My Simple Guide: Getting Your Cool AI Art Models into AUTOMATIC1111 WebUI!

Hey everyone! So, you've seen all those amazing AI images online, right? And you probably downloaded the AUTOMATIC1111 WebUI to try it out. But then you download a fancy new model, and you're like, "Uh, where does this go?"

No worries! I'm gonna show you the super easy steps to get those new AI art models working in your WebUI. We'll also break down some of the techy words so you know what's what. Let's dive in!

For this guide, I'll be using a fantastic newer model: the Stable Diffusion 3.5 Large model. It's great for detailed and realistic images, so it's a perfect example to walk through!

💡 Quick Note: To download the AUTOMATIC1111 WebUI repository, you'll need to use git: git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

---

First, a Quick Chat About "Models" (and what all those other words mean!)

Think of an AI Model like a super-smart brain for your AI. It's basically a giant file (usually a few gigabytes!) that has learned how to turn your words into pictures. Different models learned from different kinds of pictures, so some are great at realistic photos, others at anime, and so on.

You'll usually grab these models from cool sites like Civitai.com or Hugging Face.

Key Terms You Need to Know

Stable Diffusion (SD): This is the main AI software we're using. It's the magic behind the image-making!

AUTOMATIC1111 WebUI (or just "WebUI"): This is that cool program you run that opens up in your web browser. It's like the easy-to-use control panel for Stable Diffusion, so you don't have to type in complicated code.

Checkpoint / Base Model: These are just different names for the big, main brain file we talked about. This is what generates the whole image.

• SD 1.5: This is like the classic version of Stable Diffusion. Still good, but a bit older. Works best for smaller pictures (like 512x512 pixels).
• SDXL (Stable Diffusion XL): This is the newer, beefier version! It makes much better-looking pictures, especially bigger ones (like 1024x1024 pixels), and generally understands your prompts way better. If you want super realistic stuff, SDXL is your friend.
• SD 3.x (Stable Diffusion 3.x): This is the latest and greatest from Stability AI. Models like SD 3.5 Large (which we'll use in this guide!) are super powerful, excel at prompt understanding, and are fantastic for both realistic and artistic images. They push the boundaries of what's possible!

Finetune / Finetuned Model: Imagine taking that "brain" model and sending it back to school for a very specific subject – like "How to Draw Super Realistic Anime Girls" or "Only Make Amazing Landscape Photos." That's a finetuned model! These are what give you those awesome, specific art styles you see online.

Safetensors (.safetensors): This is the BEST file type for your models! It's super safe and helps make sure no weird, hidden code tries to mess with your computer. Always try to download these if you can. (The older .ckpt files are okay, but safetensors are better.)

LoRA (Low-Rank Adaptation): Think of LoRAs like little "add-on packs" for your main model. They're tiny files that teach your model just one specific thing, like how to draw a certain character, a unique outfit, or a cool art style. You use them with your main model, not by themselves.

VAE (Variational AutoEncoder): This is a fancy part of the AI that helps with colors, contrast, and making your images look super clear and not washed out. Some models come with it built-in, but sometimes adding a separate VAE file can make your pictures pop!

Prompt: Your text instructions to the AI! Like, "a happy dog playing in a field of flowers."

Negative Prompt: The opposite of the prompt! What you don't want to see. Like, "ugly, blurry, extra limbs, bad quality."

💡 Pro Tip: You can check out PromptHero.com to see examples of positive and negative prompts you can use to generate pictures (but make sure to choose the correct model when browsing examples!)

txt2img (Text-to-Image): This is the tab where you just type words, and the AI makes a picture. Easy peasy!

img2img (Image-to-Image): This tab is where you give the AI an existing picture as a starting point, plus your words, and it transforms the picture! Great for stylizing photos or fixing up a sketch.

ControlNet: This is a super powerful "add-on" that gives you insane control! You can give the AI an image and tell it, "Hey, copy the pose from this pic," or "Use these exact outlines," or "Match the style of this artwork." It's mind-blowing! (You'll need to install this as an extension).

Virtual Environment (venv): This is just a special, isolated little space on your computer where the WebUI keeps all its Python parts. It's like its own little sandbox, so it doesn't mess with other stuff on your PC.

---

Okay, Let's Get That Model Working!

Once you've downloaded that awesome .safetensors model file (or .ckpt if that's all you got), here's what to do:

Step 1: Find Your WebUI Home Base

First, you gotta find where you put your AUTOMATIC1111 WebUI. It's usually the folder where you unzipped everything, or where you see that webui-user.bat file.

Mine's usually something like: C:\stable-diffusion\stable-diffusion-webui\

Step 2: Drop That Model File In!

Now, for the main event! Inside your WebUI folder, there's a special place for models:

1. Go into your WebUI folder, then click on models, and then Stable-diffusion.
   So, it'll look like: [Your WebUI Folder]\models\Stable-diffusion\
   (Example: C:\stable-diffusion\stable-diffusion-webui\models\Stable-diffusion\)

2. Copy or move your downloaded .safetensors (or .ckpt) model file right into this Stable-diffusion folder.

For this guide, I've already downloaded my SD_3_5_Large.safetensors file and will place it here!

Pro Tip for SDXL/SD 3.x: Even if it's an SDXL or SD 3.x model, it usually goes right into this Stable-diffusion folder. Don't worry about making a separate folder – the WebUI knows what to do!

Step 3: (Optional but Smart) Add a VAE (if you have one)

Some models might look even better with a separate VAE file. If the model's download page said "use this VAE," or if you have a general good one (like vae-ft-mse-840000-ema-pruned.safetensors), pop it here:

1. Go to: [Your WebUI Folder]\models\VAE\
   (Example: C:\stable-diffusion\stable-diffusion-webui\models\VAE\)

2. Drop your .vae.safetensors file into this VAE folder.

Quick heads-up: Many new SDXL and SD 3.x models already have the VAE built right in, so this step might not be necessary. If your pictures look a bit dull or washed out, trying a separate VAE can often fix it!

Step 4: Give Your WebUI a Fresh Start!

For the WebUI to notice your new model, you gotta restart it.

1. Close any black command prompt windows you have open from running the WebUI.
2. Go back to your main WebUI folder (C:\stable-diffusion\stable-diffusion-webui\).
3. Double-click webui-user.bat (or webui.bat if that's what you use).
4. Just chill for a minute while it loads up in your browser.

Step 5: Pick Your New Brain!

Once the WebUI is back up and running in your browser:

1. Look for the "Stable Diffusion checkpoint" dropdown menu. It's usually at the very top-left.
2. Click that little arrow, and voilà! You should see the name of your new model, including our SD_3_5_Large.safetensors file!
3. Click on your new model to select it. The WebUI will load it up – might take a few seconds.

Step 6: (Optional) Select Your VAE (if you added one)

If you added a separate VAE file in Step 3:

1. Go to the "Settings" tab in the WebUI.
2. Click on the "Stable Diffusion" sub-tab on the left.
3. Look for the "SD VAE" dropdown menu.
4. Pick your VAE file from the list.
5. Don't forget to click the "Apply settings" button!

---

Now, Let's Talk About Those Fun Dials and Buttons!

Once your model is loaded, you'll see a bunch of sliders and dropdowns. These are super important for tweaking your results!

Sampler (Sampling Method)

What it is: This is like the "drawing technique" the AI uses to turn the jumbled noise into your final image. There are tons of different ones (Euler A, DPM++ 2M Karras, DPM++ SDE Karras, etc.).

Why it matters: Different samplers can give you different looks – some are faster, some produce more detailed images, and some have a unique "feel." There's no single "best" one, it depends on the model and what you like!

Tip: DPM++ 2M Karras and DPM++ SDE Karras are very popular for high-quality, realistic results. Euler A is fast and good for quick tests.

Sampling Steps

What it is: This is how many "steps" or iterations the AI takes to draw your image. Think of it as how many times the AI refines its drawing.

Why it matters: More steps generally mean more detail and a more polished image, but it also takes longer. You usually hit a point of diminishing returns where more steps don't make much difference.

Tip: For most models, somewhere between 20-30 steps is a good balance for speed and quality. For SDXL and SD 3.x, 25-35 is common. Going higher than 50 often just wastes time.

CFG Scale (Classifier Free Guidance Scale)

What it is: This controls how strongly the AI follows your text prompt.

Why it matters:

• Low CFG (e.g., 3-6): The AI will be more creative and "interpretive" of your prompt. The results might be surprising but sometimes less accurate to your exact words.
• Medium CFG (e.g., 7-10): This is the sweet spot for most general generation. The AI follows your prompt well but still has some creative freedom.
• High CFG (e.g., 11-15+): The AI will stick very, very closely to your prompt, almost "forcing" the image. This can sometimes lead to less artistic or more distorted results if the prompt is too restrictive.

Tip: Start around 7 for SD 1.5 models and 5-7 for SDXL and SD 3.x models. Adjust up or down based on how much freedom you want to give the AI.

Seed

What it is: This is like a unique "starting point" number for the AI's random noise. Every image generation starts from a random pattern of noise, and the seed determines that initial pattern.

Why it matters: If you use the exact same seed, prompt, negative prompt, model, and settings, you will get the exact same image every time.

Tip:

• If you want to create variations of an image you liked, copy its seed!
• A value of -1 (the default) means the AI will pick a random seed for each new image, giving you unique results every time.
• You'll see buttons next to the seed like a "recycle" icon (for the previous seed) and a "dice" icon (for a random seed).

Scripts

What it is: This is a powerful dropdown that lets you run special actions or processes on your image generation. They're like mini-programs built into the WebUI.

Why it matters: Scripts can do cool things like:

• X/Y/Z Plot: Generate a grid of images, testing how different settings (like CFG Scale or Sampler) change your output.
• Prompts from file or textbox: Take a list of prompts and generate images for each one automatically.
• Loopback: Feed an image back into img2img multiple times to iteratively refine it.
• ControlNet: (Yes, ControlNet can also be accessed via a "Script" if you prefer that method, though it has its own dedicated section now).

Tip: Don't be afraid to explore these once you're comfortable with the basics! They can really speed up your workflow.

---

By understanding these settings, you'll have even more control over the amazing art you create! Go forth and generate!