Catgirl's Image Making Guide

197

439

by:@edgycatgirl

Learn to make images like I do!
The setup, how to find models and LoRAs, and how to make your first AI image.

1. MAIN REQUIREMENTS

You really only need two things to get started:

A somewhat powerful PC.
The most important part is your GPU. It's needs to be Nvidia and something like a RTX 2060 can do just fine, but around 12 GB of VRAM is ideal based on ComfyUI docs.
It should be obvious, but the stronger your GPU is, the better. I personally use an RTX 5070 Ti with 16 GB of VRAM.
Download and install Stability Matrix.
I'll call it SM for the rest of this guide.
So, what actually is Stability Matrix?
Stability Matrix is basically a manager and launcher for local AI image-generation programs. Instead of manually installing Python, dependencies, folders, and every individual program yourself, SM handles most of that annoying setup for you.
It can install, launch, and update programs like ComfyUI, while also helping manage things like your models and LoRAs.
Basically, Stability Matrix is the thing that keeps all your AI image-making shit organized.
You could install ComfyUI manually without it, but for a beginner I think SM makes the whole process way easier.
Follow the guide on GitHub to install it. It's super easy:
Download Stability Matrix

And that's basically it. :)

2. SETTING UP COMFYUI

What Is ComfyUI?

Before we install it, you should probably know what ComfyUI actually is.

ComfyUI is the program we're going to use to actually build and run our AI image-generation workflow.

Unlike some AI image generators that just give you a prompt box and a Generate button, ComfyUI uses a node-based interface.

Those boxes you'll see everywhere are called nodes. Each node handles a different part of making the image. For example, one node might load your model, another holds your prompt, another controls how the image is generated, and another saves the finished image.

Those nodes are connected together to create a workflow. Basically, nodes are the individual steps, and the workflow is all of those steps connected together.

It can look intimidating as when you first open it, but don't worry. You do not need to understand every node to follow this guide. I'm giving you my workflow later, so you won't need to build one from scratch.

Install ComfyUI with SM

Once SM is installed, click Packages, then click Add Package at the bottom.

Search for ComfyUI under the Inference section:

Click on it, then hit the big green Install button:

SM will handle setting up ComfyUI and the stuff it needs to run, which is exactly why we're using SM instead of doing all of this manually.

Launch ComfyUI

Once ComfyUI is installed, go back to Packages and hit Launch here:

This starts ComfyUI on your computer. There are a few different ways to open the web GUI from here:

Don't let the "web" part confuse you. You're still running ComfyUI locally on your own PC. The interface itself just opens in your web browser.

Once it opens, you should be brought to a mostly blank ComfyUI interface:

If you're staring at the giant empty grid wondering what the you're supposed to do with it, that's normal. That's where workflows come in.

3. LOADING MY WORKFLOW

As I mentioned above, a workflow is basically a bunch of ComfyUI nodes connected together that tell the program how to make your image.

ComfyUI can get extremely complicated if you want it to, but I'm not going to explain every individual node or walk you through building an entire workflow from scratch.

You don't need to know how to build the car before you learn how to drive it.

Instead, I've made my own workflow that you can download below:

Download My ComfyUI Workflow

Once you've downloaded the .json file, simply drag and drop it onto the grid inside ComfyUI. ComfyUI will read the file and automatically recreate all of the nodes and connections from my workflow.

You should now see something like this:

Missing Nodes?

You'll probably have some boxes outlined in red and receive an error notification like this:

Don't panic. This just means the workflow uses some custom nodes that you don't have installed yet.

Custom nodes are basically extra pieces of ComfyUI made by other developers that add features the base program doesn't have. Since my workflow uses some of them, you'll need those same nodes for everything to work correctly.

Click View Details, then install the missing node packs it lists:

Restart ComfyUI after installing the missing nodes. They may not work properly until you've restarted it.

You'll probably still have errors for missing models or LoRAs. That's normal too. The workflow knows which files I was using when I created it, but those files aren't magically included inside the tiny .json workflow file.

So now we need to get the actual models.

4. DOWNLOADING MODELS & LoRAs

This is where we'll get the actual models and LoRAs that ComfyUI uses to make your images.

What Is a Model?

A model, which you'll also sometimes see called a checkpoint, is basically the main AI that generates your image.

If ComfyUI is the machine we're using, the model is the part that actually knows how to turn your prompt into an image.

Different models are trained or fine-tuned to produce different kinds of results. Some are designed around:

Anime
Realistic images
Semi-realistic art
Illustrations
Specific aesthetics
And basically anything else people decide to train them for

The model you choose has a HUGE effect on what your finished image looks like.

What Is a LoRA?

A LoRA is much smaller than a full model and is usually loaded alongside your main model to influence it toward something more specific.

LoRAs can be made for things like:

Specific characters
Art styles
Outfits
Poses
Body types
Expressions
Concepts
And a bunch of other stuff

For example, you might use an anime model as your main checkpoint and then add a character LoRA to help it accurately generate a specific character.

Model = your main generator.
LoRA = extra instructions or knowledge you add on top.

You'll probably end up collecting a shitload of these. Don't say I didn't warn you.

Using the Stability Matrix Model Browser

This is another reason I had you install Stability Matrix.

SM has a shared model library that helps keep things like checkpoints, LoRAs, VAEs, and other model files organized instead of making you manually hunt through a bunch of different program folders.

You can use the Model Browser inside SM to search for and download a ton of this stuff directly. I personally get most of mine from CivitAI.

Installing WAI Anima Manually

The one annoying exception is the main model I use, WAI Anima.

For whatever reason, it doesn't currently give us a nice download option through the SM Model Browser. :(

So for this one, you'll need to manually download three files:

The main WAI Anima model
The text encoder
The VAE

Download all three from the WAI Anima page on CivitAI:

WAI Anima on CivitAI

Where Do the Files Go?

Once all three files are downloaded, they need to go into the correct ComfyUI model folders.

Your exact Stability Matrix folder location may be different from mine. Wherever you installed SM, find your ComfyUI package and follow the same folder structure below.

WAI Anima Model
waiANIMA_v10Base10.safetensors
C:\PATH_TO_SM\StabilityMatrix-win-x64\Data\Packages\ComfyUI\models\diffusion_models

Text Encoder
qwen_3_06b_base.safetensors
C:\PATH_TO_SM\StabilityMatrix-win-x64\Data\Packages\ComfyUI\models\text_encoders

VAE
qwen_image_vae.safetensors
C:\PATH_TO_SM\StabilityMatrix-win-x64\Data\Packages\ComfyUI\models\vae

Once they're in the correct folders, refresh or restart ComfyUI.

These two nodes should now be fixed:

LoRA Install Made Easy

Okay, now the annoying part is out of the way.

We can use the Stability Matrix Model Browser to install the LoRAs used by my workflow. The four nodes shown below should still be red for you:

Head back to the Model Browser. You can filter the results by model compatibility — Anima in our case:

LoRAs Used in My Workflow

These are the four LoRAs I use in the workflow and the ones you'll want to find:

Downloading Them

For example, to install the detail LoRAs, search for detail in the Model Browser and find the one shown below:

Once you open it, you'll see that it contains multiple files. We need both the Anima Detailer and the Background Detailer, so download both:

Then repeat the same basic process for Textify and Light Concepts:

Once all four LoRAs are installed, refresh or restart ComfyUI. The red errors on those LoRA nodes should now be gone.

And now you're finally ready to make your first image.

5. MAKING YOUR FIRST IMAGE

Once your models and LoRAs are installed, you're ready to actually make something.

At this point you should have:

Stability Matrix installed
ComfyUI installed through Stability Matrix
My workflow loaded
The required custom nodes installed
WAI Anima, its text encoder, and its VAE installed
The four required LoRAs installed

Now we can finally go through the parts of the workflow you actually need to care about.

You'll notice I'm not going to explain every single technical thing ComfyUI is doing behind the scenes. You don't need to. Learn what you actually need to make images first. You can always dig into the nerd shit later.

The Prompts

The prompt nodes are right in the middle of the workflow.

Green is your positive prompt — the stuff you want in the image.

Red is your negative prompt — the stuff you don't want in the image.

You can edit both of these however you want for your own images. I'm not going to teach you prompt best practices because if I know 'em yet.

Using Anima's Built-In Styles

One little tip: Anima has a bunch of built-in artist styles you can use in your prompts. I personally use @ratatatat74 a lot.

To use one, just add the artist tag to your positive prompt. For example: @artistname

There are a ton of different styles available, so mess around with them and find one you like. You can browse them using the Anima Style Explorer below:

Browse the Anima Style Explorer

Generate Your First Image

Once everything is set up, hit Run in the top-right corner of ComfyUI.

If you did everything correctly and haven't changed the prompt or settings yet, it should generate the same image of my OC shown at the top of this post.

If you got an image, congrats — the hard part is over.

Saving Your Images

To download an image directly from ComfyUI, click Assets on the left side. Find the image you want, right-click it, and download it.

Keep in mind: the images shown in the Assets panel may disappear from there when you restart ComfyUI.

Don't worry — the actual image files are still saved on your computer.

You can find the images generated by this workflow here:

C:\PATH_TO_SM\StabilityMatrix-win-x64\Data\Packages\ComfyUI\output\anima

Your exact path may be different depending on where you installed Stability Matrix.

6. TRY HARDER!

I'm going to show you how I do things.

I'm not going to hold your hand through every possible problem you might run into.

This guide should give you everything you need to get started, but AI image generation is one of those things where you're going to learn way more by actually doing it than by reading a guide.

You're going to install something and have it break. You're going to download a LoRA that looks like shit. You're going to spend twenty minutes trying to figure out why ComfyUI is throwing an error only to realize you forgot something stupid.

You're going to make an image that looks absolutely cursed.

Good. That's how you learn.

Google the error. Read the GitHub page. Look through Reddit. Search CivitAI. Watch a tutorial. Read what the model creator actually wrote instead of blindly downloading shit and wondering why it doesn't work.

Try different settings. Change your prompt. Lower the LoRA strength. Raise it. Try another model. Look at what other people are doing and figure out why it works.

And when something doesn't work, don't immediately give up because there isn't a step-by-step guide telling you exactly what button to press next.

Experiment. Research. Fail. Learn. Try again.

There isn't one magical workflow, model, prompt, sampler, or set of settings that's going to make every image perfect.

The people making really good stuff didn't get there by copying one workflow and never touching anything.

They fucked around with it. A lot.

So use this guide as a starting point, then start figuring shit out for yourself.