Generating inclusive images to represent students – Animated Inclusive Personae (Part 1)

by Katie Stripe, Imperial College London.

Developing Inclusive Curricula Using Digital Personae’ (Imperial College London, 2024b) is a workshop run by the ‘Attributes and Aspirations’ (Imperial College London, 2024a) (AA) team based on their work using inclusive personae to make their course more inclusive. This workshop was also run as a CPD webinar for ALT in 2021 (Stripe, 2021). The Graduate School at Imperial wanted to use the theories presented in their workshop in their provision. However, much of their content is delivered as animations. This raised a question around how to source appropriate imagery for different educational scenarios.

The personae created for AA are represented by photographic headshot style stock images, which are hard to source. They also do not offer the flexibility needed for transferral to other scenarios, such as animation. However, bespoke graphics and animations are expensive and have a long development period. This makes them challenging for use in most teaching and learning scenarios.

The Animated Inclusive Personae (Stripe and Meadows, 2024) (AIP) project aims to address some of these issues by developing a solution that, by using templates, will enable any user with minimal training to create an inclusive character. It will also enable them to develop a representative digital image that goes with it. This project started in August 2023. There will be more to share when characters are developed. In this post, we share some of the issues with ‘off the shelf’ content that led to this project.

Stock Photos

The stock photo route has been used so far in the creation of the personae for AA (Stripe, 2024). Due to the nature of the programme, we not only had to find images that were diverse, but also to find images that would be suitable for a LinkedIn profile of our hypothetical students (it is a career skills development programme). This is challenging for a number of reasons and has led to feedback that all our personae look very similar in terms of body shape and style.

Representing ethnicity

Finding appropriate images to represent different ethnicities is challenging. For AA, we use the Articulate 360 content library (Articulate 360, 2024), as it comes as part of the package which we use to develop the content. Searching for ‘Black Male Student’ returned the images shown in fig 1. One of these images is clearly female (and Asian), one of them is white, and one of them is a firework. Some of the issues shown in this selection are created by the way images are tagged and databased rather than the images themselves. Nevertheless, there is limited choice.

a set of images that contains two images of Black males with laptops, two males with lighter skin carrying backpacks, an individual with technical drawings, two black males in suits, a white male with long blonde hair, an East Asian female, and a firework
Figure 1: Articulate 360 search ‘Black male student’

Shutterstock (Shutterstock, 2024a) produces a slightly better array of images (fig 2) for the same search term, at least they are all people and all present as Black males. Nevertheless, the images all show people of a similar body type.

a set of images 11 or which show young Black males is some form of head shot, 3 include computers. One image is a set of vector cartoon graphics
Figure 2: Shutterstock search ‘Black male student’

The ability to purchase vector image cartoon characters does offer an element of flexibility and a range of poses.

Representing gender

Finding images that present as either male or female is relatively simple. However, within the AA programme we wish to be as diverse as possible and required images that do not represent an obvious gender. The first issue to navigate is what search terms to use. Searching for ‘androgynous student’ and ‘non-binary student’ return similar results none of which are appropriate (fig 3), and in the Articulate content library, one of them is a burger.

a selection of images including an individual who appears male, an individual in graduation robes and a hijab (?), three individuals who appear to be female, and a burger with a flag in it
Figure 3: Storyline 360 search ‘non-binary student’

While it is true that anyone of the individuals pictured may use they/them pronouns, if the aim is to show someone that does not present with an obvious gender, then these do not work. As above, this is an issue of image tagging but highlights some significant gaps in the image banks.

Shutterstock (Shutterstock, 2024b), again, produces slightly better results on the same search in terms of diversity (fig 4) but there are very few images of a person on their own and none are really appropriate for the ‘headshot’ image that would be ideal for the purposes of AA. Furthermore, the cartoon style image portrays a very odd body shape and could be seen as perpetuating stereotypes.

a set of images containing a cartoon of an oddly shaped figure, 4 images of groups of people smiling, someone painting, two individual in everyday poses
Figure 4: Shutterstock search ‘non-binary student’

AI generation from a photo

It is possible to create cartoon style images from a photo using AI tools. While this approach would never be appropriate for the AIP project, it is nevertheless worth exploring the graphic styles that could be produced, and looking at the positive and negatives of AI image generation. (, 2024) is an online tool which takes a photographic image and converts it to a variety of different styles, some ‘realistic’ and some cartoon style (fig 5). Below from left to right show the original stock image and the filters ‘Disney’, ‘Kawaii’, and ‘Big Eyes’.

Photograph of an East Asian male, three cartoon images that are all similar but present a similar set of features to the original image
Figure 5: AI generated images

While obviously cartoon images, they all reflect the original image quite well.

AI Nero (Nero AG, 2024) also offers an option to translate a photograph using AI to create a semi realistic digital avatar. The results here are not ideal (fig 6). The avatar generated from the image used above, which in AA represents a student from Singapore, returned an avatar with light hair and blue eyes. Similarly, the image used in AA that represents a student of Black heritage returns an avatar that has a completely different skin tone.

photograph of East Asian male the same as above next a white male with blue eyes and a young Black male next to a young male with similar features but much lighter skin
Figure 6: AI Nero images from photographs

While this was done on the free version, it shows how AI tools can misrepresent racial profiles.

Online avatar creators

There are numerous websites available that offer the ability to create a digital avatar the ones discussed next are those which are free and do not require an account of any kind, although others have been investigated and offer the same general options but on a wider scale, including in some cases the ability to design a body as well as a head.

The first issue is that most tools request you start by selecting a gender. Get Avataaars (Stanley, 2024) does not, it works on a single, generic, head shape and allows you to change the hair, accessories, and clothes within a set of limited parameters. This kind of create your own kit highlights the second issue, which is the limitations of using anything that has defined sets of characteristics.

Get Avataars allows you to change eyes, mouth, and skin tone which allows me to generate a pale, crying, bald man, who is in disbelief (fig 7).

An illustration of a bald white man with a brown beard and glasses crying. There is a single blue tear shape underneath his left eye and his smile is upside-down.
Figure 7: Bald crying man in disbelief

Which may be fun, but with seven skin tones – one of which is Simpson’s yellow – this definitely does not give you the ability to represent a range of students. While in an attempt to create something to represent the two personae shown above produced slightly better results than AI (fig 8) it still does not produce something representative and is certainly limited in the ability to scale up and create more images.

the same East Asian and Black males show above with avatars which reflect their hair quite well but very little else
Figure 8: Get Avataaars images

Avatar Maker (Avatar Maker, 2024) and Cartoonize (Colorcinch, 2024) both work on the same set of parameters and offer a wider range of options than Get Avataaar including 15 head shapes (with options for eyes, nose, mouth and ears). They also offer hairstyles and outfits, but these change, depending on the gender. The main benefit of these creators is the availability of a full colour palette, allowing skin tone and eye colour to be changed by HEX colour codes. Using these tools, I was able to create something that was more representative and with more variety (fig 8), but still limited the headshot style of image.

the same East Asian and Black males show above with avatars which reflect their features quite well
Figure 9: Cartoonize images

Bespoke images

All this leaves us at a point where we have decided to create our own using artists. Watch this space to find out what happens.


Leave a Reply

Your email address will not be published. Required fields are marked *