Illustrating eLearning with AI: 3 Tips for Prompting Beautiful and Consistent Image Sets

Illustrating eLearning with AI: 3 Tips for Prompting Beautiful and Consistent Image Sets

Following up on a post yesterday where we shared our “Fruit or Veg” mirco-learning exercise, we'd like to offer some reflections from using #midjourney to generate lovely and consistent feeling images for the Rise lesson and video explainer. Note, we don't suggest using Midjourney images for commercial purposes!

A light yellow banner flanked by caution symbols that reads "We are not currently using AI tools (Midjourney, ChatGPT, etc.) in client projects.  We are currently only using these tools for research purposes."

In our “Fruit or Veg” mirco-learning exercise, we used Midjourney to generate all the visual assets. Once you decide on a style, how can you develop prompts that give a reliable and consistent output? In the process, we learned 3 lessons worth passing on to other folks exploring these technologies, today.

1. Try to keep your prompt simple

(*this tip applies to Midjourney 4 and Midjourney 5.1, not so much true for MJ5)

Many experts will encourage you to be as specific as possible, with long prompts that contain lots of information. While this is super helpful for generating one specific image, I found that it did not produce consistency across a set of images. Instead, choosing just 2-3 terms that describe the style was more helpful.

No alt text provided for this image
A simpler prompt generates more stylistically consistent results.

2. Be flexible when necessary

If you can’t get one specific image to work, consider editing the content if there’s wiggle room there. We had a hard time with watermelons, so we didn’t use them in any of the examples. There are times when you need a specific image and it's worth fishing 100x to get it. But in cases like this, don’t beat your head against the wall trying to get a specific fruit if it can be swapped out without harming the objective.

4 illustrations that show variations of the theme "a boy eating a watermelon.” They are humorous and nonsensical. Three of the images show a watermelon-person, sometimes eating watermelon and often surrounded by watermelons. In one case, the boy’s face is peeking through a huge watermelon slice and he is eating a smaller watermelon slice on the end of a stick.
Some fun watermelon fails.

3. Lean into the magic

Midjourney will /imagine things that do not exist. This can be frustrating when you are hoping to get a specific image of a kid eating a watermelon. But, this can also inspire some imagination of your own. We decided to build a final matching activity where users apply what they learned by sorting imaginary fruits and vegetables. Try to consider how you can use the unpredictable nature of AI to your benefit.

There are three square images in a row, each image has 5 imaginary fruits and vegetables. For example, one looks like a cross between grapes and a lime. Another looks like a mashup of an eggplant and an ear of corn.
The interesting outputs occur when you let AI use it’s “imagination”

Have you used AI to produce image sets for educational purposes? What was the outcome? What challenges did you face? What lessons did you learn??Let us know! And, don't forget to check out our micro-lesson on the botanical difference between fruits and vegetables!

#AI #ELearning #BlendedLearning #LearningDesign #TechForAll #Education #Innovation

要查看或添加评论,请登录

Integral Ed Services的更多文章

社区洞察

其他会员也浏览了