AI images got some serious upgrades

2 major hurdles down, what's left? and how AI image models can be useful in work

Welcome to another edition of what we’re determined to make the best damn newsletter in AI. Here we’ll break down AI topics that matter, open your mind to use cases, and keep you ahead of the curve.

Our #1 goal is to be useful. So please shoot us an email 📩 if you have questions or feedback, and especially if you implement something we share!

Here's what we're covering today:

  • AI-generated images just got a huge upgrade, but they're still far from perfect

  • Use AI to personalize images specifically for your brand

  • Is AI a new form of computers? Some investors think so.

... and if someone forwarded this email to you, thank them 😉, and subscribe here!

Let’s get to it! 👇


Stability AI introduced Deep Floyd, an AI image model that can generate text inside its images

Stability AI is the company behind Stable Diffusion, the main AI model that powers most of the image and art tools we use today. They had been teasing an AI image model that can “do text” and on Friday, it arrived!

While still currently in research preview, these roll outs tend to happen fast so we’d expect to see this in AI image tools we are using fairly soon.

We made a little comparison for you with the prompt "Deep Floyd Street" in Midjourney, vs the demo photos out of Deep Floyd... THE TEXT!

Let’s check in with AI image models for a second, shall we?

The way we see it, there are 5 big limitations AI image models need to overcome to reach mainstream usefulness.

AND researchers are knocking them down fast!

  1. The hands problem - initially, image models struggled to know that people only have 5 fingers, or dogs only have 4 legs. But the latest Midjourney image models solved that ✅

  2. The text problem - image models understand images and shapes, but have struggled with text in those images (think product names, store signs, etc). With Deep Floyd, now that’s solved ✅

  3. The resolution problem - these models still can only generate images up to about the recommended resolution for Instagram photos. Still in the works.

  4. The copyright problem - we’ve talked about this a lot, but there needs to be transparency on training data and the rights associated. Adobe Firefly is a very good start. But we want to see more.

  5. The control problem - this is more of a nice to have, but for more professional and creative use cases for AI image models, we definitely want to have more control on specific pieces of images or details. Some promising research coming here 👀

How can AI image models be useful in work applications? We’re glad you asked 👇


Personalize images for your brand with AI

Having custom images that fit your brand's look and feel used to be unrealistic for most businesses. But now it’s easier and much more affordable.

We personally use Midjourney to generate all of our blog/newsletter images. We created a template that we’ve iterated on over time to continue getting a look and feel that we like.

Now it’s extremely easy for us to generate new brand assets, and anyone can do it.

Here’s how: create templates for your visual brand.

You can do this in Midjourney. First, spend time with their documentation and community feed to get prompt inspiration. Then build a template that can be used to create consistent imagery in a set style.

Here's roughly how we do ours:

✨ Premium Subscribers: Check out The AI Exchange playbook for getting started guides on DALL-E, Midjourney, and Stable Diffusion (or ask AIxBot in Slack!)


For your reading list 📚

We're continuing to see AI alter more spaces...

And if you're really nerdy...

We've had a lot of fun with our AIxBot beta release last week! ✨🤖

Subscribers are getting advice on how to bring AI in to their business, tool recommendations for their use cases, and even getting to ask AIxBot how it's built 😉

We've got more coming soon - stay tuned!

Premium subscribers → Log in to Slack and check out AIxBot, your friendly chatbot to help you adopt AI 🙌 powered by The AI Exchange knowledge base

That's all!

We'll see you again on Thursday. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected].

... and if someone forwarded this email to you, thank them 😉, and subscribe here!


🪄 The AI Exchange Team