20/04/2023

5 observations on how AI can help us create images for scenario stories

Images are powerful and speak to us in completely different ways compared to text. So far it has been a difficult, time-consuming task to create them for complex, multidimensional scenario stories, especially since there is often no existing material, images and sometimes even words for potential futures or future elements. Hence, we had to be creative and imagine and create new drawings or collages ourselves. The use of AI generated images is opening up new opportunities to boost our creativity and ease the way of image production. We are only at the beginning of this journey to discovering its potential. Here are 5 observations from our experiments with using AI to generate images for scenario-stories:

1) Human imagination is still needed

The result of copying a scenario story into Chat GTP, telling it to create a prompt for an AI Image Generator and using this to create an image was disappointing and not reflecting the scenario sufficiently. But the more precise we imagined an image or elements of the image and described it to the AI Image Generator, the more targeted and useful were the outputs. To not be stuck on certain ideas how an image could look like, Chat GTP can certainly help to create a basis for the description/prompt, give ideas for variations or what else could be added.

2) It is still a development process, not one-click

There are so many possibilities of how an image for a scenario could look like, trial and error is an effective approach to finding out what content and style reflects the messages best. Therefore, it might not be sufficient to write only one prompt for the AI Image Generator and to generate only one image. It is an approximation process, and especially in the first attempts we still have to learn which inputs for the AI Image Generator are helpful or needed e.g. style of the image, size, colours and atmosphere. In this process it can help to use existing images and ask the AI to describe it for you in order to learn how a prompt could look like.

3) Dealing with limitations

The AI Image Generator has limitations and cannot display all that we imagine. In our case, the AI was struggling to create an image with several layers, e.g. “a frog in the front, a person behind it going into a building, while the sun is shining and you can see an airplane in the sky”. The image result was focusing only on one or two elements and not able to cover all of them. It was also not able to create visualisations of all elements we wanted to display, e.g. flying pianos.

4) It helps to use other tools for post-processing

Additional tools can be useful to customise the generated image further according to the needs. The following were especially helpful for us

vectorizer.ai to create a vector graphic from your AI-generated material, to be able to change details in Illustrator
the (AI-) neural filters from Photoshop e.g. to enlarge the images or transfer the style of a reference image
to use Photoshop or Indesign to continue working manually on the image or create collages of several AI generated images