Stability AI’s text-to-image models arrive in the AWS ecosystem

This is a toolbox story. AWS Bedrock is an AI toolbox, and it’s getting loaded up with a few new power tools from Stability AI. Let’s talk about the toolbox first, and then we’ll look at the new power tools developers can reach for when building applications.

Also: 6 digital twin building blocks businesses need – and how AI fits in

TL;DR: Developers can now call on text-to-image features from within their Bedrock-powered apps. That means they can easily incorporate AI-generated images within their own code.

AWS (Amazon Web Services) is Amazon’s powerful cloud service. The company offers a wide range of services and capabilities within the giant AWS umbrella. The key thing to know is that all of the services are on-demand and scalable. That means you don’t have to build a data center and forklift in a pile of server racks before you can start deploying applications.

Bedrock for AI choice

In the context of our story, if you want to incorporate powerful AI capabilities in your code, you don’t have to develop those capabilities from scratch, train the AI large language models (LLMs), or even figure out what server configurations you’ll need. All you need to do is enter your credit card digits, read some documentation, and start writing code.

I built my first Internet company back before there were cloud services. Trust me when I say that something like AWS is a vast and amazing game changer compared to building out server infrastructure on your own, especially for founders working on a startup’s budget.

AWS Bedrock is Amazon’s Large Language Model on Demand (LLMoD) offering. These LLMs are called foundation models. AWS Bedrock offers foundation models from AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, and Amazon.

Also: 6 digital twin building blocks businesses need – and how AI fits in

This is powerful for developers because they don’t have to implement those models. They just have to learn the protocols for talking to them and then use them, paying as they go.

But there’s another huge win with a service like this: you can try different models and see what works best for your application. If you had to build out your own infrastructure and then set up and configure each model, you’d probably just pick one and live with it because the work to switch is just too time-consuming and stressful.

But with Bedrock, you just switch a few parameters, and you’re off to the races and testing different foundation models. It’s easy and fast and gives you a way to compare and contrast AI solutions in action, rather than just guessing from what’s on a spec list.

Stability AI for image generation choice

This announcement is about Stability AI adding three new power tools to the toolbox that is AWS Bedrock. Each of these models takes a text prompt and produces images, but they differ in terms of overall capabilities.

Stable Image Ultra Excels in creating very high-quality, photorealistic output. It’s used in projects that require high resolution and fine details, like high-end marketing campaigns.
Stable Diffusion 3 Large (SD3 Large): This model excels at cranking out a ton of images quickly. The images are generally high quality, but they’re not going to be the main image in a fancy magazine spread. Instead, they’re more likely to put out a bunch of game assets or catalog entries or anything that requires a lot of images, done fast and with good quality.
Stable Image Core: This is the budget option. It creates nice images but without ultra-high detail. It’s also fast.

You can see how the tool analogy fits here. When you open your toolbox, you’re able to choose which power tool fits your project. Sometimes, you want a hammer drill; other times, you want a power screwdriver. Likewise, sometimes you want a graphics tool that generates an insane level of detail. Other times, you just want fairly cheap and fast.

Also: 6 digital twin building blocks businesses need – and how AI fits in

By adding all three models to Bedrock, Amazon and Stability AI are giving developers those choices at any time and anywhere they need them.

How could these be used?

Let’s use one example. My wife runs an e-commerce hobby-related company. Every month, she posts a theme on social media that inspires her followers to create a project. Back before good text-to-image generative AI, I created an image for her based on some brand assets using Photoshop. It took me a few hours, and the images were never all that great.

Also: How I used ChatGPT and AI art tools to launch my Etsy business fast

But then I got a Midjourney subscription. Now, each month, she gives me the theme, and I write a quick Midjourney prompt. Then, she chooses from four or more images for the one that best fits the theme. Instead of hours, it takes me minutes. And instead of looking like I pasted up clipart, each theme image is ideal in how it represents her business and theme.

She then takes those images, copies and pastes them, and posts them on social media and her site.

But what if we used Bedrock with Stable Diffusion? We could add a feature to her e-commerce dashboard for the theme of the month right from within the dashboard. She could just type in a prompt, get back a few samples, and click to have those images posted to her site. The process would all be integrated within her dashboard workflow.

Doing that integration wouldn’t require a ton of AI theory and practice. All it would require would be a series of API calls from her current dashboard to Bedrock and handling the image assets that came back from those calls. The AI task could be integrated right into the rest of her very vertical application, specifically tuned to her business.

Also: 6 digital twin building blocks businesses need – and how AI fits in

So, that’s the potential. Think about it for media and entertainment, where marketers could quickly produce marketing assets; retail, where managers could quickly produce photorealistic product visuals; advertising, where very high-end images with great detail need to be created fairly quickly and easily; and in-game design, where game designers could build a full library of assets using these tools.

Of course, some would argue this eliminates artists and takes jobs. Without a doubt, that will be the case in some applications. But the opposite is also true. My wife’s small business would never hire a dedicated artist. Instead, I put on my art director hat (one of the many roles I wore as a small company founder back in the day) and produced fairly mediocre images. The AI just simply upped our game and saved us time at the same time.

For many businesses, that’s a win-win.

You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, and on YouTube at YouTube.com/DavidGewirtzTV.