Image Creation Challenge
The Image Creation Challenge tasks participants with using generative AI tools to create images related to their work or interests.
The purpose
Ensure participants are adept at utilising generative AI tools to create images.
Reflect on the quality of AI generated images, AI’s algorithmic bias, and the ontology of creativity.
Discuss and share strategies for image generation prompts, in particular iterative or chain-of-thoughts prompting to get better results.
What you need
Setting it up
Participants require access to generative AI tools with ability to generate images, such as Copilot, Stable Diffusion, Adobe Firefly, or Midjourney. (At the time of writing Gemini’s image generation functionality is not yet available in the UK.)
How long does it take?
For asynchronous online play, we suggest it will take individual participants 5-10 minutes. We suggest leaving the game open for a week to 10 days depending on the game host’s availability for engagement.
For synchronous in person play, we suggest 5-10 minutes of individual game play and discussion time.
How it works
Select a generative AI tool and direct it to create an image. E.g. “Create an image of a hamster in a space suit walking on Mars.” Refine your prompt as appropriate.
Save and share your image.
Suggested follow-up
Compose a short reflection on your experience with the image creation process, including any insights or challenges encountered.
Reflect on the nature of creativity. Is generative AI creative? Can outputs from generative AI truly demonstrate beauty, originality and creativity? Do you feel differently when the output is in a different format–text, audio, video?
Have generative AI generate an image with text. Can it spell correctly? Can you ask it to change only the word?
For asynchronous online, we suggest participants share their screenshots and reflections in an online forum. The game host should plan on reviewing submissions, encouraging engagements, and assigning points on the leaderboard if there’s one.
Where it works well
This game can provide insight into topics that interest your participants, how they write their prompts, and how they may currently be using images in their work.
This game provides a good opportunity to
point out biases in generative AI’s training and algorithms, especially as an analogy for discussing textual biases that may often be harder to spot;
coach participants on being specific in their prompting;
coach participants on iterative prompting;
coach participants on how to get around generative AI’s safety guardrails when it refuses to generate images.
What to watch out for
Depending on the topic and on the phrasing of the prompt, generative AI’s safety guardrails sometimes cause it to refuse to generate images.
When there’s high traffic, image generation can be slow and takes up more time than intended.
Many image-generators are now behind paywalls, e.g. Adobe Firefly, Midjourney, ChatGPT.
Authorship
This entry was written by Cecilia Lo.