This is how AI sees the 27 EU countries, according to DALL-E 3 and ChatGPT.

We tasked DALL-E 3 with creating a portrait for each EU nation, and the outcomes were wonderfully bizarre. Here's how AI envisions the 27 EU countries, as interpreted by DALL-E 3 and ChatGPT. Generative AI might seem like an old companion, but in truth, our journey with this novel technology began just about a year ago. We're intrigued by its current capabilities and potential future advancements.

This piece highlights our use of ChatGPT and DALL-E 3 to craft engaging, informative content about the EU's 27 countries. Our project served as a playground for delving into AI technology, refining ChatGPT's knack for crafting precise descriptions of states, and testing AI's ability to distill and articulate complex data in a clear, impactful manner.

Simultaneously, we harnessed DALL-E 3's prowess to depict each EU country's unique attributes. This venture offered insights into AI's interpretation and graphical representation of varied landscapes and cultural icons, based on its training data.

The prompt

To generate 27 unique images, we sought a single, reusable prompt. Following experimentation, we finalized this prompt, utilized for creating all state-specific images and facts.

The Challenge: To produce 27 distinct images, we sought a versatile prompt, applicable to each nation. After experimenting, we settled on a template prompt, modified for each country. The results varied, as expected in a project of this scale, and given the current state of generative AI. However, persistent efforts yielded intriguing outcomes.

Upon presenting each country's image, we'll conclude with our observations. Now, let's embark on a European journey powered by generative AI.

Austria

Belgium

Bulgaria

Croatia

Republic of Cyprus

Czech Republic

Denmark

Estonia

Finland

France

Germany

Greece

Hungary

Ireland

Italy

Latvia

Lithuania

Luxembourg

Malta

Netherlands

Poland

Portugal

Romania

Slovakia

Slovenia

Spain

Sweden

Project observations

Our project, aimed at visually representing each European Union country, extended over an unexpectedly lengthy period, spanning nearly two days. Generating images for each country varied significantly in time, ranging from 2 to 10 minutes. We encountered challenges with the AI, requiring multiple attempts, sometimes 5 to 10, to adhere to specific parameters. Our initial strategy involved modifying prompts to guide the AI, but we soon discovered that requesting ChatGPT to regenerate the images was more effective in achieving the desired outcomes.

We faced particular difficulties with the AI regarding the 16 x 9 aspect ratio. It tended to produce abstract elements within the main image and frequently overlaid maps, which was not our intention. Working with the AI felt akin to collaborating with a highly skilled yet somewhat obstinate intern.

Despite these challenges, the AI successfully captured the essence of most countries. However, its selection and placement of landmarks were somewhat arbitrary, offering an artistic rather than documentary portrayal of the 27 EU nations. This result aligned closely with our aim to present a picturesque representation of each country's unique landscapes and landmarks.

This endeavor served as a test to understand the dynamics of such a project. We often received additional landmarks in the images, which added unexpected elements to our work. A useful strategy we employed was to initiate a new session whenever ChatGPT began to deviate from our requirements. This approach generally allowed us to obtain two or three accurate images before the AI lost focus.

We noticed a pattern where the AI functioned more reliably at the start of a session, gradually becoming more erratic. Restarting the session and using our master prompt usually helped realign the AI's outputs.

It's important to note that we did not conduct thorough fact-checking regarding the specific details or geographical accuracy of each state. While our knowledge of European geography is robust, and no glaring errors were evident, our primary focus was on exploring the AI's capabilities rather than meticulously verifying each piece of information it provided.