|

7 Best SDXL Models (Compared)

When Stability AI launched its first text-to-image generation model, many users trained it further to create their own models to improve or customize the model to be trained on various image styles. 

That’s how we got so many Stable Diffusion models that were not only better than the base Stable Diffusion model but also trained on specific styles. 

Now, Stability AI has officially launched SDXL and there are already so many new models based on it. 

In this article, I’ll be sharing the best SDXL models you can use to experience the full power of the new SDXL model but with more refinements and training. 

Since SDXL is still new, there aren’t a ton of models based on it yet. So, I’ve kept this list small and focused on the best models for SDXL. 

However, as new models emerge, I’ll be updating this list and adding newer models that are worth mentioning here. 

That being said, let’s get started. 

What Is SDXL (Stable Diffusion XL)?

Before we get to the list of the best SDXL models, let’s first understand what SDXL actually is. 

SDXL or Stable Diffusion XL is the next iteration of the open model built by the Stability AI team. This model significantly improves over the previous Stable Diffusion models as it is composed of a 3.5B parameter base model. 

Unlike the previous Stable Diffusion 1.5 model, SDXL is well-tuned for vibrant colors, better contrast, realistic shadows, and great lighting in a native 1024×1024 resolution. 

The images being trained in a 1024×1024 resolution means that your output images will be of extremely high quality right off the bat. Whereas the Stable Diffusion 1.5 model was trained on a 512×512 aspect ratio.

Another reason why SDXL is a far superior model is its ability to generate legible text. That wasn’t the case in the previous models as they weren’t able to generate text. 

Lastly, with SDXL, you no longer need long complex prompts in order to generate high-quality images. 

Often with Stable Diffusion 1.5, you’d have to use words such as ‘masterpiece’, ‘realistic’, ‘high-quality’, etc to ensure your images come out looking good. That’s not the case for SDXL as it understands short prompts well. 

Related: Stable Diffusion vs Dall-E: Which One Is Better?

Best SDXL Models

Here are the best models for Stable Diffusion XL that you can use to generate beautiful images. 

Stable Diffusion XL Base

This is the original SDXL model released by Stability AI and is one of the best SDXL models out there. All the other models in this list are trained on this model. 

This SDXL model is a two-step model and comes with a base model and a refiner. The base model is used to generate the desired output and the refiner is then used to add more detail to the generated image. 

The model can generate several artistic styles and produce stunning images. 

Key Features: 

  • Can generate several art styles
  • Doesn’t achieve perfect photorealism
  • No need for lengthy prompts
  • Can generate legible text

RunDiffusion XL

RunDiffusion XL is a model based on SDXL and is more geared toward generating photorealistic humans, animals, and cars. 

This model is very good at generating portraits or shots that are up close. It’s also better at generating hands and eyes compared to the main SDXL model. 


While the model is still in beta version, the images it generates are pretty impressive. The images have high detail and look very realistic.

You can use the standard SDXL refiner with this model. 

Key Features: 

  • Good for portraits and close-up shots
  • Realistic images
  • Better hands and eyes

DreamShaper XL

DreamShaper XL is one of my favorite SDXL models as it doesn’t require you to use any refiner with it. You can just do a hires fix while generating images with this model. 

The model is good at generating pretty much everything. The images are highly detailed with less blurry edges and more realistic-looking people.

I find the images generated with this model the best compared to most of the SDXL models I’ve seen. 

Key Features: 

  • No refiner required
  • Less blurry edges
  • Generates good-looking people

Copax Realistic XL

Copax Realistic XL is another great model for Stable Diffusion XL and can generate very good realistic images. 

This is the v3 of the model which indicates the author is frequently training and updating the model. There’s also a colorful version of this model available to generate images in a colorful style. 

It’s still not perfect when it comes to generating eyes but it’s still quite an improvement compared to the base SDXL model. 

Key Features: 

  • Generates realistic images
  • Good at different poses and styles
  • Perfect for portraits 

Mysterious

Mysterious is an SDXL model tailored to generate fantasy-style images in Stable Diffusion. This model leans towards generating images of Asian cultures, urban settings, ancient times, etc. 

If you like fantasy images, this SDXL model is something you’d like to tinker around with. However, I found the images generated to be a bit too cartoonish or sharp in detail. 

Key Features: 

  • Perfect for fantasy images
  • Leans toward Asian culture, ancient and urban styles
  • Can also generate images in a Cyberpunk style

ShikiAnimeXL

ShikiAnimeXL is the best SDXL model for anime images as it generates stunning anime and cartoon images. 

The model is still in the trial version so don’t expect much from it. But from my tests and the images shared by the community, this model is definitely worth a shot. 

But if you were to ask me if this model is better than various Stable Diffusion anime models, then I’d say no since I didn’t find too much of a difference between the images this SDXL model generates versus the images generated by Stable Diffusion 1.5 anime models. 

Key Features: 

  • Perfect for anime images
  • Well-trained
  • No need to use a refiner 

SDVN6-RealXL

This SDXL model is trained to generate beautiful faces or portraits in Stable Diffusion. The model has multiple versions to generate detailed faces, slim faces, and beautiful faces. 

The model doesn’t require you to use a refiner and a hires fix can be used instead. But it’s still a good choice to use a refiner with SDXL. 

Personally, I found the images generated from this model to be beautiful but they look too airbrushed for my liking. The details are either lost and when the face does have details, it’s often in the form of moles. 

Key Features: 

  • Multiple versions available
  • No need to use a refiner
  • Perfect for portraits 
  • Generates beautiful faces 

SDXL Models Comparison 

The SDXL models in our list are very good and like always, I love doing comparison tests to see how different they are from each other. 

While these models are still new, they generate pretty stunning and mind-blowing images. For my SDXL model comparison test, I used the same configuration with the same prompts. 

The test was done in ComfyUI with a fairly simple workflow to not overcomplicate things. I also used the refiner model for all the tests even though some SDXL models don’t require a refiner. 

Here are the configuration settings for the SDXL models test: 

Positive Prompt: 

(fractal cystal skin:1.1) with( ice crown:1.4) woman, white crystal skin, (fantasy:1.3), (Anna Dittmann:1.3)

Negative Prompt: 

blurry, blur, text, watermark, painting, anime, cartoon, render, 3d

Configuration: 

ParameterKey Value
Sampling MethodDPM SDE GPU Karras
Total Sampling Steps30
CFG Scale5
Width 1024
Height1024
Upscaler4x_NMKD-Siax_200k

The images were also upscaled to 4096×4096 in my ComfyUI workflow. However, the images shared below are 1024×1024 as the upscaled versions are very large in size. 

Here are the results of the SDXL model comparison test: 

Based on the results, I’m just in awe at the level of detail and intricacies in each generated image. 

Besides that, here are my key findings and summaries of the test: 

  • DreamShaper XL and Copax Realistic XL models generated the best eyes among all the other models. 
  • The face looks too airbrushed in all the models but this could be because of the prompt rather than the models. 
  • The image generated by Copax Realistic XL looks the most realistic of all. The SVDN-6RealXL model comes a close second. 

Honestly, it’s hard to find any negatives about the images generated by these models as all of them look stunning. 

With these results, I’d definitely recommend trying each one of these models at least once. But if you haven’t tried any SDXL model, you should start with the main SDXL model first. 

Related: Stable Diffusion Male Models

How To Install SDXL Models

Installing SDXL models in Stable Diffusion is very simple. Here are the instructions on how to install SDXL: 

Step 1. Download SDXL Models

The first step is to download the SDXL models from the HuggingFace website. 

Here are the models you need to download: 

For both models, you’ll find the download link in the ‘Files and Versions’ tab. Click on the download icon and it’ll download the models. 

Related: Best SDXL Model Prompts

Step 2. Move The Models In Your Stable Diffusion Directory 

Once your models are downloaded, you need to move them into your models in your Stable Diffusion directory. 

Depending on the Stable Diffusion UI you’re using, this directory can be different. Paste both the model files in the following directory in your Stable Diffusion folder. 

For Automatic1111 WebUI: 

/stable-diffusion-webui/models/Stable-diffusion

For ComfyUI: 

ComfyUI/ComfyUI/models/checkpoints

For InvokeAI: 

/invoke-ai/models/sdxl/main – For the main base model 

/invoke-ai/models/sdxl-refiner/main – For the refiner model

Note: I recommend using ComfyUI instead of Automatic1111 because it’s more optimized to handle SDXL models. Check out this guide to learn how to install and get started with ComfyUI and this guide for using SDXL in ComfyUI.

Step 3. Load SDXL Models In Stable Diffusion 

Now, you have to load the SDXL checkpoint models in Stable Diffusion. Depending on your web UI, this workflow can be different. 

But for this tutorial, I’ll be using Automatic1111 as the example as it’s the most common UI for using Stable Diffusion. 

Load your Automatic1111 UI for Stable Diffusion and select the SDXL base model from the checkpoints dropdown. 

This will load the SDXL base model in Stable Diffusion and now you can generate images with this model. 

For using the refiner in Automatic1111, you’ll have to send your output image to img2img and then switch to the refiner model and generate the image in img2img. 

Related: Does Stable Diffusion Need Internet? 

FAQs

Here are some frequently asked questions about the SDXL model. 

Is SDXL free? 

SDXL is an open-source model that is available for everyone for free. You can generate as many images as you want using Stable Diffusion XL.

How Do I Run SDXL? 

You can run SDXL by downloading the model and running it on any Stable Diffusion UI such as Automatic1111 or ComfyUI. 

What is the difference between SDXL and Stable Diffusion 1.5? 

SDXL is a model trained on a much bigger dataset and is trained on 1024×1024 images whereas Stable Diffusion 1.5 is trained on 512×512 size images. 

Conclusion

Despite being launched just a few weeks ago, SDXL already surprised the Stable Diffusion community with how powerful and good it is. 

And this is evident not only by the discussions in communities but also by the new checkpoint models being made for SDXL. 

I hope this list helps you choose the best SDXL model for your AI art journey. Hopefully, as there are more updates and developments into SDXL, we’ll see more models pop up in the community. 

And you best bet that I’ll be covering those models in this article and this blog in general.

Stable Diffusion Prompt Organizer

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.