This post was created as a support resource for participants of the image generation pilot. It may also be helpful for anyone who’s interested in the applications of generative AI and image generation in education.
We’ve had some fantastic insights shared during our image generation pilot drop-in sessions. Recently, much of the focus has been on prompting techniques and strategies to achieve desired outcomes using various image generation tools. We thought it’d be helpful to share some of these insights with you. In this blog, we’ll explore how to enhance your prompting skills and share a few helpful resources too.
Consider your use case
We’re often asked about prompting and whether it is better to keep prompts simple or to include more detail. We think the answer often depends on your use case. Perhaps you’d like to highlight bias in image generation tools, in which case you can achieve meaningful results with a straightforward prompt like “[Create an image of] a group of police officers/doctors”. On the other hand, if you’re aiming to generate a highly specific image, adding more detail to your prompt can be much more effective. Regardless of your approach, it’s important to consider your purpose for using the image generation tool and the output you’re aiming for.
Making use of generative AI tools
A popular technique involves using a chatbot (Copilot, ChatGPT, etc.) to help with prompting, which can be especially helpful for people looking for guidance or a starting point. That said, it’s also a valuable tool for more experienced users, as the suggestions provided by the chatbot can help to refine and enhance skills further. Various reports on the subject of AI and the environment have found that using generative AI for text to text output is far less energy intensive than for text to image. Using a chatbot to work through your prompt before inputting it into an image generation model is a much more efficient way to achieve your desired end result. If you’d like to read more about AI and the environment, you can do so in our blog on taking a responsible approach.
Within the team, we’ve given this method a try, and found that chatting through what you’d like to create is an excellent way to start. It helps you think through your ideas and consider the details before jumping in.
As an example, I inputted the following into ChatGPT Plus:
‘I’m a textiles student who wants to create some high quality, high fashion images of clothing I’d like to design as part of my project. Can you help me create a prompt to input into an image generation tool? I want the images to be like those in Vogue.’
It then provided me with a template for a prompt that I could customise:
“Create a high-fashion editorial image of a clothing design inspired by [specific theme, era, or concept]. The garment is crafted from [describe fabrics: e.g., luxurious silk, velvet, wool blend] and features [specific details: intricate embroidery, asymmetrical cuts, draping, tailored structure, etc.]. The model poses with elegance, set against a minimalist backdrop with dramatic lighting that highlights texture and craftsmanship. The setting evokes a sense of sophistication, with muted tones that allow the clothing to be the focal point. The style and quality should match a Vogue magazine aesthetic, with attention to detail in every element, from makeup to lighting to fabric texture, capturing an aura of luxury and refinement.”
Within the same output, it also provided this information:
“Tips for tailoring the prompt to your needs:
- Fabrics & Texture: Describe specific textures like “smooth silk with a soft sheen” or “structured wool with a matte finish.” This will help convey the feel and richness of the fabric.
- Colour Palette: Mention colours and tones to define the mood (e.g., “deep emerald and gold for a regal look” or “neutrals with a pop of bold colour”).
- Design Details: Be clear about design elements like necklines, sleeve styles, lengths, or embellishments.
- Mood & Lighting: Describe lighting as “soft and diffused” or “harsh spotlight,” depending on the mood you want (e.g., “ethereal and romantic” vs. “bold and edgy”).
- Background: If you envision a particular setting, like a studio or an urban backdrop, specify that for added context.
Let me know if you want any specific adjustments, like including particular poses, accessories, or additional elements!”
I found this incredibly useful, as it encouraged me to consider aspects of the clothing and imagery that hadn’t occurred to me. This could be especially useful for students in creative subjects, helping them to think through designs, from fabrics to the lighting of an image.
Throughout the pilot, we’ve also heard from photography teachers who have used image generation to help students understand how different lighting setups might impact their work. Returning to my experiment, I edited my prompt to include some of the guidance from ChatGPT.
My prompt became:
“Create a high-fashion editorial image of a clothing design inspired by Tudor England. The garment is crafted from black velvet and features gold intricate embroidery. The model poses, set against a minimalist backdrop with dramatic lighting. The muted tones allow the clothing to be the focal point. The style and quality should match a Vogue magazine aesthetic, with attention to detail in every element, from makeup to lighting to fabric texture, capturing an aura of luxury and refinement.”
This was my result, which I was really happy with:
It’s also worth noting that tools like Adobe Firefly and Canva offer features that allow users to modify and change elements of images once generated. This is another great way to encourage creativity and think more about the style of image you’d like to produce. More information can be found in our respective blog posts about these tools, and can be accessed on our Image Generation Pilot Resource Bank post.
Further prompting advice
Here is a roundup of further guidance that has been shared so far.
Dave Holloway from the University of Sheffield shared some excellent prompting ideas with us. We received a question about the best approach for creating a consistent style of image for a comic book. Dave recommended experimenting with prompts using numerous different styles and influences until you find something that closely matches your vision. He also shared that many tools will generate prompts for you from images you have created or uploaded but, if you’re using a tool which doesn’t do that, simply Googling ‘prompt generation from image’ will bring you to a number of free services that will generate prompts.
Once you’ve developed a prompt that works for you, you can maintain the same prompt structure while changing the subject to suit. Dave’s example was using Midjourney, but I gave this a try using ChatGPT Plus.
My initial prompt was “Create a black and white illustrated style image of a small cartoon dragon with an evil, adventurous personality standing in front of a castle.”
I then asked:
“Now, create a black and white illustrated style image of the same small cartoon dragon with an evil, adventurous personality standing in front of a forest.”
My results aren’t perfect, but I think it’s done a good job. The character it has created isn’t exactly the same, but it has kept the style relatively consistent. It provided some great inspiration, rather than providing a finished product, which is arguably a useful use case in itself.
Ideas for structuring your prompts
Here are some examples of how you can structure your prompt by following a specific format.
- Prompt example: ‘Stock photograph of three university students conducting a chemistry experiment in a university science lab. Diverse mix of gender and ethnicity. shallow depth of field.’
I find that the key things to include in your prompt are…
- Type of image: (Stock photo, candid photo, editorial photo…)
- Subject: (Three university students…)
- Background: (university science lab)
(This particular example was used when creating images in Ideogram, and can be read more about in our blog post).
2. Use this acronym to remind you of what to include in your prompt:
- I: Idea – Define the main concept or theme for the image (e.g., “A bustling city market at dawn”).
- M: Mood – Specify the mood or atmosphere (e.g., “Lively and vibrant”).
- A: Angles – Indicate any specific angles or perspectives (e.g., “Bird’s-eye view of the market”).
- G: Guidelines – Offer specific adjustments or refinements (e.g., “Ensure the stalls are colourful and crowded”).
- I: Inspiration – Provide examples or references to guide the AI (e.g. “In a vibrant night sky with glowing stars and a peaceful village below”).
- N: Notables – Include key elements and details (e.g., “Include street performers and food vendors”).
- E: Effects – Specify lighting, shadows, and other effects (e.g., “Soft morning light with long shadows”).
3. Follow these steps:
- GOAL > CONTEXT > SOURCE > EXPECTATIONS
i.e. create an image > of a malamute in space > using [website]/> in a cartoon style
Useful resources
- The DALLE 2 prompt book pdf
- More general prompt engineering but could be applied to image generation too: Learn about Copilot prompts – Microsoft Support
- How to Make AI Illustrations | Microsoft Copilot
- Repository – Hwb: Toolkit for AI driven blended learning
Find out more by visiting our Artificial Intelligence page to view publications and resources, join us for events and discover what AI has to offer through our range of interactive online demos.
For regular updates from the team sign up to our mailing list.
Get in touch with the team directly at AI@jisc.ac.uk