GIFT-AI: Damn!.jpg—visual literacy through image-making with generative AI

This article addresses the growing need for AI literacy by introducing a classroom activity that combines critical theory with hands-on engagement using generative AI image tools. Students were guided through theoretical framing, image selection, AI experimentation, and group analysis. The activity emphasized how prompt design shapes visual outputs and explored the implications of generative systems through selected theoretical frameworks. It created opportunities for students to engage with the aesthetic and epistemological dimensions of AI-generated media. More broadly, he exercise high-lights how image-making can serve as both a critical and creative method for interrogating algorithmic systems across digital art, media education, visual culture, science and technology studies, and critical AI research. The following course framework outlines how the activity was implemented and contextualized within interdisciplinary learning environments.

Intended course

This exercise engages students in critical explorations of AI literacy by bringing AI-generated images into dialogue with key theoretical concepts and hands-on experimentation. It is particularly suited to undergraduate and graduate courses in the Arts, Critical Media Aesthetics, Digital Art and Humanities, Science and Technology Studies (STS), Visual and Cultural Studies, and Critical AI and Algorithm Studies. Appropriate for seminar-style classes of various sizes, this workshop can be delivered in person, in a hybrid format, or fully online, either as a 180-min session or as an extended module. The exercise is accessible to students with foundational knowledge of digital media platforms and theoretical frameworks. Sessions incorporate pre-assigned critical readings, a hands-on prompting activity, and reflective discussions on the broader implications of AI-generated content, using freely available images and generative technologies. This approach encourages students to situate computational media practices within wider social and discursive contexts.

Objectives

A single-class activity framed to enhance students’ AI literacy by developing their ability to: (1) Analyze key theoretical concepts and apply them to image-making tasks; (2) Assess how prompt variation influences visual output; and (3) reflect critically on how generative AI shapes perception, representation, and meaning.

Introduction and rationale

AI literacy, in the context of generative and multimodal systems, can be understood as a methodological inquiry into the aesthetic and epistemological conditions of machine-generated media (Crawford, 2021; Long and Magerko, 2020). While generative AI systems often produce content that resembles established media forms such as photography, their underlying mechanisms are rooted in statistical modeling rather than semantic comprehension or authorial intention (Amoore et al., 2024; Agüera y Arcas, 2022). In the case of visual outputs, generative AI is not photography, yet it remains inseparable from the history of the photographic image—borrowing its visual grammar while departing from its material and indexical foundations (Blaschke et al., 2025; Sekula, 1984).

This activity centers on photographic materials uploaded to a generative AI system as an entry point for investigating how statistical probability manifests visually. Students are introduced to the idea that AI models interpret images not as representations but as data, which are encoded as numerical features aligned with probabilistic distributions derived from training corpora. As Steyerl (2023) argues, this shift displaces the indexical authority of the photograph and replaces it with a logic of stochastic discrimination, in which visual outputs reflect statistical averages rather than singular references.

The pedagogical approach of the activity is also informed by Louise Amoore et al.’s (2024) concept of a “world model,” a framework that describes how machine learning systems generate representations by simulating likely patterns across massive data sets. Although Amoore et al. primarily address large language models (LLMs), similar architectures underpin many generative image systems. Multimodal AI, which integrates text, image, and sound, further illustrates the convergence of linguistic and visual computation (Gu and Ericson, 2025). Understanding these systems requires what Amoore et al. call a critical literacy of modeling itself: not just how outputs are produced, but how meaning is shaped by the infrastructures that generate them.

In this context, critical thinking entails a reflective inquiry into how generative AI systems mediate knowledge, shape perception, and influence how meaning is produced and interpreted. Selwyn (2024) emphasizes that AI in education should be approached critically, with attention to the social, institutional, and epistemological values embedded in its use. Within the activity, students are prompted to analyze how generative models transform photographic materials through probabilistic logic and to question how these transformations reflect broader shifts in authorship, meaning-making, and visual authority. This orientation positions critical thinking as central to AI literacy, enabling students to interrogate the conditions under which machine-generated media gain credibility, coherence, and influence within educational and cultural domains.

Taken together, these perspectives support a recursive pedagogy in which students move between theoretical concepts and hands-on experimentation. By crafting and analyzing AI-generated imagery, they examine how prompt-based user input is restructured through probabilistic modeling, and how such systems contribute to the construction of a statistical “world model,” one that shapes what is visible, knowable, and imaginable.

Materials needed

Each student must have access to a networked computer and either GPT (or a comparable large language model) or a generative AI platform capable of both producing images from text prompts and modifying uploaded images. Free versions of these tools are sufficient for the purposes of this exercise. A projector is also required to facilitate whole-class discussions and collective analysis.

Step by step implementation

Step 1: introduction and theoretical framing (30 min)

The activity begins with a brief introduction and theoretical framing, which provides the conceptual and practical foundation for the following five steps: (1) Introduction to the exercise and discussion of theoretical concepts; (2) image collection from open-access platforms; (3) AI image generation and experimentation; (4) group analysis; and (5) whole-class presentation and debriefing.

This activity is designed for the midpoint of the semester, after students have read and discussed key texts such as Steyerl (2023), Amoore et al. (2024), and Agüera y Arcas (2022). These readings should be assigned during the previous week, and students are expected to arrive prepared to apply these theoretical frameworks throughout the image generation and analysis stages.

The session begins with an introduction and theoretical framing (30 min), during which students revisit key arguments from the assigned critical texts. They are introduced to two guiding questions that serve as a conceptual foundation for the exercise—questions intended to remain present in the background of their engagement, shaping their critical orientation as they progress through the task.

1. What aesthetic patterns, visual tendencies, or biases—such as platform-specific norms—emerge in AI-generated images, and what might they reveal about how the model operates?

2. How does the concept of a “world model” help us interpret these outputs and understand the assumptions behind generative AI systems?

Step 2: image collection (30 min, including 10 min for written reflection)

Having established a shared conceptual framework, students now move into the practical phase of the exercise by gathering visual material for manipulation. This step introduces them to the curatorial dimension of prompt-based image generation.

Students begin by selecting images for use in the AI image generation and experimentation phase. The chosen images must be downloaded onto a computer to ensure that the AI model modifies the original image directly, rather than generating a new interpretation based on a prompt.

Selection criteria:

1. Select an image that invites expansion or reinterpretation—one that requires the AI to generate new details or extend visual elements.

2. Do not use AI-generated images. The selected image should be an existing, non-AI-generated visual to ensure that the model expands upon a photographic reference.

Before proceeding, students document their image selections and write a brief reflection outlining their expectations for how the AI model might transform the image. The instructor ensures that 5 min are allocated for completing this written reflection.

Step 3: AI image generation and experimentation (30 min, including 10 min for written reflection)

With source images selected, students now explore how prompting practices influence the visual logic of generative AI, engaging in iterative experimentation with their chosen images.

Students upload their selected images into a generative AI model and experiment with various prompt formulations to observe how the model processes and modifies visual inputs. This phase emphasizes the relationship between prompt phrasing and AI-generated output, highlighting how the system interprets instructions related to image transformation.

Prompting guidelines:

(For examples, see Figures 1–6)

1. Upload the image and prompt the AI model to extend it to the right, left, top, or bottom.

2. Prompt the AI model to repeat a specific visual element from the image.

3. Prompt AI model to further develop or refine a previous prompt, observing how it builds upon earlier transformations.

Figure 1

Figure 1. Damn! Photo by Kevin Dooley, Chandler, AZ, USA. Uploaded to Wikimedia Commons on 20 December 2020. Licensed under CC BY 2.0, via Wikimedia Commons.

Figure 2

Figure 2. AI-generated extension of Damn! by Kevin Dooley (Figure 1), created by prompting a generative model to “use the uploaded image and extend it by expanding it to the right.” Source image licensed under CC BY 2.0, via Wikimedia Commons.

Figure 3

Figure 3. AI-generated extension of Damn! by Kevin Dooley (Figure 1), created by prompting a generative model to “Use the uploaded image, expand it further to the right, and make it resemble the original film with random light leaks.” Source image licensed under CC BY 2.0, via Wikimedia Commons.

Figure 4

Figure 4. Habitat 67. Photo by Maela Ohana, uploaded to Wikimedia Commons on 28 February 2022. Licensed under CC BY-SA 4.0, via Wikimedia Commons.

Figure 5

Figure 5. AI-generated extension of Habitat 67 by Maela Ohana (Figure 4), created by prompting a generative model to “Use the uploaded image and extend it by adding more units to the right.” Original photo by Maela Ohana, licensed under CC BY-SA 4.0, via Wikimedia Commons.

Figure 6

Figure 6. AI-generated extension of Habitat 67 by Maela Ohana (Figure 4). Created by prompting a generative model: “Use the uploaded image, expand it further to the right, and make it resemble the original film with random light leaks.” Original image by Maela Ohana, licensed under CC BY-SA 4.0, via Wikimedia Commons.

Students must document each prompt used, save all generated outputs, and organize them in a designated folder. This record will be essential for the final analysis and discussion phase, where students reflect on how the AI system processes visual patterns and responds to variations in user input. The instructor ensures that 10 min are reserved for written reflection, addressing the following questions:

1. How does prompt phrasing affect the generated image?

2. How does the AI model interpret ambiguous visual patterns, such as light leaks or blurry edges?

Step 4: group analysis and presentation preparation (30 min)

After individual experimentation, students transition to group work, synthesizing their insights and preparing to present their findings.

In this phase, students work in groups, with the class divided into a maximum of four groups to present during the whole-class discussion. Each group uses the written reflections from Steps 2 and 3 as a foundation for their work.

Guiding questions:

1. How did the AI’s modifications compare to your initial expectations?

2. What generative models were used, and how did their outputs differ?

3. How did variations in prompt phrasing influence the resulting images?

Each group will review their outputs, identify key insights, and select the most relevant examples. These findings will be compiled into a shared group document in preparation for a projector-based, whole-class presentation in the following session.

Step 5: whole-class presentation and debriefing (60 min)

In the final phase, students collectively reflect on their findings, drawing explicit connections between AI outputs and the theoretical models introduced earlier.

Each group presents their selected outputs, key observations, and reflections to the class, drawing on their own group document. The instructor facilitates a debriefing discussion that invites students to connect their findings with the broader theoretical and conceptual frameworks introduced earlier in the session.

Suggested debriefing questions:

1. What patterns or tendencies emerged across different groups’ outputs, and how do these reflect the probabilistic logic of generative AI systems?

2. In what ways did prompt phrasing shape the outputs, and what does this suggest about authorship and control in AI-assisted image production?

3. How do critical concepts introduced in previous class sessions, such as “mean images,” (Steyerl) or “world models” (Amoore et al.) help explain the visual strategies or biases observed in your outputs?

4. How did the AI handle ambiguity, irregularity, or open-ended visual elements, and what does this reveal about the model’s assumptions or limitations?

5. What could be the potential broader social or cultural implications of image generation practices like these?

Appraisal

The activity was implemented in an MA program in Digital Communication within the social sciences, as part of a course titled Critical Topics in Digital Discourse, which focuses on close reading and discussion of contemporary and emerging issues in digital media. In an anonymous survey administered at the end of the semester, students reported that the exercise supported their understanding of how generative AI systems function, particularly in relation to the aesthetic, technical, and political dimensions of machine-generated imagery.

Several respondents highlighted the value of combining hands-on experimentation with theoretical concepts, noting that working directly with AI tools helped clarify and contextualize the readings. The group-based discussion format was also seen as productive, encouraging meaningful discussions about authorship, bias, and the ways AI shapes visual representation. One student reflected, “I expected the AI to fill in the photo naturally, but it created something uncanny—it misunderstood what a building looks like.” Another noted, “It’s like the AI draws from what it’s seen too many times, not from the image itself.” These reflections suggest that students actively engaged with and applied the critical material through practice.

While originally developed for graduate-level instruction, the activity can be adapted for undergraduate settings or shorter class formats lasting between 60 and 90 min by streamlining the workflow. For example, instructors may pre-select image inputs and limit AI experimentation to a single prompt modification per student or group. The theoretical framing can be condensed into a brief instructor-led overview, and group discussion can focus on a smaller selection of the debriefing questions. These adjustments preserve the core pedagogical aim, which is to engage students in critical inquiry about generative AI, while accommodating limited time and a broader range of student experience levels.

Variations

To build on the exercise, students can further explore how working with a variety of generative models deepens their understanding of how these systems interpret inputs, produce outputs, and shape visual results. They may engage in iterative prompting by re-uploading AI-modified images into the same or different models to examine how layers of transformation accumulate and affect meaning.

Additionally, comparing results across open-source and proprietary platforms can prompt discussions about how factors such as access, transparency, and training data influence not only the visual outputs but also the ways those outputs can be interpreted. While generative AI exercises are becoming more common in media and design curricula, this activity is distinct in its integration of photographic source material with prompt-based manipulation. It encourages students to investigate not only how images are generated, but how such outputs reflect deeper epistemological questions about how meaning is produced, structured, and interpreted by computational systems. Framing the activity through concepts like “world models” foregrounds the stakes of AI literacy as a matter of understanding how machines participate in shaping what becomes intelligible, credible, or imaginable within mediated environments.

These variations support the course’s goals by encouraging students to critically assess the creative possibilities and limitations of generative AI across diverse contexts.

Limitations

While the activity combines theoretical inquiry with hands-on experimentation in productive ways, several limitations emerged during implementation. The use of freely available generative AI tools, although pedagogically valuable and accessible, introduced variation in output quality, model responsiveness, and interface functionality. Some platforms restricted prompt complexity or image upload options, which occasionally affected the consistency of student outcomes across groups.

Data availability statement

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

Author contributions

KK: Writing – original draft.

Funding

The author(s) declare that no financial support was received for the research and/or publication of this article.

Acknowledgments

This learning activity was developed following a guest presentation by artist Klara Källström, reflecting on her collaborative practice with artist Thobias Fäldt. Invited to the course by Professor Bernard Dionysius Geoghegan, their work in visual arts and photography served as a key inspiration for the exercise. The artists gratefully acknowledge the opportunity to contribute, as well as the collaborative dialogue that informed the conceptual development of this activity. They would also like to thank Steven Tester and his colleagues at Scholarminds.ai for their support and for providing resources that helped shape the ideas explored here. Finally, they thank those who offered insightful comments on the work along the way.

Conflict of interest

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author declares that Gen AI was used in the creation of this manuscript. Generative AI was used as both a subject of critical inquiry and a creative tool in the classroom activity. Students engaged directly with image-generating models, using prompt-based experimentation to explore how these systems transform visual inputs and reflect broader aesthetic, technical, and epistemological dynamics in AI-generated media.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Agüera y Arcas, B. (2022). Do large language models understand us? Daedalus 151, 183–197. doi: 10.1162/daed_a_01909

Crossref Full Text | Google Scholar

Amoore, L., Campolo, A., Jacobsen, B., and Rella, L. (2024). A world model: On the political logics of generative AI. Polit. Geogr. 113:103134. doi: 10.1016/j.polgeo.2024.103134

Crossref Full Text | Google Scholar

Blaschke, E., Bonhomme, M., Joschke, C., and Somaini, A. (2025). Introduction: Photographies et algorithmes. Transbordeur 9, 6–19. doi: 10.4000/13dwn

PubMed Abstract | Crossref Full Text | Google Scholar

Crawford, K. (2021). The atlas of AI: Power, politics, and the planetary costs of artificial intelligence. New Haven, CT: Yale University Press.

Google Scholar

Gu, X., and Ericson, K. (2025). AI literacy in K-12 and higher education in the wake of generative AI: An integrative review. arXiv. Available online at: https://arxiv.org/abs/2503.00079

PubMed Abstract | Google Scholar

Long, D., and Magerko, B. (2020). What is AI literacy? Competencies and design considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (pp. 1–16). Association for Computing Machinery. doi: 10.1145/3313831.3376727

Crossref Full Text | Google Scholar

Sekula, A. (1984). Photography against the grain: Essays and photo works, 1973–1983. Halifax, NS: Press of the Nova Scotia College of Art and Design.

Google Scholar

Selwyn, N. (2024). On the limits of artificial intelligence (AI) in education. Nord. Tidsskr. Pedagog. Krit 10, 3–14. doi: 10.23865/ntpk.v10.6062

Crossref Full Text | Google Scholar

Steyerl, H. (2023). Mean images. New Left Rev 140/141, 82–97. doi: 10.64590/uhm

PubMed Abstract | Crossref Full Text | Google Scholar