This part two of a series of articles I will write on this subject. This covers the walkthrough. You can find part one here:
https://civitai.com/articles/28131/workflows-a-beginners-tutorial-and-hands-on-walkthrough-part-1
Introduction:
I love building stuff. I'm an engineer and a carpenter and now build waterparks for a living. How things work and creating is my passion. I create workflows that get used by a lot of people and I get asked a lot of questions, especially about ComfyUi in general and how to build with them, so I thought I'd take the time to put together a few articles over the next few (weeks, month, however long) to help those underratnd the basics, the move on to more advanced workflows.
In this part, I will go through the following:
A hands on walkthrough a basic workflow.
downloading the models and required nodes
familiarizing yourself on basic node layout
using groups
adding, copying, & changing nodes
wiring fundamentals
Generating your first image
Along with some cool💡tips & tricks along the way.
Getting started:
⚠️⛔🛑✋ If you have not already read part one of this article, please do so before beginning.
Downloading Models:
Open up the workflow. On the right hand side, you will see a list of the models we are going to use and the folders that they go in.
Note: You can use any checkpoint model from SDXL, Illustrious, or Pony for this exercise. Just be aware that your image will be altered accordingly
You can also do this from the model manager in the Comfy Manager
Open the manager and search for the appropriate files
These will automatically be put in the correct folder
Custom Nodes:
At the top of the manager, click "Install Missing Custom Nodes" (i just used an example below. you do not need these ones)
install all nodes, then restart ComfyUi.
Note: If you are on Portable, you will also need to refresh the page.
The workflow:
Looking around
Take a moment to look around at the workflow. Refer back to part one of this article if you need clarification on what each part is.
for practical purposes, most workflows are laid out from left to right.
if you look closely at a node, you will see the input is on the left, and the output is on the right. In this case, the clip goes into the encoder, it reads the prompt, and the results (conditioning) come out of the right side
Notice how some nodes are purple and hard to read
These nodes are Bypassed. That means they are not used or the informaton passes through them without it being used. right click on one of them
You can bypass or unbypass this here.
You can also hover over a node (or group) and hit ctrl+b to bypass or unbypass it.
⚠️Make sure you leave the node bypassed for now
Go to the "LoRA loader" section and click on the blue square around them. This is a group.
hold down the left mouse button and drag the group around. Notice how they all move together? This eases organization
Now click on one of the loaders and drag it out of the square. It is no longer part of that group.
Now delete the blue box around the nodes by just clicking on it and hitting delete on your keyboard
Now hold down control and your left mouse button. Drag a square around the nodes and let go. Both nodes shoudl have a whit dashed line around them
right click (anywhere but on top of the nodes or in the dashed lines) to display the secondary menu
Click "Add Group For Selected Nodes". The group will reappear
💡double click on the name at the top of the group to rename it.
You can also have groups within groups. This is useful for when you want to have a group of nodes together, but do not want to bypass all of them at the same time.
Copying, adding, pasting nodes:
Lets begin in the upper lefthand corner of the workflow where the Load Checkpoint is:
As you can see, I have deleted the reroute for the VAE. Let's add it back in. There are several ways to accomplish this
Hover over where you want the reroute to go and double click, a menu will appear.
begin typing in "reroute" it will appear in the list. then click on it.
depending on how updated your comfy is, it will either appear as a floating block under your cursor for you to set, or will just set itself.
Set it where you want it
Don't worry about the color or name or what it looks liek for now.
Delete the reroute you just created.
click on the "clip" reroute
now hold down the 'alt' key on your keyboard and drag down. A duplicate copy of the node should follow
💡on other workflows, the nodes are simetimes pinned, so it will create another pinned node right on top of it and it will not move. The "P" button is how you pin and unpin nodes/groups/etc.
Notice how the node is the same color and has the same name? Let's change that
when you click on the node and it is highlighted , this menu pops up. click on the yellow circle to display standard colors and choose one.
🤫 Pro tip: If you right click, go to colors, then go to custom at the bottom, you can create any color you want.
🤫🤫 Super pro tip. right click directly on the dot on the left of the reroute. This is how you name them.
Now do the same thing for the setnodes below so that your final product looks like this:
⚠️Notice the name "Model" by the left side of the purple node and the * by the left side of the yellow 'Clip" node? Once a setnode is defined, it cannot be used for another purpose (in this case changing from model to clip). You can use ALT to copy on an undefined one, but not one one that is already dedined. You must create a new node.
Duplicating connected nodes.
hold the left mouse button and make the connections for the reroutes like this
now hold down ctrl and click on each seperately so they light up, or draw a box around them as you did wthe teh LoRA nodes.
hit ctrl+c
now move the cursor over several inches and hit shift+ctrl+v
You will see that the nodes duplicated themselves while the connection remained.
⛔ delete the added nodes
💡many times you will find that nodes are renamed in a workflow for convenience. If you are ever confused about a name of a node, right-click and at the top in a slighty darker area you will find the node name.
Creating the Workflow:
I've deleted the unnecessary node groups for simplicity. You should just move the group out of the way
Wiring the workflow
Lets start by connecting the Model to the LoRa loader. click and hold over the purple checkpoint dot and drag it to the reroute, the repeat for the other side of the reroute to the LoRa Loader
Try hooking the wire up to the clip. notice how it does not connect? only the proper wire will work on the proper connection.
Repeat this procedure with the clip. we will address the VAE connection a bit further down
make the remaining connections as follows:
The model from the output of the LoRa loader to the Ksampler
The clip from the LoRA loader to the positive and negative text encoders
The empty latent from the latent area to the latent connection input
Take the VAE from the checkpoint area we left unconnected and take it to the VAE decoder
hook up the latent from the ksampler to the VAE decoder, then hook up the image from the VAE decoder to the two image nodes
Note: You don't really need the preview one for this exercise. It is there for a later lesson, but hook it up anyway
Congratulations!!! You just hooked up a workflow!!! 🥳🎉🎊🎈🍹
Settings:
I will just list these out per area:
In the checkpoint, use SDXL (or another checkpoint model)
In the LoRa loader, place a LoRA. I used the "colorful burst" LoRa, but you can use anyone that is compatible with the base model. tehye get placed in the Comfyui\model\lora folder
Set both weights to around 0.8. This works as a good basis to start from.
Set the latent size to 1024x1024, and the batch to 1
Ksampler settings:
leave the noise and the control after generate as is, or set it to fixed
Set steps to 30
Set CFG to 5.0
Use Euler as a sampler
Use Simple as a scheduler
leave denoise at 1.0
Hit RUN!
Hopefully you get some version of my little friend here.
Next steps:
Mess around with the settings.
adjust the weights of the LoRA and see what happens to the image
change the steps
Try changing the sampler and scheduler
Try changing the CFG
use different prompts.
Summary:
I hope that is was a valuable article and that you enjoyed it. Please leave me a feedback.
I will be doing another article (or two) in the near future about advanced options including:
Image to Image
Upscaling
Lora chaining
Ksampler chaining
detailers
other stuff
Please comment and let me know what you think
Instagram: https://www.instagram.com/synth.studio.models/
Buy me a☕ https://ko-fi.com/lonecatone
This represents Many of hours of work. If you enjoy it, please 👍like, 💬 comment , and feel free to ⚡tip 😉