Hello, my curious fellows! Want to share our experience in using CHAT GPT/DALL E3 AI for making complex action images involving animals for our newly published book "Animal Fairy Tales for All Ages". The first part is writing a detailed prompt that the AI artist (DALL E3) can understand. This is done by Chat GPT based on your instructions. However, our experience is that DALL E3 does not always follow the prompt faithfully. You and Chat GPT keep on tweaking the prompt till you get the desired image. We also noticed that fewer elements in the prompt, fewer errors DALLE3 makes. As an example, the Front cover (complex image with many animals) of the book took a lot of hit and trial whereas the back cover (simpler images) was easier to generate quickly.
Animation : AI images in Animal fairy Tales for All Ages by Khawar Sohail Siddiqui
Other topics in Animation:
Check out Trending Discussions
Register / Log In
AI-generated Front and Back covers
1 person likes this
This really shows that, although AI can create individual pieces of interesting looking art, for a project like this, you still need a designer. There’s no cohesive style between these pieces. The cover as a whole doesn’t resemble a professional book cover, needs a layout artist.
The original back cover was symmetric, but the images were moved at the last time to make way for the Bar code image. You are right that a professional designer would certainly add a spark so that the images are even more eye-catching. I add here an AI-generated image using chat GPT 4 from inside the book (story 2).
AI drives me nuts. It's really hard to get good interesting shots and compositions. Its always that same boring old eye level boring poses. The way it has so much trouble with fingers cracks me up and when I was playing with it, the amount of times it turned PoC into white people was frustrating. I was trying to get a shot of a village with native americans and it ket sneaking in this one white guy that looked like Joe Biden wearing native american traditional clothing....lol It can be very useful still even if it's struggling.
1 person likes this
You are right. All AI image creators are in infancy and even the best combo Chat GPT/DALL E3 is not perfect and make lots of mistakes. It put things in the image that I never asked and leave out things I asked it to add. The example is one scene for my story where two pre-historic hominids meet a proboscis monkey and a cockatoo is watching. Like your experience it will always put one very modern white American in the scene (see the image). I had to repeat it many many times till I got it right by trial and error. I learned that if it is making same errors again, switch to a New chat. With time it will get better. I asked Chat GPT to make a bullet point prompt which the Image creator can follow more faithfully. In the end I got many nice images for my book.
In another scene where only animals should have been shown voting, it created one image where animals are dressed up as modern humans dressed in 3-piece suits. Funny! but not what I wanted to fit my storyline. In the end after many tries, I got the right image.
In this image, I wanted Justice jaguar to be giving a verdict about a dispute between animals. In the background, the image creator showed Jesus on the cross, which it added by its own choice. Again, after numerous trials it got it right. You have to tell it what elements you do not want in the image. Hope you enjoyed these misfit images.
Khawar Sohail Siddiqui I have to admit it’s super useful for pitching ideas! Especially live action. Your images look really good though!
1 person likes this
It is these animal action images where the chances of errors increase. More the elements, higher is the probability of making an error. Here is a good image accordioning to the storyline.