GenAI image exercise

Students use GenAI to generate images and critically analyse the results.

Purpose

This activity provides a novel and creative introduction to prompt engineering, focusing on image generation models. Students are encouraged to think about ethics in GenAI, especially around bias, accuracy, and refusals. It may be suitable to explore these themes in creative arts subjects.

What will students achieve?

Students will be able to:

  • Design and iteratively refine prompts to generate images.
  • Critically evaluate the reliability and accuracy of GenAI content.

Required resources

Students should be able to access multiple GenAI models that can generate images, such as Copilot, ChatGPT, Canva, or Leonardo.

Activity instructions

  1. Students individually select a GenAI image generator of their choice and prompt it to create an image on a relevant topic.
  2. Students form into groups and discuss the results, noting similarities and differences, and points of interest. Groups then evaluate their results according to:
    1. Accuracy. Is the image an accurate depiction of the topic? Has GenAI reproduced misconceptions about the topic ? Are there any primary sources (photos, films, written accounts) they can check the GenAI generated image against?
    2. Bias. Are there biases present in the image? If there are people in the image, is there realistic diversity in their depiction?
    3. Quality. Is the image well-rendered? Are there errors in the image? Does the style of the image reflect the topic at hand?
    4. Impact. Has GenAI added any unexpected elements? How are visual elements like colour, light, and composition used?
    5. Refusal. Sometimes GenAI models won’t generate images of certain topics. Did the model refuse or agree to generate your image? Was this justified?

Considerations

  • GenAI models can have biases. Discuss with students what they understand bias to be in relation to GenAI.
  • GenAI models can be inaccurate. Discuss with students why this might be.
  • GenAI models can refuse to generate some content, particularly images. Some GenAI models will not generate content that is considered explicit, controversial, or political. Discuss with students why this is, whether they feel it is fair, and how these refusals may relate to bias.
  • This exercise can be used in a number of different disciplines. Students could try generating images of scientific discoveries, abstract concepts, inventions, scenes from works of fiction, metaphors, or scenes from the future   .
  • To generate images, GenAI models are frequently trained on large datasets of existing images and photographs. The companies that train GenAI models do not always have the full consent of the authors of these images before they are used. Discuss the ethics around GenAI training and image generation with students.