The Uefa EURO 2024 is over and what a tournament it was. With Spain as the well-deserved winner, if you ask me, it is now time to reflect back. As mentioned before I would like to share my lessons learned when it comes to:
- AI Prompt scripting
- AI Builder Prediction model
For this blog, I would like to focus on AI prompt scripting. This will become a very important skill. Will AI prompting a new class in the future? I think it will, since this will be incorporated in lots of people their daily activities.
All my AI-related posts, such as the ones during the Uefa EURO 2024, I created with Microsoft Designer. This is using DALL-E artificial intelligence that allows you to create unique images using AI Prompts. The beauty of this if you ask me that is that the possibilities are endless. I sometimes compare it with if you ask a little kid to draw a painting about a zoo on the moon with dancing elephants and monkeys on it you will get the most amazing results, no limitations. Below are the lessons that I learned, but don’t forget, to get better at this you need to practice this every day to improve your “AI prompting skills”.
Concept of labels
During the tournament, I wanted to provide updates on how well the prediction model was doing. I wanted to use a scoreboard, how ironic, but I had quite a few issues with coming up with the right prompt. What I found out is that if you want to have a single label on a single object in your image it would not take many attempts to generate the right image. With more objects and more information in a label, like the example below, you never know how it will be generated and how it is spelled. So if you want to use the concept of labels or text, please choose them wisely.
AI Prompt: “A realistic photograph of a robot in 2080 sitting on the bench in a soccer stadium and looking at the scoreboard that says “Team AI – Team Opponent 2-2“
Concept of sports rules or actions
If you were about to ask someone, can you draw someone taking a free kick and have the opponents standing in the wall, you probably would not end up with something like below. My experience is that concepts within a sport, in my example soccer, it should be able to generate better images if ask me. In addition to this, it perhaps does not help that English is not my primary language, I get that 🤣
- bicycle kicks, I had plenty of examples where robots were flying around
- Free kicks, Even though I wish that the Netherlands would have walls like these in their goal, it is not what you would expect when you try to generate it
- Penalty kicks, I had examples where goalkeepers were standing in the net of the goal or on the wrong side of the post
AI Prompt: “A photorealistic image of a robot wearing a Dutch orange soccer jersey in the year 2080 playing in the Olympia Stadium in Berlin. The Opponent robots are wearing a Turkish jersey standing in a wall and the robot taking the free kick is wearing an orange jersey“.
AI Prompt: “A photorealistic image of a robot in a Dutch soccer jersey, scoring an amazing bicycle kick against a robot standing in the goal and wearing an Austria Jersey in a sold-out Olympia Stadium in Berlin“.
Too many objects
When your AI Prompt does contain too many and detailed requests, like in the example below, it gets confused and objects will either not be generated or it is mixing up instructions. The difficulty on top of that is that in the examples I wanted to generate you had in a lot of cases a robot wearing a Dutch jersey and one for the opponent. Having multiple people / robot objects can be challenging if you want too much details in it as well.
AI Prompt: “a photo realistic image of a Robot wearing a Dutch soccer jersey in the year 2080. The Robot is ready for taking a penalty and the goalkeeper wearing a England jersey is holding in his left hand a bidon“
Scenes
Within my prompts, I tried to mention the stadium names so for example “Dortmund Signal Iduna Park”. Being a big soccer fan myself, I did recognize most of the stadiums that it generated quite well. This is absolutely amazing! What it also did a few times, is that because you used Dortmund Signal Iduna Park as the stadium, that a lot of details around it were all of a sudden getting the colors or logos from Borussia Dortmund who would normally play in that stadium in yellow. This can save time if you want to prompt about Borussia Dortmund, but in my case, it was about the Dutch National Team playing there and that was making it a bit challenging.
AI Prompt: ” A photo realistic image in the year 2080 of the Dortmund Signal Iduna soccer stadium with a robot with an orange jersey in the goal, building a wall with bidons“
Microsoft Designer
If you would like to play yourself using AI prompts to generate images, you can do this yourself using Microsoft Designer. It is absolutely amazing to see what it can do and it will only get better of course. Using your imagination to create your own fantastic images, greeting cards, icons, wallpapers, backgrounds or many other things is absolutely awesome. It does require AI prompting, but Microsoft has provided templates and examples to help you get started. I can only encourage you to try this yourself, since I think this will become more and more important in our daily lives. I expect our children will grow up with this and will be using it as their second language and probably get classes at primary school to become experts in this.
Microsoft Designer: How to use AI image prompts to generate art using DALL‑E | Learn at Microsoft Create
