My initial attempts at using DALL·E 2

DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language that everyone talks about on social media. However, it is lesser known that it can also create images starting from existing images, so I decided to test this second functionality. Here you can see a quick snapshot of the results.

I uploaded an image and a “generation frame” of a certain size was added to it by the tool. The generation frame is where the AI applies its algorithms and completes the image based on your brief.

TENTATIVE #1

Original image: a drone picture I took of an island in Lombok, Indonesia

Generation frame: below the image

My request: “add a painted dragon seen from above”

Result: the tool suggested 5 images and I chose the one that got closer to my brief

Comment on the final result: I was not 100% satisfied with the result. However, surprised by how well the image got completed in the generation frame, I decided to take it a step further starting from the same original image.

TENTATIVE #2

This time I placed the generation frame above the island and I asked for something simpler (I already knew DALL·E 2 was good at it) but also more complicated.

Generation frame: above the image

My request: “complete the image of the island (the easy part) and recreate a realistic war scene with ships attacking the island with cannons seen from above (the difficult part)”

Result: the tool suggested again 5 images and I chose the first one

Comment on the final result: the 4 other images generated were far away from resembling a realistic war scene with cannons, so “the difficult part” of my request was not fulfilled. I can’t deny that the “easy part” of completing the island is actually impressive and I could not help but ask myself how many people other than me would have not been able to do something better in such a short time with other tools…

Conclusions: If this is just the first attempt, I am definitely excited by all the possibilities that will surface in the coming years and I will keep testing it together with other tools such as Midjourney and Stablediffusion.