Welcome to this week’s edition of Architecture Insights.

In this week’s 5 minute read:

News & Updates

Adobe Firefly has improved its AI-powered image generation with Firefly Image 2. The latest version is currently only accessible on the web, but will soon be available on Photoshop and other Creative Cloud applications. Check it out here.

Europe has enforced strict rules on AI development while Asia adopts a more business-friendly approach. Both announcements are significant in terms of their impact on the development of AI safety protocols and guidelines. Read more about it here.

DALL-E 3

Overview

As of October 2023, anyone with a Microsoft account can access the latest version of DALL-E 3.

This opens up new possibilities for creating images specifically for architectural rendering purposes.

If you have any experience with the older versions of DALL-E you might have found it underwhelming when it came to creating high-level renderings and perspectives that could be used for presentation. DALL-E 3 brings us one step closer.

While we would like to acknowledge that in overall capabilities Midjourney is still the leader in generating high-quality images. This post focuses on DALL-E and how you can start using it today to generate your ideas to images.

How to access DALL-E 3

To access DALL-E 3, visit bing.com/create and log in with your Microsoft account. A free version is available as of October 2023.

Check out one of four images DALL-E 3 created for us using the following prompt:

“Create a victorian style castle in the forests of the Amazon with an overcast setting.”

This example was just to show you how the quality has improved and how creative the AI can get when you give it a specific prompt.

If DALL-E 3 can achieve similar results for more site-specific requests, could it be the end of manual image rendering tools like photoshop?

Let’s try another prompt:

“A bungalow by the beach with a wooden deck leading out to the water.”

From one perspective, this prompt is still a very basic and high-level request but if we are looking at things like quality, detail, and realism, then DALL-E checks off a few of those boxes.

Understanding that the more detail you give your prompt the more likely you can predict the outcome.

Let’s try the same prompt and be a little more specific:

“A bungalow by the beach with a glass roof that features a hammock and chairs. A stone pathway leads to an adjacent beach and gets lit up with bollards in the sand.”

These were the 4 images generated.

Once again they are all high quality, fairly accurate (one of them doesn’t feature the glass roof), and offer unique variations to choose from.

We can see the concept design phase as a great place to start experimenting with DALL-E for idea development and/or inspiration.

One limitation is that AI software still falls behind when it comes to site-specific designs. They cannot yet use exact locations via Google Earth or alternative map software as a base point for images. While there are diffusion models out there with these capabilities, for a tool like DALL-E, while OpenAI’s main focus continues to be ChatGPT it will likely never reach the full potential of other tools out there like Midjourney and Stable Diffusion.

Resources

Read how OpenAI improved DALL-E 3 from a computing and software perspective, Link.

Soon, video creation AI will reach the capabilities that image-generating AI has. Check out Moonvalley.ai for software that has started this development.

Adobe’s prompt guide helps with writing better requests and delivering better results when using AI image generators.

AI Image of the week

In case there weren’t enough AI generated images for you in this email already, here is one more.

Thank you for reading this week’s issue, check past issues here. Sharing this newsletter with colleagues, friends, or anyone interested in the combined world of architecture and artificial intelligence is greatly appreciated.

Until next Friday,

AI