf6a48f82954b20d8993189a2a4ce2d4401899fdcf859931a
termcolor==3.1.0
tifffile==2025.8.28
tokenizers==0.22.0
tomlkit==0.12.0
torch==2.8.0+cu128
torch_scatter==2.1.2+pt28cu128
torchvision==0.23.0+cu128
tqdm==4.67.1
transformers==4.56.1
trimesh==4.8.1
triton-windows==3.4.0.post20
typeguard==4.4.4
typer==0.17.3
typing-inspect==0.9.0
typing-inspection==0.4.1
typing_extensions==4.15.0
tzdata==2025.2
urllib3==2.5.0
uvicorn==0.35.0
wandb==0.21.3
websockets==11.0.3
wrapt==1.17.3
xformers==0.0.32.post2
xxhash==3.5.0
yarl==1.20.1
zipp==3.23.0

Stable-Zero123

需要通过 threestudio 使用

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
# model
git clone https://huggingface.co/stabilityai/stable-zero123
git clone https://github.com/threestudio-project/threestudio.git
cp stable-zero123/stable_zero123.ckpt threestudio/load/zero123/

# dependencies
cd threestudio
pip install ninja
# 注释 nerfacc envlight nvdiffrast tiny-cuda-nn
pip install -r requirements.txt
pip install nerfacc==0.5.2 envlight
pip install libigl==2.5.1
pip install huggingface_hub==0.25.*

git clone https://github.com/NVlabs/nvdiffrast.git
cd nvdiffrast
pip install .

python -m pip install pip==23.0.1
pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

添加环境变量

1
D:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\14.44.35207\bin\Hostx64\x64

直接跑预估 68h，trainer.max_steps 配置为 600，改成 5 只需要半小时

1
python launch.py --config configs/stable-zero123.yaml --train --gpu 0 trainer.max_steps=5 data.image_path=./load/images/hamburger_rgba.png

但是导出模型报错，暂未解决（太难用了）

1
python launch.py --config "outputs/zero123-sai/[64, 128, 256]_hamburger_rgba.png@20250901-142614/configs/parsed.yaml" --export --gpu 0 resume="outputs/zero123-sai/[64, 128, 256]_hamburger_rgba.png@20250901-142614/ckpts/last.ckpt" system.exporter_type=mesh-exporter

参阅