How to Generate WAN 2.2 Videos with Forge Neo: Easier Video Generation Than ComfyUI
In this article, I introduce “Stable Diffusion WebUI Forge – Neo”. The original Automatic1111 Stable Diffusion WebUI appears to be designed specifically for U-Net (Stable Diffusion models), and support for the recent Transformer (DiT-based) models seems to have made little progress. The last major update was SD3 support in July 2024, with only minor bug fixes in February 2025. While you can run Flux.1 on Stable Diffusion WebUI Forge, it does not support next-generation models like WAN 2.2. The fork “Stable Diffusion WebUI Forge – Neo”, developed by Haoming02, supports a wide range of the latest models and attention mechanisms, making it especially recommended for those who find ComfyUI difficult.
What You’ll Learn in This Article
- Features of Stable Diffusion WebUI Forge – Neo and its differences from the original Forge
- How to install uv and FFmpeg, and setup steps for Forge-Neo
- How to configure WAN 2.2 T2V (text-to-video generation) and recommended parameters (💎Members only)
- How to use WAN 2.2 I2V (image-to-video generation) and how to generate input images (💎Members only)
- How to choose the optimal resolution for your VRAM and save disk space by sharing models (💎Members only)
What Is Stable Diffusion WebUI Forge – Neo?
Stable Diffusion WebUI Forge – Neo is a fork born from taking over the original Forge project developed by lllyasviel. It is designed with an emphasis on “lightweight design” and “practicality”, and the developer’s policy is clear: it adheres to a simple philosophy of “providing a GUI that lets anyone easily use the latest models while eliminating all unnecessary features.”
It also uses the official latest stable version of PyTorch (2.10.0+cu130), so it runs very stably on RTX 50 series GPUs as well. It is one of the most recommended combinations at this time.
Greatly Expanded Model Support
The February 2026 update added support for quite a large number of new models. The most notable points are as follows.
- The latest image generation models such as Flux.2-Klein and Anima
- Video generation functionality via Wan 2.2
- The ability to input multiple images simultaneously with the Qwen-Image series
- Support for LoRA for Flux and Qwen via Nunchaku (SVDQ)
These additions have expanded the tool’s capabilities beyond simple image generation to include video generation and advanced editing tasks.
Major Performance Improvements
Technical improvements are remarkable and make a significant difference in practical use.
- Redesigned preset system: Automatically remembers the models and settings you used, improving workflow reproducibility.
- UV package manager support: Significantly reduces the time required for initial setup.
- Multiple attention optimizations: SageAttention and FlashAttention achieve faster processing while reducing memory consumption.
- GPU-accelerated tile processing: Processing during upscaling has become dramatically faster.
These improvements allow larger models to be run even with limited VRAM.
Features Removed from the Original Forge
The following features have been removed from the original Forge. If you want to use these features, use the original Forge.
- Support for SD2 and SD3
- Hypernetworks and Textual Inversion training features
- Tagging tools such as CLIP Interrogator and Deepbooru
- Many built-in extensions
By removing training features, it achieves a lightweight design as a pure inference environment.
Important Notes for Use
- Supported environment: Basically Windows + NVIDIA GPU. Linux, macOS, AMD, and Intel are not officially supported.
- Restoring removed features: The developer’s policy is not to respond to requests for removed features.
The greatest feature of the Neo version is its commitment to a “clean implementation.” While the original Forge automatically clones external repositories, Neo operates with only the minimum necessary dependencies. It also does not include bitsandbytes by default, adopting an approach where users add it as needed.
How to Install Stable Diffusion WebUI Forge – Neo
Prerequisites: Installing "uv"
Before installing Forge – Neo, first let’s install “uv.”
“uv” is a fast Python package and project management tool written in Rust, developed by Astral. It integrates the functionality of pip, pip-tools, virtualenv, poetry, and more into one, and is significantly faster than conventional tools.
First, let’s check whether “uv” is installed on your PC. Open PowerShell or Command Prompt and check with the following command.
uv --versionIf it is installed, version information such as uv 0.10.9 (f675560f3 2026-03-06) will be displayed.
If an error occurs, it is not installed, so install it with the following command.
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"After installation is complete, run the following command again to confirm it is correctly installed.
uv --versionPrerequisites: Installing "FFmpeg" (Optional)
If you want to use WAN 2.2, you need to install FFmpeg.
The easiest method is to go to the 🔗”FFmpeg” download page, click the Windows logo under “Get packages & executable files,” and download and install one from the list under “Windows EXE Files.”
Cloning the Stable Diffusion WebUI Forge-Neo Repository
✅ Cloning the repository requires 🔗git.
In Command Prompt, navigate to the directory where you want to install (such as %USERPROFILE%), then clone with the following command.
git clone https://github.com/Haoming02/sd-webui-forge-classic sd-webui-forge-neo --branch neoSetting Up Python
Once the clone is complete, navigate to the Forge-neo directory with the following command.
cd sd-webui-forge-neoOnce there, use the following command to set up the venv environment and install Python. ✅ This method does not conflict with the Python installed on your main Windows system.
uv venv venv --python 3.13 --seedvenv venv is a uv subcommand that creates a venv named “venv,” and the final --seed is a command for maintaining compatibility with pip, setuptools, and wheel.
Finally, open webui-user.bat with Notepad or similar and enter --uv in the set COMMANDLINE_ARGS= line.
Loading Existing Models (Optional)
If you already have the original Forge or A1111 WebUI installed in your environment, you can save disk space by sharing models.
The sharing method is to add commands to the set COMMANDLINE_ARGS= line in the webui-user.bat mentioned earlier. As an example, sharing checkpoint models is done as follows. ⚠️ Please replace the USERNAME with your own username.
--ckpt-dirs "C:\Users\USERNAME\webui_forge_cu121_torch231\webui\models\Stable-diffusion"This example is for when Forge is installed in %USERPROFILE% (the root directory of your username).
Common commands for model sharing are as follows.
- --ckpt-dirs: Specifies the path to checkpoint models.
- --lora-dirs: Specifies the path to LoRA.
- --vae-dirs: Specifies the path to VAE.
- --text-encoder-dirs: Specifies the path to text encoders.
- --embeddings-dir: Specifies the path to embeddings.
- --esrgan-models-path: Specifies the path to the upscaler model ESRGAN.
- --forge-ref-a1111-home: For those who find the above commands cumbersome and already have the original Forge, you can add all of them at once.
- --forge-ref-comfy-home: Use this to share models from ComfyUI.
✅ If you are already running WAN 2.2 in ComfyUI and want to try it in Forge-neo, you can easily share it by adding the following command. This example is for the ComfyUI portable version. ⚠️ Please replace the USERNAME with your own username.
--forge-ref-comfy-home "C:\Users\USERNAME\ComfyUI_windows_portable_SA\ComfyUI"This example is also for when ComfyUI is installed in %USERPROFILE% (the root directory of your username).
Installing Attention Mechanisms (Optional)
If you want to install SageAttention or similar, add the --sage command to the set COMMANDLINE_ARGS= line.
Also, if your GPU is an RTX 30 series or higher, adding --cuda-malloc is also recommended.
✅ If you want to know all available commands, you can see the full list by running the following command in the root directory of “Forge – Neo.”
venv\Scripts\python.exe launch.py --helpLaunching Forge-neo
Once you have come this far, simply run webui-user.bat as with the original Forge to launch it. ✅ The first launch will take time as packages are installed, but thanks to uv, it is significantly faster than conventional methods.
After a while, the UI will start up. You can basically use it the same way as the original Forge.
Running WAN 2.2 with Stable Diffusion WebUI Forge – Neo
Let’s actually run WAN 2.2 in Forge-neo. At this time, T2V 14B and I2V 14B are supported, while TI2V 5B, Fun, and S2V are not supported.
Generation Examples
T2V 14B Generation ExampleIn actual use, the operation is simple and easy to handle, but compared to ComfyUI, detailed settings options are more limited, so generation quality is slightly inferior. Also, Hires. fix and ADetailer cannot be used.
Now let’s look at how to use it.
Summary
Stable Diffusion WebUI Forge – Neo is a lightweight WebUI that allows anyone, even those who find ComfyUI’s complex operations difficult, to easily work with the latest video and image generation models. Let’s review the key points of this article.
- Forge – Neo supports a wide range of the latest models including Flux.2-Klein and WAN 2.2, and runs stably even on RTX 50 series GPUs.
- Using the uv package manager greatly reduces setup time.
- With a single command such as
--forge-ref-comfy-home, you can directly share existing Forge or ComfyUI models, saving disk space as well. - Both T2V and I2V of WAN 2.2 can generate videos in approximately 7 minutes using Lightning 4Step LoRA (in an RTX 3090 24GB environment).
- Even with limited VRAM, lowering the resolution allows support for GPUs with less VRAM.
- WAN 2.2’s operation is simple and has a lower learning curve than ComfyUI, but Hires. fix and ADetailer cannot be used.
Even if you are already running WAN 2.2 in ComfyUI, give Forge – Neo a try if you want to experiment easily through a GUI.
Thank you for reading to the end.
If you found this even a little helpful, please support by giving it a “Like”!