DALL·E 2 is officially in beta!

Today I received an email from DALL·E that it is officially in beta!

Having had the privilege of an early access to the DALL·E platform (over the past 2 months), I have been meaning to write this article for a while now. DALL·E is simply exciting, intriguing and revolutionary - think about it, all these images are generated by a computer, they don't exist in the real world and are not based on a CAD / 3D design!!! Through my experiences, and having let my kid's, colleagues and clients imagination loose on the platform, I have developed a good sense of DALL·E's potential, limitations and copyright challenges in the future, which I wanted to share with you in this article.

Let's start with the potential - sorry to say to photographers and graphic designers out there, but DALL·E is officially threatening a big part of your lunch. The ability to put a requirement in a short sentence and have DALL·E produce an image in almost any artistic or photo realistic form, in under 30sec, is incredible. Any of the images above could have been used for cues in promotional campaigns, marketing material, presentations, etc. I often use stock photo platforms for some of my presentations, and having used DALL·E, I found this experience to be quicker and, in most cases, nailed the image I had in mind, compared to the alternative. The level of abstraction and merging of concepts is incredible - the images on the right half of the collage demonstrates this, where the interpretation of various styles and concepts are blended very well to produce a magnificent result. Even the angle and direction of the shot can be altered, which is truly mind blowing. It creates faces and people is a hyper realistic form and is good at limiting bias (their content policy prevents me form sharing examples of this). This is going to be a game changer in the stock photo and graphics design industry.

Moving to the limitations - it does not always hit the mark. You need to master key words and phrases to refine what you are looking for and there are some ideas that an AI can't draw learnings from in order to deliver the intended result. Examples of this is when I would ask DALL·E to create an oil painting of a scenic environment or people, it would do this well as it had many old oil paintings to learn from. However, when you ask it to paint the earth, a robot, etc. it tries its best to replicate the art style, but is far from the mark, producing an amateurish output. In addition, you can't control changes to specific aspects / elements of the produced imagery, as when you expand on the description, it can radically change the interpretation and move away from the original intent. They have two ways of providing variability and flexibility in this regard - each image can be expanded into similar looking alternatives or it allows you to delete the areas you don't like fill in the space with a new prompt. However, my experience with both these options is that it did not hit the mark and I would imagine that this would improve in the future releases.

The copyright challenges is the big question mark I have on this platform. I don't know the legalities behind how copyright laws works, but I would imagine that there is a limitation to creating and commercializing content that reflects existing brands, concepts, characters, people, etc. About a month ago DALL·E changed their policy to reject uploading of realistic faces or requesting for images of public figures / celebrities (which makes sense and I would have thought that they would have done this from the outset). However, I was surprised by the generation of the two images to the top and bottom right (basketball player dunking and robotic super heros) - in both these images there is a very high alikeness to the Air Jordon logo and Iron Man (first character) / Flash (second character) respectively. The key point I want to make is - given that DALL·E has learned from other photos, pictures, imagery, etc to produce what it does, it is technically drawing on IP owned by others to create content that it claims and commercialises on. Furthermore, if it used my face to learn how to draw faces, there is a chance that I may land up in an AI generated photo in the future without my consent. I'm curious to get a legal view on this from someone in my network.

This is an exciting solution and looking forward to seeing how this progresses.

#dalle2 #artificialintelligence #4thindustrialrevolution