In today's newsletter, I am going to talk about DALL-E 2, the new artificial intelligence image generator created by OpenAI, one of the biggest companies in the AI space right now.
If you are not familiar with it, DALL-E 2 let's you add any text prompt (i.e. a photo of a delicious pepperoni pizza slice in a white plate), and it will generate a photorealistic image of it in around ~10 seconds.
As crazy as it sounds, for many requests it works terrifyingly good.
So after ~60 hours of research about DALL-E 2, and with all my background and experience in the creative industry (I own a web design agency since 8 years ago), I will approach this newsletter with 3 questions:
1. What kind of designs can DALL-E create: Along with detailed examples of positive/successful design results.
2. What does this mean for current designers: I am going to be replaced by a robot anytime soon?
3. How to prepare for the future of AI-design: How to take advantage of DALL-E 2, along with some valuable resources to learn more about it.
If you work in the creative industry (design, development, photography, marketing, etc), I highly recommend you to read this in full, I promise it will be worth every second of your time.
After playing around with DALL-E 2 for hours, and using over 1,000 credits (one credit is equal to one request that generates 4 images), I figured out the design applications where it works best.Let's go over each of them:
After trying out many different prompts, I was very surprised that by submitting the right request, DALL-E can generate pretty nice 3D icons.
This can be a game-changer for web designers or graphic designers, among many others, because having the opportunity to use 3D graphics as assets for any website or graphic design work, can easily elevate the final design quality without adding much extra work.
I also got great results with simple 3D characters, such as the ones below.
It can produce more complex 3D renders or characters too, however, the more details and polygons added, the less 'polished' the final result is. Here are some examples:
That being said, even if the results for high complexity 3D renders are not great, the resulting images can still be a great starting point for 3D character design.
Another great strength of DALL-E is generating realistic photographs, and this is nothing short of amazing.
If you are searching for a very particular image and you are not able to find it on any of the popular stock sites (i.e. Adobe Stock, Shutterstock, etc), DALL-E 2 can be of great help, as it can generate very impressive images, as the 8 I generated below:
This is by far what impresses me more about DALL-E, because, without looking very specifically into the details, it's close to impossible to guess these are AI-generated images.
While I don't think stock photography will be replaced by AI-generated photos as of today, I believe that in the coming years (3~5 maybe), as the models get better, it is definitely possible that stock photographers won't be spending as much time with cameras, but instead submitting prompts of the photo they are looking for to the computer.
As of today, you can generate pretty specific photos using DALL-E as needed, and use them on any design application or marketing material (we had already done so in our agency, and no one has guessed yet that those are not real pictures).
Icons seem to be easy work for DALL-E. Similar to the 3D graphics mentioned on point #1, by using the right text prompt, the resulting 3D icons generated by DALL-E are incredible, and here are some examples:
Amazing, right? Typically designing 3D icons like this can take tens of hours (we have done it many times for high-end websites), however, in a matter of minutes you can generate a complete set using DALL-E.
It's the turn of logos, one of the most thoughtful and detailed design pieces in the creative industry...
The DALL-E results for logos are pretty variate. While for heavy-graphic logos the results can be pretty promising, and serve as a great starting point, the results for simple logos are not as good.
Here are some examples:
For logos, in particular, I believe if you are designing high-detail logos (i.e. restaurants, sports teams, local businesses, etc), DALL-E can help you save 50-70% of idea concepting time simply by generating many different logo options, and then it's just a matter of choosing the right concepts and direction, and design something completely polished based on it.
Oh, another big category! Faces.
DALL-E can generate photorealistic faces based on any text prompt (only limited to not generating famous or known faces, to avoid fake news and similar problems), and it works terrifically well.
Here are some examples:
However, the powerful thing about DALL-E is that you can really request anything with as much detail as you need, resulting in pretty specific face photos:
The applications for this are... endless?
From interesting applications such as giving a 'real' face to an AI assistant or generating more realistic avatars on videogames, to shady things like creating fake photos for fake testimonials.
As uncle Ben said: With great power comes great responsibility.
Another interesting DALL-E application is product idea concepting. Just like Paul Yacoubian shared on Twitter, he created a concept of an Avocado Backpack, and then he just send that as an example for the manufacturer to produce it. Simply incredible.
From my side, I tried this out with sneakers too, and it didn't disappoint.
The truth is that my fashion sense is not that great 😅, so there are some more cool product concepts created by DALL-E on the web.
But as you had seen, product idea concepting can be incredibly easy by using DALL-E, and I believe that will be a game changer for many people in the e-commerce industry.
If you ever found yourself searching for textures, you may know how specific they can get. The good thing is DALL-E knows how to generate them easily.
This has many obvious applications, from creating textures for traditional graphic design to creating textures for architecture 3D design, 3D rendering software, or even texture-fixing for 3D models as shared by Shahriar Shahrabi on Twitter.
Last but not least is one of the most fun (but unfortunately not so business-oriented) DALL-E applications: Generation of a variate set of illustrations or art-related images.
The DALL-2 Subreddit is full of incredible examples that will blow your mind as you scroll, however, in most of these, it is hard to find the pragmatical application (it's very fun tho).
Here are some of mine. At this point, you may already guess that I am a big fan of dragons 😂.
After going over all of these DALL-E use cases, you may be wondering the next big question:
For many designers, it can be scary to see an artificial intelligence create in a matter of seconds something that may take hours for a 'human' designer.
So it raises the natural question: I am going to be replaced by a robot anytime soon? And I believe the short answer is no.
While something like DALL-E can create pretty specific design results (i.e textures, 3D icons, etc), it cannot produce the complete final work (i.e. a website, a brochure, a polished logo, etc), so designers are still needed and will be needed for a long time.
That being said, as of today the cost of using DALL-E is $15 USD for 115 credits (with each credit being equal to one request that results in 4 images), meaning the cost per credit is $0.13, or $0.032 per image.
As you may have guessed, it is not expensive at all, and I believe it is going to get much cheaper as time passes by and more competitors join the AI-image generation space.
So I believe that now more than ever, the gap between 'beginner' and 'expert' designers is expanding more and more.
Just like Canva 'commoditized' business cards and slide presentations by offering hundreds of high-quality templates for free, I believe tools like DALL-E will also commoditize stuff like simple 3D illustrations, certain types of graphics, illustrations, and some other things.
But this isn't necessarily bad, as this will incentivize the market to elevate the quality of work, and while this comes with downsides too, I believe there is still a bright future for those who embrace change, new tools, and ongoing learning.
Also, research from other DALL-E competitors (like Google Parti: Pathways Autoregressive Text-to-Image Model) has proven that the more data input the model has, the more the results improve. This means that as DALL-E and other AI-image generation tools evolve, it is only going to get better.
Just like my favorite villain of Marvel, one time said:
Dread it. Run from it. Destiny still arrives.
Even though some may not like it, it is true that AI-design tools have arrived, and it's better to be prepared.
So that raises our 3rd and last question of today's topic:
The nice thing about all of this is that if you are reading this, it means you are already interested in learning about AI design (so kudos to you 🙌🏻).
The next step is to keep learning so you master the tool, so you take advantage of it and not the other way around. So here I share with you a few resources that can be relevant and useful:
DALL-E is getting so popular, that there is even a new marketplace called 🧠 PromptBase, where you can purchase 'prompts' (strings of texts) that work well to produce specific results from DALL-E. Really impressive for something like this to appear so fast.
As time passes, it is very likely that many more software solutions and platforms will integrate with the future DALL-E API to do all sorts of incredible things. So stay tuned about it, as this is just the beginning.
Along with the above, I highly recommend you to stay tuned about some of the other relevant DALL-2 competitors (Not all are available yet, but it's likely they will be available in the future).
Finally, a short conclusion about all of this:
Take a look, read, experiment, and learn. For me, trying out and learning the basics of DALL-E was very fun, intellectually inspiring, and very interesting.
I honestly believe that DALL-E 2 is the most impressive piece of software that I have tried in my lifetime, and every time I show it to my friends and family, they are just amazed by how it instantly generates images based on a simple line of text.
I believe if you genuinely like design, it won't feel like 'work' to you, but more like learning what seems to be the future of our industry, so it shouldn't be too hard.
As history has told us, typically the ones that adopt new technology sooner, are the ones that benefit most from it, so I'm really confident it will be the same this time.
And that's all for today.
P.S. The beautiful tiny camera 3D icon used in this newsletter cover was created using DALL-E 😉.
Follow me on Twitter, and let's chat.
As a tradition of this newsletter, I always ask a question at the end.
If you liked it, I invite you to reply and answer the question so it is not only a one-way channel of communication (I promise I will reply back).
How do you think AI will shape the world in the next 10 years? I personally have strong beliefs on how AI will shape the way we live, so it would be interesting to hear your thoughts too.
And that's all for today.
P.S. Liked this newsletter? Scroll a little, and subscribe to receive the next one in your inbox.
See you next week!
We will be talking about website calculators, and I don't mean literally a calculator just like the one you have on your iPhone, but instead a pricing or quote website calculator...
Until recently, building voice assistants or apps (aka skills on Alexa, or actions on Google Assistant) was very hard and required a lot of technical knowledge, however, now it's much easier thanks...