Artificial Intelligence and the Teacher Image? “That’s Funny, You Don’t Look Like a Teacher” ()
1. Introduction
People spend a considerable amount of time consuming visual images on a daily basis, particularly since the Internet was established in 1992 (Comenius, 2016; Sturken & Cartwright, 2001). Comenius (2016) insightfully noted that, “Visuality and human being [sic] are the interconnected vessels” (p. 193). Images are powerful and have a critical influence upon people’s thoughts, perceptions, understandings, and opinions. Pictures indicate what people esteem and what societies appreciate, implying human beings’ values, needs, and desires (Weber & Mitchell, 1995). The visual, as Martikainen (2020) pointed out, is a primary part of humankind’s commonsense knowledge. In countless political regimes, including such notable ones as the Soviet Union or Nazi Germany, in order to maintain control of their people, those in power employed tightly controlled still and moving imagery, which they censored or altered to manipulate, persuade, and influence public understandings and actions (Barroso, Estrada, & Toldy, 2020; Cohen, 1991). What are the reasons for this? Some are to bend the truth, to convince, to make ideas appear in a different light than reality, to erase undesirable experiences for propaganda, and for glorification or vilification of particular people (Barroso, Estrada, & Toldy, 2020). Moreover, pictures tell stories and shape understandings of self and others (Weber & Mitchell, 1995). In the last century, mass media—including such genres as film and television—saw a proliferation of research regarding media’s influence on people and society (Nerlino, 2023). There has been a particular fascination regarding the representation of teachers in television and cinema notable with numerous studies produced by such researchers as Ahn & Leggo (2019), Bullman (2014), Butler, (2000), Cap & Black (2014), Crume (1988), Dalton (2005, 2010), Ewing, Ewing & Copper (2021), Maruri (2012). Others, like Alhamdan et al. (2014), Martikainen (2020), and Weber & Mitchell (1995), approached the pictorial representation of educators using a wider lens, analyzing not only television and film but also other visual media genres, including fiction, comic books, published articles, newspapers, and even dolls. From this analysis, many findings about the mass media portrayal of teachers prove interesting. These images are not innocuous: not only do these portrayals affect the public perspectives of teachers and teaching (Alhamdan et al., 2014; Nerlino, 2023) they also affect and shape teachers’ attitudes of themselves as educators and of their perspectives regarding their own teaching profession (Ahn & Leggo, 2019; Beauchamp & Thomas, 2009, Buchanan, 2015; Mockler, 2011; Nerlino, 2023). Social science and education researchers Weber & Mitchell (1995) wrote that:
Images are constructed and interpreted in attempts to make sense of human experience and to communicate that sense to others. Images in turn become part of human experience and are thus subject to reconstructions and reinterpretations. While images always maintain some connection to people, places, things, or events, their generative potential in a sense gives them a life of their own, so that we not only create images, but are also shaped by them. (p.21)
Of particular interest to Weber & Mitchell (1995) are the images created by these educators; their findings included discussions of teacher stereotypes, attitudes, and identity issues related to age, gender and race. With the proliferation of artificial intelligence (AI), we have turned our attention to visual arts teachers as depicted by AI technologies. We ask, “Have teacher depictions changed with the proliferation of generative AI?”
2. Teacher Imagery
2.1. Teacher Imagery Created by Artificial Intelligence (AI)
We are living in an era wherein artificial intelligence is spewing out imagery. Shortly after ChatGPT launched in November 2022, users of the image generator DALL-E generated more than 2 million images per day (Ali et al., 2023, p. 11). Although it is difficult to know for sure, according to Valyaeva from Everypixel Journal (Valyaeva, 2025), it is estimated that people are using a variety of generative artificial intelligences (GAIs) to create approximately 34 million images daily.
Civilization is entering a new technological epistome, leaving far behind the Gutenberg industrial age, the mass media era, and even the digital period, as we are now entering a new technological artificial intelligence epoch (Black, 2023). According to Buchanan (2015), it is a new epistemology, and because of this, it is important to examine, comprehend, and sustain ways educators can foster and build their own comprehension towards a conception of what constitutes a professional educator in our AI times. We turned to the seminal book by Weber & Mitchell (1995), entitled “That’s Funny, You Don’t Look Like a Teacher”, as a basis for studying new imagery and stereotypes of teachers created by artificial intelligence to see if they have changed or remained the same.
2.2. “That’s Funny, You Don’t Look Like a Teacher.” Past Non-AI Teacher Imagery
Weber & Mitchell (1995) analyzed the images of teachers created by mass media (including movies, television, books, cartoons, and dolls) as well as 600 international drawings of educators created by children, teachers in training, and professional educators, particularly honing their research focus on stereotypes and teacher self-identity formation. Mitchell & Weber acknowledged at that time that few scholars were looking at teacher cliches in relation to self-identities1 (Weber & Mitchell, 1995: p. 27). Studying these images, they specifically examined cultural markers including gender, race, and appearances, in order to scrutinize the paradoxical educator image. Critical questions were asked, particularly about ways in which mass-produced texts created constructions of ideas pertaining to gender, schooling, authority, and race (Weber & Mitchell, 1995). They studied images of teachers’ poses, gestures, dress, and what they term, “markers” (desks, pointers, apples, books, in fact any background items and props) that are a visual vocabulary used to delineate educators. Also, they considered the cumulative cultural texts, which are pictures of the present affected by past imagery, as well as the image as an intertextual and intergenerational text. Looking at the stereotypes that are used to construct the image of the teacher socially, culturally, historically, and at the personal and societal levels, they revealed constructed knowledge and feelings about teachers and their profession. Weber & Mitchell (1995) describe diverse examples of stereotypes ranging from the dreary, asocial, to the beautiful, elegant, and trendy female educator and contrasting the elementary feminine “soft” teacher to the secondary level, male “tougher” ones.
3. Perspectives on Teacher Stereotypes
Perspectives regarding teachers are complex: there is a far-reaching range from the exceptionally positive to the extremely negative, varying from, for example, the teacher as saint, hero, to that of buffoon and drug addict. Firstly, we will address the positive image of the educator.
Going back further than one hundred and fifty years ago, as far as the 1870s, there have been countless portrayals of good, morally upright, dedicated teachers depicted in such television shows as Little House on the Prairies (Landon & Friendly, 1974-1983) set in the United States or the series such as Anne of Green Gables (Walley-Beckett, 2017-2019) that took place in Canada. Indeed, this concept of the good teacher carries through to such films as Goodbye Mr. Chips (Jacobs, 1969), situated in Great Britain in the 1920s, which has a depiction of a persnickety, surly teacher (Mr. Chips himself) whom the audience watches on his career-path trajectory as he matures to eventually become admired and well-loved by his students. Not much has changed in the depictions of teachers as we jump to the 1950s and 1960s with educators who are challenged yet remain dedicated, caring individuals, as is apparent in such movies as Dead Poets Society (Weir, 1989), in which the protagonist, John Keating, inspires his students through the teaching of poetry. Similarly, in the film Mr. Holland’s Opus (Herek, 1995)—set from 1964 to 1995 in America—Mr. Holland is a musician and teacher: this film is a depiction of his struggle over thirty years, in which he is portrayed as being conflicted between his time working as a musician trying to produce a symphony and his teaching. At the end of the movie, with cuts to the arts, Mr. Holland is finally laid off. He ends his career with a touching scene in which the Governor of Oregon gives him recognition as an outstanding educator. Additionally, in this final film part, there is the display of evident adulation from his students when Holland finally presents a live performance of his own work to his enrapt, supportive audience of pupils from his thirty-year career. Thus, it is apparent that, for over a hundred-year period, teachers like these have been depicted in films and television as inherently good. They are filled with knowledge (Martikainen, 2020; Weber & Mitchell, 1995), are hardworking, stalwart, possess a moral compass, and, as such, are role models who are thought of by students as so dedicated that they never leave their schools (Alhamdan et al., 2014; Weber & Mitchell, 1995). Teacher caring is mentioned (Alhamdan et al., 2014), that is often kind, selfless, and sensitive in which educators have compassion leading to the perspective of them being saints, rescuers, protectors, and saviors who go above and beyond for students and for the schools within which they work (Butler, 2000; Black & Cap, 2014; Dalton, 2005, 2010; Ewing, 2021). This media depiction can romanticize the teacher as a superhuman entity: a person impossible to be in reality (Ewing et al., 2021). At other times, teachers are portrayed as renegades pushing back against societal expectations, school policies, and traditional, outdated teaching methods as aptly delineated by Dalton (2005 & 2010).
Weber & Mitchell (1995) and Tsaousi (2019) also scrutinized educators’ clothing and appearance. Teachers have often been portrayed as prim, proper, and concerned about what is in the mind over their appearance (Tsaousi, 2019). As such, the drab, shapeless, colourless female clothing and nondescript, scruffy, ill-fitting male clothing as described by Weber & Mitchell (1995) are often conjured in the mind when thinking about “the teacher”. However, Weber and Mitchell do discuss alternative counter-cultural images to the ones described above that sometimes appear in media: this is of the teacher as cool, beautiful, and enticing. They wrote:
There are and always have been many dramatic and colourful exceptions that provide unconventional responses to the question “What do teachers wear?” The fuchsia shirt of the substitute teacher in the girls’ book Barnyard Battle, the elegant, soft clingy fabrics, and draping silk scarves worn by Jean Brodie in the movie The Prime of Miss Jean Brodie, the impeccably tailored suits worn by Mark Thackeray in To Sir With Love, the real-life teacher you once had who was beautiful or handsome or fashionable or eccentric or colourful—they all momentarily rupture the image of teacher as asexual and drab, offering alternative possibilities: Teacher-as-cool; teacher-as-sex symbol/object; teacher-as-goddess; teacher-as-hero; teacher-as-gentleman; teacher-as-rebel, and teacher-as-artist. (1995, p. 59)
As the authors point out, these alternative depictions are exciting, positive, romantic teacher portrayals and are, unfortunately, the rare exceptions.
Secondly, we turn to negative teacher portrayals. Over recent years, the teacher image has increasingly become pejorative (Ahn & Leggo, 2019; Alhamdan et al., 2014; Ewing et al., 2021; Nerlino, 2023). Researchers have noticed this swing towards a darker vision of educators and the teaching profession. Ewing et al. (2021) points out the damaging, disapproving portrayals from the 1960s onwards, where teachers are no longer worthy of being emulated; they are ineffectual. By the 1980s, the media image was increasingly distorted in terms of teachers themselves, the schools within which they work, and the profession as a whole. It is noted that educators are increasingly portrayed as incompetent, less attractive, and far less professional than other professional workers portrayed in television and film, such as police officers and lawyers (Ewing et al., 2021). Alhamdan et al. (2014) asked why there is a preponderance of negative teaching imagery in the media.
Ewing et al. (2021) attributed the hostility towards teachers to distrustful, eroding relationships between society and its educators, which is often aggravated and intensified by governments and those in power who control the media. Evident outright hostility and cynicism towards educators and the profession are apparent in government policy and in media texts. Negative representations abound concerning teachers’ unprofessional behaviour, misconduct, lack of dedication to teaching, inefficient functioning, insincerity, or scant desire to make important sacrifices. Ewing et al. (2021) observed that harmful media depictions of teaching over the last half century have been notably escalating, particularly over the last twenty years2. Because of this undermining situation, Nerlino (2023) called the teaching profession “controversial”: texts about teachers and their profession are, it is argued, “disseminating a crisis mentality that comes from only negative images of teachers, blaming them for students and schools not performing up to par” (p. 282). This perspective is not just located within North America or the West but is more international. In a study of South Africa, Australia, Oman, Saudi Arabia, and Bangladesh, the authors found upon examining newspaper images of teachers that there was a focus on the pejorative aspects of educators’ behaviour, demeanour, morals, and principles (Alhamdan et al., 2014).
What are the many representations of the bad teacher? One is of the overly stern, traditional, authoritarian, joyless person who, to use a cliché, “shoves learning down students’ throats” (Ahn & Leggo, 2019; Tsaousi, 2019; Weber & Mitchell, 1995). Sexuality is a concern in the media. Others have portrayed teachers—particularly female ones—as ranging from asexual (Tsaousi, 2019; Weber & Mitchell, 1995) or perversely sex-crazed (Butler, 2000; Ewing et al., 2021) to the traditional spinster and bossy matron (Butler, 2000; Weber & Mitchell, 1995) who is dowdy, shoddy, and careless. Weber & Mitchell (1995) observed that the audience often reacts to these portrayals with disinterest, contempt, or pity. These fictitious images, Weber & Mitchell (1995) argued, are so pervasive that they often override the public’s real experiences of teachers, which are far more colourful and favourable.
Another issue is the undermining of teacher expertise. Some depictions are of a docile, innocuous, boring, harmless white middle-class teacher replete with respectability who possesses, if a woman, “nice” femininity and impeccable neatness (Weber & Mitchell, 1995). More destructive teacher portrayals of declining skill and knowledge levels, for instance, are depictions of teachers as buffoons (Butler, 2000; Cap & Black, 2014; Ewing et al., 2021) or absent-minded bumblers (Weber & Mitchell, 1995). Further undermining teachers’ expertise are representations of their character and position. Teachers have been portrayed as being powerless, inept, dull, backward, and poorly trained (Butler, 2000; Ewing et al, 2021). Other delineations attack personality. For instance, they can be nerdy (Swetman, 1992; Weber & Mitchell, 1995), a bully (Cap & Black, 2014), a jerk or clown (Ewing et al, 2021; Swetnam, 1992). On a lighter note, representations of the teacher as a prankster (Martikainen, 2020) and a fool (Ewing et al., 2021; Martikainen, 2020) are now common. Recently, media depictions have become even darker with drug addicted teacher portrayals (Ewing et al., 2021) as seen in, for instance, the notable depiction of Dan Dunne by the actor Ryan Gosling in the movie, Half Nelson (Fleck, 2006). Finally, faced with increasing budget cuts to education in North America, it is not uncommon to have depictions of the poor, financially struggling teacher, as is evidenced, for instance, in the movie Bad Teacher (Kasdan, 2011), wherein the teacher resorts to devious means in order to acquire money for her breast enhancement. Nerlino (2023) quoted a teacher who comments on educators’ financial struggles, reflecting that this “…image of a teacher as an ‘invisible’ and ‘low class’ worker in particular de-professionalizes teachers’ work, removing the more reflective and intellectual dimensions required” (2023, p. 297). Nerlino (2023), Martikainen (2020), Swetnam (1992), and Weber & Mitchell (1995) pointed out that many teachers do not see themselves reflected adequately or well in the media portrayals they have viewed. In summary, these negative portrayals of teachers, which many teachers view as unrealistic and not reflective of their day-to-day experiences, attack their sexuality, appearance, personality, expertise, and socioeconomic levels.
Finally, we would like to discuss the COVID-19 pandemic. Nerlino (2023) looked at teachers’ depictions during COVID-19 and noticed that at the beginning of the pandemic, educators were looked upon favourably as heroes working within the new life structures of the pandemic. They dropped face-to-face learning, and many had to struggle to teach online at a moment’s notice in March 2020. During the pandemic, when many teachers were forced back into schools, they heroically faced possible illness themselves, long hours of cleaning, and taking necessary health precautions. However, as time passed, this positive perspective towards teachers quickly turned negative again: Nerlino (2023) discussed the creeping back of the disapproving, deleterious viewpoints regarding educators into popular discourse: specifically, teachers were again being depicted as, for example, being lazy, wanting to avoid their duties, and shirking their responsibilities.
What the literature makes clear, taken together, is that teacher stereotypes in mass media have proven durable. From the morally upright schoolmarm of the nineteenth century to the burnt-out, financially struggling teacher of recent decades, the range of depictions has shifted in tone while remaining constrained by a recognizable set of recurring figures. Race, gender, age, dress, and the visual markers surrounding teachers in these representations have changed little, even as the media contexts producing them have multiplied and diversified. As Nerlino (2023) observed, these portrayals shaped not only public perception of teachers but teachers’ own sense of professional identity, a finding that Weber & Mitchell (1995) anticipated in their description of the cumulative cultural text, the slow accretion of imagery so thoroughly absorbed that it often overrides direct experience of real teachers and real classrooms.
GAI-generated teacher imagery enters this long history at the current moment. Unlike film or television, GAI systems do not construct images through deliberate creative choices; they reconstruct them statistically from the vast body of visual material on which they were trained. In this sense, what a GAI produces when prompted to generate an image of a teacher is not an interpretation of a teacher but a probabilistic distillation of how teachers have already been visualized across the breadth of existing visual culture. Whether that distillation reproduces, intensifies, or, in some ways, shifts the stereotypes documented in the literature is not self-evident, and it is precisely this question that motivates the present study. We ask: to what extent do GAI-generated images of visual art teachers perpetuate or depart from the teacher stereotypes established in mass media?
4. Methodology
We examined GAI-generated imagery of visual art teachers using qualitative research methods drawn from the work of Human Geographer Gillian, Rose (2016), Visual Methodologies, specifically compositional interpretation as our primary analytical method, informed by her broader critical visual framework. Compositional interpretation involves attending carefully to the content and form of images, their figures, colours, spatial organisation, and expressive qualities, in order to produce a close reading of what an image shows and how it produces meaning (Rose, 2016). Rose’s critical visual methodology extends our attention to the social work that images do: the categories they construct, the assumptions about professional identity they encode, and the particular visions of who teachers are and what they look like that they present as natural or given. Complementing our compositional interpretation methods was simple frequency-based analysis. To be clear, our intention was not to conduct a quantitative study, but rather to see what patterns emerged in the baseline data of age, gender, and race found in our tables in the findings section.
4.1. Image Generation
To build our image dataset, we collected the data in January 2025 using three GAI platforms: DALL-E 3 (OpenAI, San Francisco, CA) as well as Meta AI (Meta Platforms, Menlo Park, CA). (Note that Meta AI does not publicly expose specific version numbers for its backend image generators. Instead, the company utilizes continuous, silent updates.) The third platform we used was Stable Diffusion 3.5 (Stability AI, London, UK). While collecting the data we prompted the system 25 times and there were no default settings or “re-rolls” used in our process. These are machine-learning models that use diffusion processes rather than traditional generative adversarial networks to produce imagery. The process works by decomposing visual information from large-scale training datasets and then reversing that decomposition to reconstruct and reform images. What each GAI produces is therefore not invented from nothing but statistically derived from the vast body of imagery on which it was trained, making the images it generates a direct index of the visual assumptions embedded in its training data (Diamant, 2025; Jayachandran & Arni, 2023).
The type of prompts used to generate imagery is important. We were consistent in using the same prompts as follows: 1) An elementary school art teacher; 2) A junior high school art teacher; 3) A high school art teacher; 4) A fine art university professor. Each of the three researchers generated 25 images per prompt category, with each GAI, for a total of 300 images. All image codes were recorded in a shared Google Sheet. We wanted to determine what type of imagery would be generated based on the educational level, the type of prompt, and the AI we used. The imagery was generated for 25 pictorial representations in each category. Even though we gathered diverse media depictions, ranging from paintings and drawings to cartoons of teachers, using Stable Diffusion, Meta AI, and DALL-E. After systematically reviewing imagery across multiple visual modes, which included cartoons, drawings, and paintings, we narrowed image generation to photorealistic depictions. Photorealistic imagery most closely approximates the kinds of visuals in which teacher representations circulate in contemporary life, including news media, institutional websites, and stock photography. It is through photorealistic images that stereotypes of teachers are most actively produced and consumed, and it is here that they most directly shape how teachers are perceived by the public and how teachers may come to perceive themselves (Weber & Mitchell, 1995). Generating photorealistic images, therefore, created the most direct basis for comparison with the stereotyped representations described in the literature.
4.2. Coding and Analysis
Compositional interpretation requires a close and systematic description of an image’s content before interpretive claims can be made (Rose, 2016). Our first round of coding applied basic demographic descriptors to each image, identifying perceived age, race, gender, and age. We approached these differently. In terms of our discussion of race, we referred to the Chicago Face Database by Ma, Correl, & Wittenbrink (2015) for demographic information. We used the Chicago face database (CFD) to discern between white and nonwhite persons. It was developed by the University of Chicago, and this database is comprised of 597 images of unique people consisting of self-identified Black, White Latino, Asian male and female models. These people were all recruited in the United States of America. The facial expressions of the images are neutral. A particular subset of these representations is also presented with cheerful (open mouth), content (closed mouth), irate, and apprehensive/scared countenances. Gender differences in the data were very easy to discern. We never encountered a problem with the clear visual demarcations we examined in our research. Finally, in terms of discerning the age of the person constructed in AI imagery, if there was a discrepancy, we took an average of our three tabulations from the three researchers.
Alongside written observational notes we made notes concerning composition, dress, setting, pose, and what Weber & Mitchell (1995) call markers such as the background items, props, and visual vocabulary used to signify the teacher. From these notes, we identified recurring signifiers across the dataset, such as the yellow dress frequently appearing in Meta AI-generated images of elementary teachers and the disproportionate presence of rainbows in depictions of male teachers (refer to Figure 1). These recurrences formed the basis for iterative, targeted coding in subsequent rounds.
Each set of images generated by each platform was coded independently by two of the three authors. Coding discrepancies arose when two coders assigned different classifications to the same attribute. For example, one coder classified a depicted teacher’s age as 50 - 59, while the other classified the same figure as 30 - 39. Each image was assigned a single code by majority agreement across coders. When coders disagreed on age, and there was no majority, the older age band was taken. Aside from coding discrepancies, we also noted anomalies, including images in which the GAI produced departures from the dominant patterns in the dataset, such as a teacher depicted with visible tattoos or a nose piercing. These were flagged during coding and set aside for interpretive discussion during the analysis phase, as they proved significant in what they demonstrate about the sources that GAI draws from.
We then merged the coded data across the three platforms to identify patterns and anomalies in the full dataset, with particular attention to consistencies and divergences across DALL-E, Meta AI, and Stable Diffusion. Differences were important because they illustrate how the distinct training data and algorithms of each GAI shape the images produced, rather than treating GAI output as uniform or neutral. We interpreted these patterns in dialogue with the literature on teacher stereotypes reviewed above, and specifically with Weber & Mitchell’s (1995) framework of cumulative cultural texts, as a basis for assessing whether the historical constructions they documented persist, shift, or are reproduced in AI-generated imagery.
5. Findings
We again pose the problem, “How have teacher depictions changed with the proliferation of teacher AI images?” We found that the portrayals of educators have not fundamentally altered from the constructs described in the literature review, especially those of female teachers. Tsaousi (2019) described the prim and proper teacher, and Martikainen (2020) and Weber & Mitchell (1995) the good, stalwart, kind educator, which we found is the general portrayal in the AI images of women teachers. Nevertheless, each AI generator posed particular delineations that differed from each other.
Figure 1. Depictions of art teachers with rainbows.
Women educators were represented as white, able-bodied, and typically young. Much like the proportion of men to women, the proportion of young white women represented in the data varies significantly in different GAI models. They were smiling and looking directly at the viewer without pupils in sight. We found a preponderance of young women wearing yellow dresses found in Meta AI, particularly at the elementary level. Out of the 25 images of early years educators, 19 were found wearing this colour, specifically yellow dresses, in comparison to Stable Diffusion and DALL-E, which had one early years educator each wearing yellow sweaters. Tsaousi (2019) notes that dress is directly related to identity and bodily practice, and with teachers, it has connections to morality, including “good”, appropriate, and acceptable. In reference to typical colour symbolism, yellow is a hue in Western society signifying “value” (noted its relationship to gold) and “beauty” (St. Clair, 2016) This aligns with the stereotype of the “good” teacher, one who is unencumbered with the negative aspects of teaching, such as large class sizes, lack of resources, a large quantity of administrative duties and extra work behind the scenes such as copious marking and curricula planning. Unlike in former traditional film and media depictions, there were no negative depictions of teachers, such as the depressed teacher or one who lacks money for basic necessities; these depictions show only “happy” portrayals of educators.
Male teachers were mainly depicted as bearded, able-bodied white males. At the university level, in DALL·E, most of the men were portrayed as the same person; however, just under half the males in MetaAI and in Midjourney diverged from the depicted smiling image, often portrayed as being extremely serious. Nevertheless, regarding the male teachers’ representations, there are notable changes in imagery. Both male and female art educators were often depicted wearing or having in the foreground/background polychromatic rainbows, colour wheels, and palettes as imagery. Although there is no clear meaning, we suggest that this is the GAI’s interpretation of colour wheels and palettes expressed in unusual forms that are typically equated with art teaching.
We found a difference in age between the elementary and university depictions of teachers. In the elementary and middle years, both males and females were portrayed as younger, between 20 and 35 years old, whereas in the senior years, they were depicted as in their 30s to 40s, and in university, there was a preponderance of professors aged 50 - 59.
5.1. Gender
The gender distribution in Meta was skewed heavily towards men in the university art teacher depictions, at 88% men and 12% women. Whereas, according to Statistics Canada (Government of Canada, Statistics Canada, 2025a), 43.7% of full-time faculty are women, these depictions do not match the reality of the gender composition of university professors in Canada (see Table 1).
In Meta, there was a significantly higher proportion of women in all other years, 76% in high school, 68% in middle school, and 88% in elementary. This reinforces the idea that grade school teachers are predominantly women. According to Statistics Canada (Government of Canada, Statistics Canada, 2025b) in 2023/24, 75% of grade school teachers were women. Therefore, these percentages are fairly accurate to the reality of grade school teachers in Canada.
DALL-E, unlike Meta, depicted 100% of both university professors and high school teachers as male, and 96% of middle school teachers as male. For the university and high school teachers, the Gen AI generated what appeared to be the same man multiple times (more discussion on this below), which may account for some of the gender distribution. In the elementary school data, it depicted 60% women and 40% men, having a significantly higher percentage of men than would statistically be seen in the profession.
Stable Diffusion had the highest percentage of women portrayed across all categories, with 76% of university professors, 96% of both high school and middle school teachers, and 100% of elementary school teachers depicted as women. This is much higher than the real-life percentage and may reflect the stereotype that teachers are predominantly women.
Table 1. Gender distribution in AI-generated images (n = 25 per row).
AI Model |
Male |
Female |
Meta-University |
22 (88%) |
3 (12%) |
Meta-High School |
6 (24%) |
19 (76%) |
Meta-Middle School |
8 (32%) |
17 (68%) |
Meta-Elementary |
3 (12%) |
22 (88%) |
DALL-E-University |
25 (100%) |
0 (0%) |
DALL-E-High School |
25 (100%) |
0 (0%) |
DALL-E-Middle School |
24 (96%) |
1 (4%) |
DALL-E-Elementary |
10 (40%) |
15 (60%) |
Stable Diffusion-University |
6 (24%) |
19 (76%) |
Stable Diffusion-High School |
1 (4%) |
24 (96%) |
Stable Diffusion-Middle School |
1 (4%) |
24 (96%) |
Stable Diffusion-Elementary |
0 (0%) |
25 (100%) |
5.2. Ages
Similar to Weber & Mitchell (1995), we found a wide range of ages regarding teacher portrayals: some are young, some are middle-aged, and some are in their 50s and 60s. (This is apparent in Table 2). Representations of young teachers were abundant across all three AI technologies. We found many young teachers (almost too young), as identified by Stable Diffusion and Meta AI, who looked like senior, middle, or elementary-year students themselves.
DALL-E represented university teachers as being definitely older, with 64% aged 50 to 59 and 32% over 60 years old. In contrast, at the high school level, teachers’ ages were younger, with 84% of them portrayed between the ages of 30 and 49. At the middle school level, the age again shifted downward: 100% of teachers were represented in the age range of 20 to 39 years old, and at the elementary school level, 80% of teachers were in the 20 to 39 age demographic. DALL-E did not produce any teacher images looking under 20 years of age.
Stable Diffusion portrayals were somewhat different. 58% of university professors were in the older demographic of 40 to 59 years old, and 36% were in their thirties. At the high school level, a marked shift occurred as 58% were between the ages of 20 and 29. Oddly, at the middle school year level, 60% of teachers were delineated as under 20 years old, and at the elementary level, the percentage increased to 80% of educators being in the under 20 age range. Of note, it is very difficult to become a professional teacher before the age of twenty, as most educators in K-12 education need university training in order to be hired, which would mean that they would be at the earliest in their early twenties when they begin their teaching careers. It is evident that these AI teacher portrayals are markedly askew in the years between elementary and middle school, with this unrealistic younger representation produced by Stable Diffusion.
Like Stable Diffusion and DALL-E, Meta AI delineated university professors within an older demographic. 64% of the professors portrayed were between 50 and 59 years old. Of note, 12% were a decade younger. At the high school level, the age shifted a decade downward: 48% of the teachers were between the ages of 40 and 49, with 36% being between 20 and 29 years old. In contrast to this, 56% of middle school teachers were between 20 and 29 years old, and finally, at the elementary level, 60% of the teachers were young, between 20 and 39 years old. Ane 4% of teachers appeared to be under 20 years of age.
Government of Canada, Statistics Canada (2025b) found that the proportion of educators in public elementary and secondary schools is very small: 2.9 % are under twenty-five years of age. Between the ages of 25 and 29, more teachers enter the teaching profession, and 9.1% of Canadian teachers fall within this age range. Canadian teachers do not start teaching in their early 20s: they are somewhat older, and we argue that the reason is, as stated earlier, that they need time to obtain their teaching qualifications in order to be hired. However, all three AI technologies are not reflective of this very real situation. Rather, they are portrayed as far younger teachers teaching in the early, middle, and high school levels in comparison to the older educators who are actually working within the field.
In summary, (as is illustrated in Table 2) it was found that the higher the education level, the older the portrayal of educators; and vice-versa, the younger the teaching level, the younger the educator. A general shift from early years teacher portrayals being young to university professors being older is evident. Too many young teacher artificial intelligence portrayals were out of sync with reality within schools, where, in general, the higher demographic, according to Government of Canada, Statistics Canada (2025b), for early to secondary school level teachers is in the range of 30 - 59 years old. In addition, Stable Diffusion and Meta AI were completely out of sync with reality in terms of representations of educators in middle and elementary schools, who often looked like students themselves in these teacher portrayals.
Table 2. Age distribution in AI-generated images (n = 25 per row).
AI Model |
Under 20 |
20 - 29 |
30 - 39 |
40 - 49 |
50 - 59 |
60+ |
Meta-University |
0 (0%) |
0 (0%) |
3 (12%) |
3 (12%) |
16 (64%) |
3 (12%) |
Meta-High School |
0 (0%) |
9 (36%) |
3 (12%) |
12 (48%) |
1 (4%) |
0 (0%) |
Meta-Middle School |
3 (12%) |
14 (56%) |
3 (12%) |
2 (8%) |
3 (12%) |
0 (0%) |
Meta-Elementary |
1 (4%) |
6 (24%) |
9 (36%) |
8 (32%) |
1 (4%) |
0 (0%) |
DALL-E-University |
0 (0%) |
0 (0%) |
0 (0%) |
1 (4%) |
16 (64%) |
8 (32%) |
DALL-E-High School |
0 (0%) |
0 (0%) |
6 (24%) |
15 (60%) |
4 (16%) |
0 (0%) |
DALL-E-Middle School |
0 (0%) |
2 (8%) |
23 (92%) |
0 (0%) |
0 (0%) |
0 (0%) |
DALL-E-Elementary |
0 (0%) |
4 (16%) |
16 (64%) |
4 (16%) |
1 (4%) |
0 (0%) |
Stable Diffusion-University |
0 (0%) |
0 (0%) |
12 (48%) |
8 (32%) |
4 (16%) |
1 (4%) |
Stable Diffusion-High School |
4 (16%) |
17 (68%) |
4 (16%) |
0 (0%) |
0 (0%) |
0 (0%) |
Stable Diffusion-Middle School |
14 (56%) |
9 (36%) |
2 (8%) |
0 (0%) |
0 (0%) |
0 (0%) |
Stable Diffusion-Elementary |
20 (80%) |
4 (16%) |
1 (4%) |
0 (0%) |
0 (0%) |
0 (0%) |
5.3. Race and Representation
Across all three GAI platforms, white teachers dominated the dataset. This is consistent with what Weber & Mitchell (1995) identified in their analysis of mass-media teacher imagery: the default figure in popular representations of educators has long been white and middle-class, a pattern embedded in visual culture that it registers not as a choice but as a norm. The degree of that dominance, however, varied considerably across platforms and, in two cases, shifted in ways that merit attention.
DALL-E produced the most racially homogeneous results. White teachers accounted for 100% of university professors, high school, and middle school teachers generated by this platform. The only departure appeared at the elementary level, where 68% of images depicted white teachers and 32% depicted non-white teachers. This near-total whiteness at the upper educational levels is inseparable from the broader DALL-E finding: this platform repeatedly generated what appeared to be the same white male across multiple prompt categories. The racial homogeneity of DALL-E’s output is embodied in a single recurring figure who ages across educational levels but otherwise remains unchanged.
Stable Diffusion showed a similarly white-dominant pattern at the secondary and post-secondary levels, with 80% of university professors, 96% of high school teachers, and 88% of middle school teachers depicted as white. Meta AI showed the most diverse pattern across educational levels. At the university level, 88% of teachers were white and 12% non-white. This proportion shifted progressively at each lower schooling level: 76% white at high school, 60% white at middle school, and 40% white and 60% non-white at the elementary level.
Table 3. Race and representation in AI-generated images (n = 25 per row).
AI Model |
White |
Non-White |
Meta-University |
22 (88%) |
3 (12%) |
Meta-High School |
19 (76%) |
6 (24%) |
Meta-Middle School |
15 (60%) |
10 (40%) |
Meta-Elementary |
10 (40%) |
15 (60%) |
DALL-E-University |
25 (100%) |
0 (0%) |
DALL-E-High School |
25 (100%) |
0 (0%) |
DALL-E-Middle School |
25 (100%) |
0 (0%) |
DALL-E-Elementary |
17 (68%) |
8 (32%) |
Stable Diffusion-University |
20 (80%) |
5 (20%) |
Stable Diffusion-High School |
24 (96%) |
1 (4%) |
Stable Diffusion-Middle School |
22 (88%) |
3 (12%) |
Stable Diffusion-Elementary |
22 (88%) |
3 (12%) |
Note. Non-White aggregates all codes other than White.
The dominance of white teacher depictions across all three platforms and particularly the near-total whiteness of DALL-E’s output at the secondary and post-secondary levels, reflects and reproduces the pattern of whiteness as the default of the art teacher image (This is clearly visible in Table 3). Research on the Ontario teaching profession documents a significant racial gap between the population and its educators: while racialized groups account for 26.5% of Canada’s total population (Statistics Canada, 2021), only 9% of elementary school teachers and 10% of secondary school teachers in Ontario are racialized (Turner, 2015, as cited in Abawi, 2018). The elementary representation by both Meta AI and Dall-e is more difficult to account for. Why do both platforms produce an increase above demographic norms of non-white teachers at the elementary level, while defaulting to predominantly white depictions at every other level? This finding raises questions about what these platforms are being trained on and where their image datasets are sourced.
5.4. Strange Oddities
In terms of the AI technology, each AI generator had its own particular imagery and themes. There have been some oddities noted by the authors in relation to the AI technology and its portrayals of teachers. We observed that each AI we used has its limitations.
Figure 2. Depiction of similar male imagery.
DALL-E generated the same type of male representation over and over again, ranging from a person in middle age to a university professor as is evident in Figure 2. He appears to be aging depending on the level he is teaching: in the middle years, he is young, and at the university level, older. For instance, in the middle junior high school art teacher representations, approximately seventy-five percent of the imagery is of the same white male who has wavy to curly auburn hair, sporting a mustache and beard. He is in the 30 - 39 age range and wears informal clothing such as a buttoned-down or totally unbuttoned shirt, often sporting a tight-fitting painting smock with paint on it. Situated in the center of the image, he is, in many of the images, looking directly at the camera while his students are in the background of the art studio classroom, complete with media, tables, artworks, and easels. For the high school art teacher, again, the image of a white male who is bearded and mustached is reproduced many times: the difference between the middle and high school versions of this teacher is that this male appears somewhat older. The high school teacher has more wrinkles on his face and graying hair, which is not seen in the early-year version. At the university level, the art professor is the same man as in the early and middle years, but this time having more gray hair, wrinkles, and at times, sporting glasses. In approximately one-third of the images, these professors wear suits and fully buttoned-up shirts or turtlenecks, in contrast to the younger versions’ clothing; the rest are in dirty smocks and are placed centrally in the composition, again with studios and easels behind them.
![]()
Figure 3. Depiction of women in yellow dresses.
Meta AI has its own recurrences. In the early years, eight percent of the female teachers wore yellow dresses (Figure 3 illustrates this), as has been noted earlier in this paper. Also, Meta AI has produced a few abstracted images of teachers in which their bodies appear put together like a Cubist artwork. Additionally, Meta AI portrayed atypical teachers. In the middle years, for example, tattoos are depicted on the arms of one of the teachers (refer to Figure 4). We surmise that this is affected by Facebook and Instagram imagery. Also, at the secondary level, some teachers had distorted, atypical, and coloured ears that differed from the rest of their skin tone.
Figure 4. Female art teacher with a tattoo.
Stable Diffusion has its own problem: it appears to respond inadequately to the prompts provided. This AI portrayed students rather than teachers in all the groups except fine art university professors. For the prompt ‘an elementary art school teacher’, 80% of the images portrayed teachers 20 years or younger, of which at least 8 of the images represented children in early or middle years. Additionally, using the prompt for the middle years, we found that 56% were under 20, of whom 6 images portrayed youth in the early or middle years. By high school, this misrepresentation had been reduced. Using the prompt for secondary level teachers, it was found that 16% of the pictures of high school teachers were 20 years of age or younger, and only 3 of these depictions looked like students in the early or middle years. At the university level, this problem ceased: all images depicted were of older professors, at least 20 - 29 years old. A further anomaly was noted as well: art professors wearing lab coats in Stable Diffusion reflect on art studios drawing upon science imagery to depict the art process. Consequently, in summary, we have discussed the diverse problems that the differing AI technologies reveal in the generated imagery produced.
6. Conclusion
The researchers are aware that in using the Canadian population and Canadian teacher benchmarks as the main comparison point, the analysis emerges from a Canadian reference case rather than from a global teaching workforce. We made this distinction primarily due to the fact that we produced these images from Canadian IP addresses. What is not fully known from this research is the ways in which users’ IP locations could influence what is produced. We fully recognize that it would not be effective for researchers in future studies to use a VPN because the investigators’ browsing history and language settings could potentially influence the data. This would make an interesting future research study: to have investigators research this same subject using different IP addresses from around the world.
We began this study by asking whether GAI-generated images of visual art teachers perpetuate or depart from the teacher stereotypes established in mass media. The answer, taken across our dataset of 300 images, is largely yes, there is a continuation, but with a significant qualification. The negative portrayals of teachers extensively documented in film, television, and news media were absent from our findings. The depressed teacher, the burnt-out cynic, the drug addict, and the buffoon were nowhere to be found. What GAI produced instead was an overwhelmingly positive, idealized teacher: smiling, able-bodied, positioned confidently in a well-resourced classroom, unencumbered by the administrative burdens and financial pressures that define so much of real teaching life. This is the good teacher of Weber & Mitchell (1995), reproduced at scale. We attribute this, at least in part, to the sources GAI draws on. The reasonable explanation is that these systems are trained on imagery that skews toward stock photography and social media, both of which tend toward idealized depictions of teachers. Social media in particular functions as a highlight reel of everyday life, and stock photography is produced explicitly for promotional and institutional purposes. GAI, in this sense, does not generate new images of teachers; it concentrates and reproduces the images that already circulate most widely in digital visual culture.
The three platforms, however, behaved very differently from one another, and this finding, we postulate, has implications extend beyond our study. DALL-E produced near-total gender and racial homogeneity at the secondary and post-secondary levels, effectively generating the same white male figure across multiple prompt categories, aging him across educational levels while otherwise leaving him unchanged. Meta AI produced the most racially diverse outputs overall and was the only platform to invert the racial pattern at the elementary level. Stable Diffusion, in several categories, generated students rather than teachers, revealing a significant limitation in how that platform interprets prompts. These differences matter because a researcher or institution looking to advertise a program, generating a single image on a single platform, would have no basis for understanding how another platform might respond differently to the same prompt. Weber & Mitchell (1995) did not differentiate among media forms; our findings suggest that, with GAI, the platform itself is a variable that cannot be ignored. Some platforms produced more inclusive images of teachers than previous media portrayals have, while others produced less inclusive ones. This becomes evident when generating multiple images with the same prompt, but it would not be apparent if only a single image were generated.
Age representation across the three platforms broadly reflected reality, with caveats. Older teachers were depicted at the university level, high school teachers were generally shown in their thirties to fifties, and teachers at the elementary and middle school levels were depicted as younger. Weber & Mitchell (1995) found that teachers’ ages in traditional media ranged widely from the twenties to the sixties. Our findings were similar, though with notable distortions: Stable Diffusion and Meta AI produced images of teachers at the elementary and middle school levels who appeared young enough to be students themselves, which has no correspondence to the reality of teacher qualifications and hiring in Canada or the United States.
The whiteness and gender defaults we found have an impact beyond representation. Weber & Mitchell (1995) noted that “teachers were usually portrayed by everyone as a smiling white woman” (p. 46). Our findings confirm that this default persists in GAI imagery and, on some platforms, intensifies. Near-total whiteness dominated at the secondary and post-secondary levels, and university professors were depicted as 88 to 100% male on the DALL-E and Meta platforms, with 24% from Stable Diffusion. None of this matches the actual demographic composition of the Canadian art teaching profession today. When GAI imagery is used in institutional contexts, in recruitment materials, educational media, or communications, it shapes who sees themselves as belonging in teaching. These are not neutral images. Because GAI companies do not disclose their weighting processes for gender and racial representation, we cannot fully account for why these platforms produced the images they did. The plausible interpretation is that these images reflect assumptions already embedded in the visual culture on which these systems were trained. These technologies are also shifting rapidly. More advanced image-generation tools, such as NANOBanana, are already in use, and research indicates that people’s ability to distinguish AI-generated imagery from photography is becoming increasingly difficult (Roca, 2025). As GAI images become more and more realistic and difficult to detect, it becomes more important to understand how these images can concentrate stereotypes around professions, and in this case, teachers. Since each GAI model uses different images as sources, they do not all replicate the same stereotypes, and each may change over time as their sources evolve. It is important to consider how these GAI images fit into our cultural assumptions regarding types of people.
NOTES
1Teacher Self-identity and how it has been affected by mass media imagery of educators has been a focus of many researchers including Ahn & Leggo (2019), Beauchamp & Thomas (2009), Buchanan (2015), Mockler (2011), Nerlino (2023), and Weber & Mitchell (1995).
2Ewling et al. (2021) wryly observe that while teachers have been notably harmed by deleterious, nasty media portrayals, specifically in film and television, the criminology profession has been downrightly positively affected through affirming, encouraging portrayals (pp. 506-507).