ComfyUi is a web-based application to generate images and videos using Stable Diffusion technology. It is a framework to integrate modules such as ControlNet, IP-Adapters, AnimateDiff to work together in one workflow. It allows you save and re-use the workflows that carry out complicated tasks. This guide provides the steps to download and install ComfyUI portable in Github.
ComfyUI consists of connected nodes. Nodes are building blocks of workflows. Some nodes come with the default ComfyUI installation. But most time, you need install customer nodes for certain modules, such as ControlNet, AnimateDiff etc. In this post, I will introduce how to install ComfyUI and its must-have customer nodes before you load any sample workflows from online sources.
Table of Content
- Install ComfyUI and ComfyUI-manager
- Install ControlNet in ComfyUI
- Install IPAdapter in ComfyUI
- Install AnimateDiff in ComfyUI
- Install AnimateDiff-Evolved in ComfyUI
1. Install ComfyUI and ComfyUI-Manager
1. Go to 7zip to download an installer for Windows. Run the installer.
2. Go to ComfyUI GitHub download page. Read the release notes and select ComfyUI_windows_portable_nvidia.7z file to download comfyui portable. Right click the file and use 7-Zip to extract it. The root directory is “ComfyUI_windows_portable.”
3. Go to “ComfyUI_windows_portable\ComfyUI” subfolder, rename “extra_model_paths.yaml.example” to “extra_model_paths.yaml.” Open a text editor to edit this file. Uncomment the section for comfyui. Change base_path to the path of your “ComfyUI_windows_portable\ComfyUI” directory.
4. Next is to install ComfyUI-Manager. It is an extension for you to install and remove various customer nodes and models. On Windows, the easy way is to copy the content of install-manager-for-portable-version.bat. Create a new file with the same name, paste the content you just copied, save the file. Then put this file at “ComfyUI_windows_portable” directory. Double click to run it. This will add “ComfyUI-Manager” folder at “ComfyUI\custom_nodes” directory.
5. Download Stable Diffusion models from Huggingface and civitai. For example, go to Counterfeit-V3.0. Download “CounterfeitV30_v30.safetensors” file. Copy this file to your “ComfyUI\models\checkpoints” directory.
6. Now let’s run ComfyUI. Under “ComfyUI_windows_portable”, there are two batch files, “run_cpu.bat” and “run_nvidia_gpu.bat.” Run one of them based on your Windows env. A console window will show the status of launching. Moments later, ComfyUI runs in your browser. The default url is http://127.0.0.1:8188/.
7. You can test your installation by running a workflow, which is a json file or a PNG image containing the parameters to generate images or videos. Go to ComfyUI_examples, and click Img2Img. Drag the screenshot image on that page to your ComfyUI in the browser.
8. In Load Image node, load an image from your local drive. Note the default input images are located at “ComfyUI\input.” In Load Checkpoint node, select a checkpoint that available in the checkpoints directory.
9. Click Queue Prompt button at the top of the right panel. The workflow starts to run. Monitor the progress in the ComfyUI console. You can also get the status by looking at the green edge or red/pink edge around a node in ComfyUI interface.
10. When it finishes, a new image is showing at Save Image node. The output image is also saved at “ComfyUI\output” directory.
2. Install ControlNet in ComfyUI
ControlNet is a group of neural networks that can control the artistic and structural aspects of image generation. It gives you more controls over images created by Stable Diffusion.
1. Open ComfyUI in a browser at http://127.0.0.1:8188/.
2. Click Manager button in the right panel. The ComfyUI Manager Menu opens. Click Customer Nodes Manager button, a window with a list of customer nodes pops up. In this window, search for “controlnet.” Select “ComfyUI’s ControlNet Auxiliary Preprocessors” and click Install.
3. When the installation finishes, click RESTART to restart ComfyUI.
4. Go to ControlNet v1.1 to download ControlNet models, such as “control_v11p_sd15_scribble.pth”, “control_v11p_sd15_openpose.pth,” “control_v11p_sd15_lineart.pth,” “control_v11p_sd15_softedge.pth,” “control_v11f1e_sd15_tile.pth,” etc. Put them at “ComfyUI\models\controlnet” directory.
5. To test the installation, go to ComfyUI_examples, and click ControlNet and T2I-Adapter. In this page, there are a few ControlNet examples. Drag the Scribble ControlNet workflow image to your ComfyUI in the browser. Meanwhile, download the scribble image beneath and save it at “ComfyUI\input” directory.
6. Let’s look at the nodes in the workflow. There are two nodes used for ControlNet (See the image above). One is Load ControlNet Model, the other is Apply ControlNet. In Load ControlNet Model node, select “control_v11p_sd15_scribble.pth.” If you don’t see any data from the dropdown, click Refresh button in the right panel to get the ControlNet models showing up.
7. To run the workflow, in Load Checkpoint node, select a checkpoint in your “ComfyUI\models\checkpoints” directory, for example, the “CounterfeitV30_v30.safetensors” you downloaded before. In Load Image node, click choose file to upload button to load the scribble image you downloaded.
8. Click Queue Prompt button to run the workflow. Monitor the progress in the ComfyUI console. Note during the run, there are a few annotators downloaded. They are saved at “ComfyUI\custom_nodes\comfyui_controlnet_aux\ckpts\lllyasviel\Annotators.”
9. When it finishes, a new image is showing at Save Image node. The output image is also saved at “ComfyUI\output” directory.
3. Install IPAdapter in ComfyUI
IP-Adapter, or image prompt adapter, is a stable diffusion add-on to generate images using image prompts. It can apply styles from your reference images.
1. Open ComfyUI in a browser at http://127.0.0.1:8188/.
2. Click Manager button in the right panel. The ComfyUI Manager Menu opens. Click Customer Nodes Manager button, a window with a list of customer nodes pops up. In this window, search for “ipadapter.” Select “ComfyUI_IPAsapter_plus” by cubiq and click Install. When the installation finishes, click RESTART to restart ComfyUI.
3. Go to ComfyUI_IPAdapter_plus GitHub page. Scroll down to Installation section. Download CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors and CLIP-ViT-bigG-14-laion2B-39B-b160k.safetensors. Rename the downloaded files to the names shown here. Put them at “ComfyUI\models\clip_vision” directory.
4. Download ip-adapter_sd15.safetensors and other ip-adapter models listed on the same page. Put them at “ComfyUI\models\ipadapter” directory.
5.(Optional) If you are going to work on face generation or face fixing, you need to continue to install insightface and download lora models by following the instruction in the installation section.
6. To test the installation, go to ipadapter_plus examples. In this page, there are many ipadapter json workflows files. Click ipadapter_simple.json and download it. Drag the file to your ComfyUI in the browser.
7. Let’s look at the nodes in this workflow. There are two nodes used for ipadapter (See the image above). One is IPAdapter Unified Loader, the other is IPAdapter.
8. To run the workflow, in Load Checkpoint node, select a checkpoint in your “ComfyUI\models\checkpoints” directory. In Load Image node, click choose file to upload button to load an image. Change the prompt to describe your image.
9. Click Queue Prompt button to run the workflow. Monitor the progress in the ComfyUI console.
10. When it finishes, a new image is showing at Save Image node. The output image is also saved at “ComfyUI\output” directory.
4. Install AnimateDiff
AnimateDiff is an AI video generator that uses Stable Diffusion along with motion modules. It provides text-to-image, camera movements, image-to-video, sketch-to-video. Here is the link of AnimateDiff paper.
1. Open ComfyUI in a browser at http://127.0.0.1:8188/.
2. Click Manager button in the right panel. The ComfyUI Manager Menu opens. Click Customer Nodes Manager button, a window with a list of customer nodes pops up. In this window, search for “animatediff.” Select “AnimatedDiff” by ArtVentureX and click Install. When the installation finishes, click RESTART to restart ComfyUI.
3. Go to download animatediff motion models, such as “mm_sd_v15_v2.ckpt,” etc. Put them at “ComfyUI\custom_nodes\comfyui-animatediff\models” directory.
4. On the same page, download motion lora models, such as “v2_lora_PanLeft.ckpt”, “v2_lora_PanRight.ckpt” … “v2_lora_ZoomIn.ckpt” and “v2_lora_ZoomOut.ckpt.” There are total eight motion lora models. Put them at “ComfyUI\custom_nodes\comfyui-animatediff\loras” directory.
5. Go to Stable Diffusion VAE model in Huggingface, download “vae-ft-mse-840000-ema-pruned.safetensors” and put it at “ComfyUI\models\vae” directory.
6. To test the installation, go to comfyui-animatediff workflows. In this page, there are a few animatediff workflow examples. Download simple.json.
7. simple.json is a txt2img workflow. Drag simple.json to your ComfyUI in the browser. There are a few nodes used for AnimateDiff (See the image above). They are: Animate Diff Module Loader, Animated Diff Sampler, and Animate Diff Combine. In Animate Diff Module Loader node, select a motion model, such as “mm_sd_v15_v2.ckpt.”
8. To run the workflow, in Load Checkpoint node, select a checkpoint in your “ComfyUI\models\checkpoints” directory. In Load VAE node, select the VAE model you just downloaded. Change the prompt to describe your video.
9. Click Queue Prompt button to run the workflow. Monitor the progress in the ComfyUI console.
10. When it finishes, a new video is showing at Animate Diff Combine node. The output image is also saved at “ComfyUI\output” directory.
5. Install AnimateDiff-Evolved
AnimateDiff-Evolved is an improved AnimateDiff integration for ComfyUI. It requires the installation of additional customer nodes such as ComfyUI-Advanced-ControlNet and ComfyUI-VideoHelperSuite.
1. Open ComfyUI in a browser at http://127.0.0.1:8188/.
2. Click Manager button in the right panel. The ComfyUI Manager Menu opens. Click Customer Nodes Manager button, a window with a list of customer nodes pops up. In this window, search for “Kosinkadink.” There are three custom nodes showing up. They are “ComfyUI-Advanced-ControlNet”, “AnimateDiff Evolved” and “ComfyUI-VidelHelperSuite.” Select all three and Install them. When the installation finishes, click RESTART to restart ComfyUI.
3. Go to download animatediff motion models, such as “mm_sd_v15_v2.ckpt” and “v3_sd15_mm.ckpt” etc. Put them at “ComfyUI\custom_nodes\ComfyUI-AnimateDiff-Evolved\models” directory.
4. On the same page, download motion lora models, such as “v2_lora_PanLeft.ckpt”, “v2_lora_PanRight.ckpt” … “v2_lora_ZoomIn.ckpt” and “v2_lora_ZoomOut.ckpt.” There are total eight motion lora models. Put them at “ComfyUI\custom_nodes\ComfyUI-AnimateDiff-Evolved\motion_lora” directory.
5. Go to VAE model page, download “vae-ft-mse-840000-ema-pruned.safetensors” and put it at “ComfyUI\models\vae” directory.
6. If you haven’t install ControlNet, go to install ControlNet in ComfyUI.
7. To test the installation, go to comfyui-animatediff-evolved samples. In this section, there are many animatediff-evolved workflow examples. Drag one of them, e.g “txt2img – 48 frame animation with 16 context_length,” to your ComfyUI in the browser.
8. Let’s look at the workflow. The animatediff nodes are hightlighed in red (see above image). In AnimateDiff Loader node, select a motion model, such as “mm_sd_v15_v2.ckpt.”
9. The Context Options node helps to extend the output video to more frames. In Latents = Animation Length node, the default batch size is 48, which means it will generate 48 frames.
10. To run the workflow, in Load Checkpoint node, select a checkpoint in your “ComfyUI\models\checkpoints” directory. In Load VAE node, select the VAE model you downloaded. Change the prompt to describe your video.
11. Click Queue Prompt button to run the workflow. Monitor the progress in the ComfyUI console.
12. When it finishes, a new video is showing at Video Combine node. The output image is also saved at “ComfyUI\output” directory.
Now you have ComfyUI and the most important customer models installed, you can load more complex workflows to test your Comfyui setup. You can also start to build your own workflow.
