Stable Diffusion XL Review: The Future of Creating Stunning AI Images

With the release of DALL-E 3, the race for the best AI image generator is now tighter than ever. What can Stable Diffusion XL offer that DALL-E and Midjourney can't? Find out now:

John Angelo Yap

Updated October 17, 2023

Reading Time: 9 minutes

Text-to-image generators is slowly becoming the key to unlocking a world of creativity and expression. All it takes is a couple of seconds to create stunning images from a single sentence. It doesn't matter if you're making thumbnails or your next masterpiece, imagination is your only limit.

And it's for everyone too! Whether you're a business owner, a content creator, or simply an enthusiast, AI image generators are here to transform your ideas into beautiful and (possibly) revenue-generating art.

So, in this comprehensive review, let’s dive into one of the most popular AI image generators: Stable Diffusion. We'll go over how it works, how much it costs, closest competitors, and more!

What Is Stable Diffusion?

As the name suggests, Stable Diffusion is a freemium text-to-image generator that creates stunning, detailed images from prompts. This latent diffusion model was developed by Stability AI and was officially launched on August 22, 2022.

Stable Diffusion XL Landing Page

Stable Diffusion has taken the internet by storm since its initial version. Unlike DALL-E and Midjourney, it's open-source, meaning that you're free to use, modify, or distribute their code legally. This is an unofficial invitation for developers to improve their model.

As of September 2023, Stable Diffusion has more than 10 million users.

So, What is Stable Diffusion XL?

Released in July 2023, Stable Diffusion XL or SDXL is the latest version of Stable Diffusion. As expected, it has significant advancements in terms of AI image generation.

SDXL adds more nuance, understands shorter prompts better, and is better at replicating human anatomy. You're now less likely to generate images with the infamous "uncanny valley" feeling.

Key Features

Advanced Text-to-Image Generation

Stable Diffusion's bread and butter is how well it can generate images from a single prompt and a couple of parameters. All it needs from you are the following:

  • Prompt: A description of the image you want.
  • Negative Prompt: Objects you don't want in the generated art.
  • Aspect Ratio: The ratio of the image's width to height.
  • Variations: How many versions of the image you want from a prompt.
  • Image: This is optional but, if you want an AI to slightly edit your image, you can also upload any JPG or PNG file to Stable Diffusion.

From there, all you need is creativity in crafting your prompt. It also helps to have some patience (or a lot) because I've noticed that Stable Diffusion isn't as accurate at understanding context as Midjourney or DALL-E

To test SDXL, I used the input “gold penguin smiling, wearing sunglasses, standing under an apple tree in summer.” I also used parameters such as Fantasy Art and 3:2 for my aspect ratio.  

Stable Diffusion Sample Images

Cute? Absolutely. But, as I previously mentioned, I'm underwhelmed with SDXL's nuance. Out of the four images, only the second and third one met all the requirements. The first one looks more like a baby duck and the penguins in the fourth one aren't gold. I wanted to see penguins :/

Regardless, you can still edit and download your outputs. You can edit to remove a background, clean up imperfections, relight, upscale, and reimagine.

Reimagine XL

You can use Stable Diffusion's Reimagine feature to generate variations of an image with a single click. No need to pay graphic designers or artists — you just need to drag and drop an image and let Stable Diffusion do its magic.

Stable Diffusion Reimagine XL Landing Page

You can use this feature through Clipdrop.co for free, as long as you don't exceed their daily limit of 400. If you need to create more and with higher resolution, it will cost you $7 per month.

So, enough talking, here's what SDXL returned when I asked the tool to create reimagined versions of the Gold Penguin I received in the previous section:

As you can see, SDXL maintains both intent and content of the input. What it changes is the artwork, which makes this feature perfect for slight tweaks to achieve the quality you need.

Negative Prompts

After using DALL-E and Midjourney for quite a while now, here's what I learned: these AI image generators are stubborn as hell. I can't tell you how many times I started a high-context prompt from scratch after they kept generating (admittedly, beautiful) images that had an element I didn't like.

Fortunately, this isn't the case with Stable Diffusion. Thanks to its negative prompt feature, you can exclude specific elements or avoid certain characteristics.

To show you an example, I asked SDXL to paint a “portrait of a lady with long hair standing in the rain with umbrella.”

It's a good output, but what if I want the lady to have less curly hair? Let’s try the same prompt again but let me add a negative prompt this time: “curly hair.”

Similar outputs, but notice how the lady now has straight hair? That's the power of negative prompts.

Let me give you a different example. To start, I'm going to use the prompt “portrait of an old man without long beard.”

See how the old man in the image has a beard, even after I specified that he shouldn't have one? This is an issue I keep encountering with AI image generators. There are times that they ignore instructions and skip to the other parts of the prompt.

With SDXL, I found it easy to fix this mistake. All it took was dividing the input into two: a positive (portrait of an old man) and a negative prompt (beard). See for yourself:

How Does Stable Diffusion Work?

Stable Diffusion Online

Since Stable Diffusion is open-source, you can actually use it using websites such as Clipdrop, HuggingFace, Playground AI, Stable Diffusion Online, or (my personal favorite!) Dream Studio. You can also Dream Studio for free for 25 images or so. It's also significantly more flexible with its parameters which allows you to customize the output better.

Below are the step-by-step procedures on how to use Stable Diffusion on Dream Studio.

1.0 Head to https://dreamstudio.ai/generate. Accept all pop-ups.

Stable Diffusion (Dream Studio) Step 1

2.0 Create a new account then sign-in. You should see the control buttons on the left side and a sample prompt. By default, the aspect ratio is set at 1:1 and the image count is 4. Feel free to tweak this as much as you need.

Stable Diffusion (Dream Studio) Step 2

3.0 From the dropdown menu, you can choose from 16 different styles, including Anime, Photographic, Comic Book, and Digital Art.

Stable Diffusion (Dream Studio) Step 3

4.0 After placing your prompt, hit the Dream button. After a few seconds, your images will appear.

Stable Diffusion (Dream Studio) Step 4

It's important to note that this is just one of the many ways you can use Stable Diffusion. This one's my preference because Dream Studio streamlines the prompt process which, if you've been using other tools such as Midjourney long enough, could be tiresome.

Stable Diffusion Prompts & Examples

New to test–to-image generators? You've come to the right place. Don't waste your credits by testing (and failing, but that's okay!) prompts on paid Stable Diffusion services. Instead, start by giving some of our examples a try:

Architecture

Prompt: Luxury apartment interior with minimalist furniture and lush houseplants and abstract murals | modern architecture, realistic, detailed, decorated with trinkets by makoto shinkai, ilya kuvshinov, lois van baarle, rossdraws and frank lloyd wright, 8k

Abstract

Prompt: Abstract rectangles colorful with ragged edges perspective oil paint drips infinite inception diffused splatter twisted lines

Portrait

Prompt: Ghostly young woman in victorian house, black and white, volumetric lighting, ultradetailed face, realistic portrait, gorgeous face, long braided hair, realistic photograph, by caravaggio, award winning photo, very sharp, ultradetailed, beautiful light, beautiful eyes

Cinematic

Prompt: Mad max: fury road. chasing cars with fire and explosions everywhere. a man and his yellow suzuki esteem in postapocalyptic desert, heavy weapons. directed by frank miller.

If you want to see more Stable Diffusion prompts, you can visit PromptHero.

Stable Diffusion Pricing

The pricing for Stable Diffusion varies, depending on how much and where you will use it. For its more popular platforms, this is how much SDXL costs:

Stable Diffusion Pricing (Dream Studio)

Dream Studio offers a free trial with 25 credits. If you need more, you can purchase them for $10 per 1,000 credits.

Now, here's where it gets tricky. Dream Studio says that 1,000 credits can generate 500 SDXL images but that's actually just for artwork generated using their default settings. In truth, depending on your step count settings and the image resolution, it could be way less than that. A single prompt can range between 0.2 and 28.2 credits.

Stable Diffusion Pricing (Clipdrop)

Unlike in Dream Studio, using Stable Diffusion via Clipdrop has no pay-per-image option. Free users are allowed to create 400 images a day but with watermarks and low quality. There’s only one price option - $9 a month for 1,500 images a day and without a watermark. Paid users can also enjoy HD images and use other features. For more information, see the Stable Diffusion Clipdrop pricing table below.

Free

  Pro

  Price per month

  $0

$9

Stable Diffusion XL

  Watermark, 400/day

No watermark, 1,500/day

  Uncrop

Watermark, 100/day

No watermark, 1,000/day

  Background Removal

1024x1024 max

HD (1000/day)

  Cleanup

1024x1024 max

HD (5000/day)

  Relight

1024x1024 max

HD

Image Upscaler x2

2048x2048 max

Image Upscaler up to x16 (1000/day)

  Web Editor

10 free clips

Unlimited

Text Remover

1024x1024 max

Unlimited

Clipdrop iOS

limited + watermark

Unlimited

Clipdrop Android

limited + watermark

Unlimited

Reimagine XL

High Resolution (2000/day)

Replace Background

High Resolution (500/day)

Stable Diffusion Pricing (API)

Basic

  Standard

  Standard

  Price per month

  $0

$9

$149

Image Generation

  999

9999

Unlimited

  API

No Controlnet APIs

Access All APIs

Unlimited - All AI APIs

  Shared GPU

  Req/s Limit

5

10

15

Stable Diffusion Pros and Cons

PROS

CONS

  • Can be used for free

  • Available offline

  • Open source

  • Easy-to-use interface

  • Highly flexible

  • Fast image generation

  • Your images are private

  • Requires powerful hardware

  • Steep learning curve for beginners

  • May require extensive experimentation

  • May not always produce desired images

  • Paid API Access

Stable Diffusion Alternatives

Midjourney

Midjourney is a prominent text-to-image generator and is an extremely good alternative to Stable Diffusion. Founded in July 2022, this powerful AI image generator is popular with enthusiasts, prompt engineers, and content creators of all shapes and sizes. It also shares some features with Stable Diffusion like negative prompts and image variations.

For more details, check out our comprehensive Midjourney review.

Midjourney Pros: Why It's Better Than Stable Diffusion

  • Affordable
  • Beginner-friendly
  • Readily accessible
  • Quality and creativity control
  • Area specific variations and zoom-in/zoom-out
  • Active community and consistent developer support
  • Comprehensive documentation, including how to use some handy parameters

Midjourney Cons: Why It's Worse Than Stable Diffusion

  • No official API
  • Only available via Discord
  • More expensive than Stable Diffusion
  • Has fewer settings than Stable Diffusion

DALL-E 2

DALL-E 2 was officially made available for public use in September 2022. Developed by OpenAI, this second iteration of DALL-E works on 3.5 billion parameters and can generate photorealistic images.

You can also expect DALL-E 3 to be publicly available in the next few months.

Pros

  • Highly scalable
  • ChatGPT Integration
  • Creates images quickly
  • User-friendly with a straightforward interface

Cons

  • No free trial/version
  • More prone to uncanny valley and hallucinations
  • Image quality is heavily dependent on the prompt

Final Thoughts

There's no other way to put it: Stable Diffusion exceeded my expectations. As one of the most parameter-rich language models today, SDXL 1.0 is a significant step-up from its predecessor.

However, with the release of DALL-E 3 and the dominance of Midjourney, can Stable Diffusion maintain its excellence and carve out a space for itself in the AI image generator market?

To be honest, it's hard to tell. However, the one thing that Stable Diffusion has over its competitors is something that many people have been requesting from its competitors: the ability to generate images for free. Features aside, this alone makes SDXL worth trying out.

Overall, I enjoyed using this AI model.

I'll definitely be back for more!

Want to Learn Even More?

If you enjoyed this article, subscribe to our free newsletter where we share tips & tricks on how to use tech & AI to grow and optimize your business, career, and life.


Written by John Angelo Yap

Hi, I'm Angelo. I'm currently an undergraduate student studying Software Engineering. Now, you might be wondering, what is a computer science student doing writing for Gold Penguin? I took up studying computer science because it was practical and because I was good at it. But, if I had the chance, I'd be writing for a career. Building worlds and adjectivizing nouns for no other reason other than they sound good. And that's why I'm here.

Subscribe
Notify of
guest

0 Comments
Most Voted
Newest Oldest
Inline Feedbacks
View all comments