Prefer to Watch a video on how to install ComfyUI and ComfyUI Manger? See this video:
Download ComfyUI: https://github.com/comfyanonymous/ComfyUI
Press Ctrl + F and type in Direct Link to find the download link and click on it.
2. Extract the folder to the location where you want to install ComfyUI.
Right Click on your downloaded file and choose "Extract to" and put it anywhere on your PC. If you can't extract it see the 7-Zip link in the picture above to download 7-Zip extractor.
3. Install GIT https://git-scm.com/download/win.
Get the standalone version and most of you should have x64 Bit Windows, if you're not sure press the windows Key + Pause at the same time to see your system info to find your version. When installing just accept all the defaults and click next until you're done. You will not need to open GIT by itself it will work through command prompt.
4. Download ComfyUI Manager: https://github.com/ltdrdata/ComfyUI-Manager
First Step is to Copy the link on the Manager site by clicking the green button labeled "Code" and then click on the double squares "copy" icon to copy the link to the repository.
5. Use GIT to extract your ComfyUI Manager into your custom nodes by going into your ComfyUI/Custom Nodes folder.
i.e. Mine is in E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\custom_nodes
Type in CMD in the Address Bar and Press ENTER and a Command Prompt will pop up that will change our directory exactly where we need to be:
Type CMD and Press ENTER
6. Now Type in git clone and press Ctrl + V or right click into the command window to paste the address we copied in step 4. It should look like this when your done. Now Press ENTER to execute.
once you execute Git will install and extract a bunch of stuff. I bet your glad you installed it.
7. After It's done you will see the ComfyUI-Manager inside your Custom nodes folder. You will also have a manager button in your ComfyUI and if you don't restart your comfyUI completely if it was running.
8. Click on run_nvidia_GPU to run ComfyUI and if you don't have an Nvidia Click on Run CPU but you will go 10x slower.
If you don't see a workflow just click on load default. The manager button will be at the bottom right. Click on it and Select update all.
It will then update, ask you to restart, Confirm you want to exit and reconnect, and open up a new tab with the refreshed comfyui
Congrats! You installed ComfyUI and the manager.
However...
You will still need Models to make this thing work. Let's keep it simple and use a SDXL Model that doesn't require separate encoders. In plain English that means you're just downloading 1 file.
Flux models are the best models, but they are slow and require a very good graphics card. So, let's just use the next best thing which is an SDXL Model (SD3 is not great right now) called RealVisXL 5.0 Lighting which you can find on CivitAI. Here is the link.
This is a custom model and one of the best.
Pay close attention to the green and red text at the bottom. This is what settings you will use. Lighting models will utilize DPM__ Karras/DPM ++ SDE Sampler and 4-6 steps and CFG Scale of 1-2. It will all make sense in a second.
Go into your default downloads folder and Cut and Paste your download (there is no need to extract it or open this file)
Cut and Paste into ComfyUI\models\Checkpoints i.e. my folder is E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\models
As you can see my Model Made it into its correct location
Now that your model is in it's correct location and we have at least one working model. Let's Start up ComfyUI.
Find the run_Nvidia_gpu file in your folders.
Mine is located in here: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable
Note 1: if you don't have nvidia you can click run_cpu but it will be very slow. Even if you have a faster AMD Graphics Card. For AMD users there are work arounds. See official Github page for comfyui for more info near the bottom of the page.
Note 2: Before you do anything Click on the Manager button and select "UPDATE ALL" to make sure you have all the latest updates to avoid hidden bugs or problems.
Note 3: You can also click on Load default for the most basic ComfyUI Workflow if it doesn't pop up when you open up the tool.
If everything went well you should be looking at this screen:
Let me explain what you're looking at with a picture.
Load Model (click in the middle of the text to get a list of the models you have in your checkpoints folder.
TIP: Navigation. Click on an empty space in ComfyUI (Not on a node) and drag around the canvas to get around or you can click the middle mouse button.
TIP: Scroll up and down to zoom in and out.
TIP: Double Click on the canvas to add a new node or note or add a background to help identify a group of nodes and move them all at the same time.
Right click on a node to change it's color or get more options. Colors don't change the function of a node.
(PROTIP: the loader just checks inside the checkpoints folder for ckpt or safetensor files so you can create subfolders i.e. SD3, SDXL to manage your models neatly and it will still find it. Also, when using a new workflow, you can use the manager to install all missing nodes, but the final missing piece of the puzzle is finding a file that goes in all the nodes that say "LOAD or LOADER" like this one. Just match the model or find one to replace it.
Type what you Want in the positive prompt and What you don't want in the negative prompt (AKA CLIP Text Encoders). PROTIP you can right click on each node and change the color of the node to help you ID things faster. I made the positive prompt on the top Green and the Negative one on the bottom red.
Change your Settings (ksampler): You can hover over each item to see a description of it. go to the model's official website to see what the best steps and CFG are. For lighting models it recommend DPM++ SDE Karras/DPM ++ SDE Samples, 4-6 steps and a CFG Scale of 1-2 so I made my settings as such.
Info from REALVISXL's Website on what to use.
Hyper, Turbo and Lighting Models are popular for their 1-3 second generations but you will have to play with the CFG and the Steps to get each model just right for the pic your trying to generate. Typically you pair it with a highrez fix and don't use it by itself on a normal workflow.
Last but not Least! Click Queue to Generate:
You can click on view queue to see how many are running or waiting to run
The Top right is your Output and the pictures at the bottom is your output history or sequence of outputs for the workflow. You can scroll up and down when inside that box.
PRO TIPS:
TIP:
The vanilla workflow will not produce great results especially with the faster lighting models. These are normally paired with HiRez fixes and other tools to refine the outputs. you can find workflows on these sites to help you do that.
When you first open a workflow there will be a lot of red nodes and warnings of missing nodes: Go into manager and click on install all missing nodes to fix that. Then go to each loader and see what's currently in the loader. That is what you will have to find on hugging face and put into it's appropriate folder or replace it with an alternative to get it to work.
Look for a workflow of your choice. Since we have a lighting model we will have to find a workflow that works with SDXL Lighting or use a basic SDXL workflow
Want to Learn how to install and use Flux in ComfyUI? See this video:
Want a Free Alternative to Midjourney or Easy way to create Ai Art without installing anything? Try PicLumen!
Comments