2024/12/14

Major Changes | 大改动

• Starting from this release, this package no longer includes ComfyUI-3D-Pack. Users of Comfy3D please use the dedicated package instead.
• 自本次发布起，本整合包不再包含 ComfyUI-3D-Pack，Comfy3D 用户请使用专门的 整合包 。

Minor Changes | 小改动

• Disable hf_transfer by default.
• 默认禁用 hf_transfer。

Versions Info | 版本信息

• Python 3.12.8
• CUDA 12.4
• PyTorch 2.5.1
• xFormers 0.0.28.post3

What's Next | 新版展望

• New releases will come with more custom nodes pre-installed.
• 新版将会预装更多节点。

2024/11/07

Temporarily add custom nodes:

临时添加节点：

• mixlab-nodes

Installing or updating mixlab-nodes via Manager will cause downgrade of setuptools (a core package of Python), and will further cause dependency chaos. So it won't be included in the next stable release.

通过 Manager 安装或更新 mixlab-nodes 会导致 Python 核心包 setuptools 降级，并进一步造成依赖项不兼容，因此下一版发布中不会包含该节点。

2024/11/04

• Doc for this release: README.adoc
• 适用于该发布版本的说明文档： README.zh.adoc

Note | 简要

• This update only includes minor tweaks to the user experience, with no major version updates.
• 本次更新仅调整使用体验，无重大版本更新。

Changes in Start Script

run_nvidia_gpu.bat:

• Enable hf_transfer lib for high-speed file transfer feature of Hugging Face Hub.
  • If HF Hub freezes while downloading, disable this setting (remove the line in start script).
• Use a local folder HuggingFaceHub to save model files downloaded by Hugging Face Hub.
  • You can copy/move your cache %USERPROFILE%\.cache\huggingface\hub to the HuggingFaceHub folder if needed.
  • If you like the old way, simply remove the line in start script.
• Add comments about setting HuggingFace Access Token.

启动脚本变动

run_cn.bat:

• 启用 hf_transfer，Hugging Face Hub 的实验性高速传输功能。
  • 如果 HF Hub 下载时整个程序卡住，在启动脚本里删除该行即可。这是个 rust 多线程传输库，不会显示下载进度。
• 现在程序根目录下有个 HuggingFaceHub 文件夹用来保存 HF Hub 下载的模型文件。
  • 以前的模型文件在 %USERPROFILE%\.cache\huggingface\hub，可以将其移动或复制过来。
  • 或者，如果你喜欢以前的保存位置，将启动脚本里对应行删除即可。
• 添加关于 HuggingFace Access Token 的说明。

Added Custom Nodes | 预装更多自定义节点

➕ ComfyUI-Easy-Use
➕ ComfyUI-GGUF
➕ ComfyUI-to-Python-Extension

Versions Info | 版本信息

• Python 3.12.7
• CUDA 12.4
• PyTorch 2.5.1
• xFormers 0.0.28.post3
• Using wheels for Comfy3D compiled by me.

---

Previous Releases | 经典版本一览

• For 3D-Pack users, keep in mind that 3D workflows are very picky about versions compatibility (due to its Torch C++ JIT nature).
  • Use the PyTorch 2.3.0 version if you only use 3D-Pack.
• Newer version of PyTorch tend to have better performance, but not always significant for 3D workflows.
• Furthermore, I made a dedictated package for Stable Fast 3D.

• 3D-Pack 用户留意，因为 3D 工作流需要运行时编译 Torch C++ 扩展，其对兼容性十分挑剔。
  • 如果你只用 3D-Pack，推荐用 PyTorch 2.3.0 版本 。
• 新版 PyTorch 性能更好，但对 3D 工作流并不明显。
• 此外，我专门为 Stable Fast 3D 制作了一个 整合包 。

Date	PyTorch	Python	CUDA	Note	备注
2024/05/21	2.2.2	3.11	12.1	Used by #7. Good if you want to use older version of 3D-Pack	适合运行老版本 3D-Pack
2024/07/01	2.2.2	3.11	12.1	Almost the same as 05/21, only update	与上一版几乎一样，仅小更新
2024/09/10	2.3.0	3.11	12.1	Most likely to successfully run all 3D-Pack workflows	目前兼容性最好，最有可能运行全部 3D-Pack 工作流
2024/10/01	2.4.0	3.11	12.1	Second most likely to successfully run all 3D-Pack workflows	第二兼容性好
2024/10/18	2.4.0	3.12	12.1	Minor improvement to the 10/01. Upgrade to Python 3.12	对上一版的小改进，升级到 Python 3.12

2024/10/31

Changes | 变动

• Upgrade to PyTorch 2.5.1
• 升级到 PyTorch 2.5.1

Known Issues | 已知问题

• If you encounter a slangtorch error while running the StableFast3D workflow, it is a compatibility issue that cannot be resolved at this time. Please use the dedictated package for Stable Fast 3D.
• 如果运行 StableFast3D 工作流时遇到 slangtorch 报错，这是个暂时无法解决的兼容性问题。请使用 Stable Fast 3D 整合包 。

Versions | 版本信息

• Python 3.12.7
• CUDA 12.4
• PyTorch 2.5.1
• xFormers 0.0.28.post3
• Using wheels for Comfy3D compiled by me.

Wheel Files for ComfyUI-3D-Pack 2024/10/31

Note: You don't need to install these manually, they're already installed in the package.

Python 3.12
PyTorch 2.5.1+cu124

Instruction for building: compiling-wheels-for-comfy3d.adoc

Note that texture_baker, uv_unwrapper and pynim are compiled for requirements.txt of stable-fast-3d. They are not needed for 3D-Pack.

2024/10/29

Changes | 变动

• Upgrade to PyTorch 2.5 + CUDA 12.4.
  • Using wheels for Comfy3D compiled by me.
• Add custom node: PuLID.
• Add a script for re-compiling and re-installing Comfy3D dependencies.
• Add a script for force-updating ComfyUI and all custom nodes.
• bugfix: wrong python_embeded\include path

• 升级到 PyTorch 2.5 + CUDA 12.4
  • 用了我自己编译的 Comfy3D 依赖项 wheel 文件 ，看起来兼容性更好了（文件也更大了）。

• 添加自定义节点： PuLID 。
• 添加启动脚本 run_cn.bat 便于大陆地区使用。
• 添加 使用国内镜像强制更新全部节点.sh 便于大陆地区使用。
• 添加 重新编译安装3D-Pack依赖项.bat 便于用户解决兼容性问题。
• 修复 bug：python_embeded\include 路径错误。

Known Issues | 已知问题

• In order to run StableFast3D workflow, you need to download unordered_dense.h into python_embeded\include\ankerl\ folder.
• 如需运行 StableFast3D 工作流，需下载 unordered_dense.h 文件到 python_embeded\include\ankerl\ 目录下。

Versions | 版本信息

• Python 3.12.7
• CUDA 12.4
• PyTorch 2.5.0
• xFormers 0.0.28.post2

2024/10/27

Changes | 变动

• Upgrade to PyTorch 2.5 + CUDA 12.4
• 升级到 PyTorch 2.5 + CUDA 12.4

• Using wheels for Comfy3D compiled by me. May have compatibility issues.
• 用了我自己编译的 Comfy3D 依赖项 wheel 文件，可能会有兼容性问题（也可能兼容性更好），暂不作为正式版发布。如果反馈良好，性能和兼容性兼得，则转正。

• Add custom node: PuLID
• 添加自定义节点：PuLID

Known issue

Wrong folder. Please move files from
python_embeded\Include\include
To
python_embeded\Include

Versions | 版本信息

• Python 3.12.7
• CUDA 12.4
• PyTorch 2.5.0
• xFormers 0.0.28.post2

Wheel Files for ComfyUI-3D-Pack 2024/10/27

Python 3.12
PyTorch 2.5.0+cu124

Instruction for building: https://github.com/YanWenKun/ComfyUI-Windows-Portable/blob/main/docs/compiling-wheels-for-comfy3d.adoc

2024/10/25 Without 3D-Pack

Note | 提示

• With the main components using the latest versions, this release may have better performance for new models like FLUX and SD 3.5.
• 本次发布版本追新，意在 Flux 与 SD 3.5 等新模型的性能。

• This release doesn't include:
• 本次发布不含：
  • 3D-Pack
  • InstantID
  • ReActor
  • AnimateDiff
  • AnimateAnyone
  • FaceAnalysis
  • WAS NS
  • MTB Nodes

Hint | 使用技巧

• If you're using 40 series / Ada GPU or newer, you can try adding --fast to the startup arguments in run_nvidia_gpu.bat.
• 如果你在用 40 系或更新 GPU，可以尝试开启实验性参数 --fast，编辑 run_nvidia_gpu.bat 添加至启动行末尾即可。
• Source: comfyanonymous/ComfyUI@9953f22

• If you want to turn-off xFormers, add --use-pytorch-cross-attention to the startup arguments in run_nvidia_gpu.bat.
• 如果你不想启用 xFormers，编辑 run_nvidia_gpu.bat 并在启动参数里添加 --use-pytorch-cross-attention。

Versions | 版本信息

• Python 3.12.7
• CUDA 12.4
• PyTorch 2.5.0
• xFormers 0.0.28.post2

2024/10/18

How to use:	如何使用
1. Download all .7z.00X files	1. 下载所有 .7z.00X 文件
2. Open the .001 file and extract	2. 打开第一个文件（.001）并解压
3. Edit run_nvidia_gpu.bat to fit your GPU	3. 编辑 run_nvidia_gpu.bat 以适配你的 GPU
4. Run run_nvidia_gpu.bat to start ComfyUI	4. 运行 run_nvidia_gpu.bat 启动 ComfyUI
Note: You can install custom nodes via "ComfyUI-Manager"	提示：使用“管理器”可以安装自定义节点

Changes | 变动

• Add script reinstall-deps-for-3d.bat that build and re-install dependencies for Comfy3D.
• 添加脚本 重新编译安装3D-Pack依赖项.bat 。

Versions | 版本信息

• Python 3.12.7
• CUDA 12.1
• PyTorch 2.4.0
• xFormers 0.0.27.post2