Beyond the Prompt – AI and the African Diasporic Image: Memory, Archives and Authorship in Contemporary African Art

Artificial intelligence is often presented as a technology of speed. A few words are entered; an image appears. This apparent immediacy has encouraged the belief that AI reduces artistic creation to a prompt, the written instruction given to a generative system, and the artist to the person who writes it.

The practices considered here suggest something more complex. Diallo develops a sustained conversational protocol before and around image-making.¹ Okaro locates cultural specificity in intention, lived experience and human encounter rather than in a medium alone.² Tsegba approaches digital collage as an archival practice in which images are gathered, read and recomposed.³ Dounia constructs project-specific datasets, curated collections of data used to train or study models, and examines how generative systems reorganise memory.^{4, 5} Atairu tests the cultural assumptions embedded in text-to-image systems, which generate images from written descriptions.^{6, 7} Attafuah uses generative AI selectively within a wider digital practice, while Àsìkò places AI alongside research, photography and editing within a practice informed by Yoruba cosmology, a broad term for Yoruba understandings of relations between visible, spiritual and ancestral worlds, and diasporic memory.^{8, 9, 10}

These artists do not share one method, one position or one visual language, and the category "African AI art" cannot describe their differences. The stronger connection lies elsewhere: in their attention to what images inherit, what archives make visible or difficult to find, how bodies are represented, and where artistic agency remains possible inside technical systems the artist does not fully control. In that sense, authorship lies beyond the prompt. It is located in research, selection, refusal, recontextualisation, material practice and the construction of conditions in which an image can mean more than the system that helped produce it. For a broader orientation in the field, see OOA Gallery's guide to contemporary African art.

A Longer History of Image Transformation

The practices examined here did not begin with generative AI. They extend longer histories in which artists have constructed, layered, animated and recontextualised images through photography, collage, digital manipulation, algorithmic processes and immersive technologies. Artificial intelligence represents a new moment within this evolution rather than a complete break with what preceded it.

This continuity takes different forms across the work of the seven artists discussed here. Diallo's use of AI grows out of analogue and digital photography, collage and a sustained investigation of the relationship between subject and observer.¹ Okaro's reflections place AI beside photography rather than after it: generative systems can visualise memory, longing and alternate realities, while photography retains the trace of an encounter between artist, subject and a real moment in time.²

Dounia's generative work develops from her engagement with images, code, datasets and archives, while Atairu uses existing generative systems to investigate gaps and distortions in Black historical and visual records.^{4, 7} From a curatorial perspective, these practices can be situated beside longer critiques of archival power without implying that a generated image itself becomes historical evidence.¹¹

Tsegba makes this continuity explicit. In her written responses, she describes collage as a continuous movement of gathering, reading, disassembling and recomposing existing images. She works from the digital archive while also contributing to it, looking to past and present in order to inform the future, and invokes Sankofa, an Akan concept centred on returning to the past in order to carry valuable knowledge forward, as a resonant framework for that movement.³ Load Gallery identifies Tsegba’s 2022 work Dambe as a still collage made using digital collage. No AI process is documented for the work, so it appears here as a digital-collage and archival counterpoint within the longer history of image recomposition rather than as evidence of AI generation.¹²

Alexis Chivir-ter Tsegba, Dambe, 2022. Still collage. Digital collage. Courtesy of Load Gallery and Alexis Chivir-ter Tsegba.

In Dambe, two dark, near-silhouetted male figures face one another in a posture close to combat, arms raised, against a composite landscape of rolling green hills crossed by a winding road. Two large white egrets, wings fully extended, occupy the centre of the frame between the figures, and a textured golden disc sits behind them like a sun. The bodies retain the grain of black-and-white photographic source material, while the birds, the gilded disc and the aerial landscape belong to different tonal and spatial registers. The elements do not resolve into a single continuous scene. According to Tsegba's own account, the work developed from photographs she took during a 2021 encounter with the Nigerian martial art of Dambe, including a portrait of two men at a local boxing contest; Dambe is a form of boxing historically associated with Hausa communities in West Africa.¹³ The work does not document that encounter so much as reorient it: montage becomes legible as method, and the photographed meeting is recomposed into a new visual space in which, in the artist's words, existing images can be seen "through new eyes" and made available to be "reencountered".³

Attafuah's documented use of generative AI belongs to a broader practice encompassing digital image-making, sound and immersive technologies, and speculative environments populated by Afrofuturistic avatars.^{8, 9} Àsìkò approaches technological experimentation through a different framework, combining research, photography, collage, film and AI while engaging Yoruba cosmology and diasporic experience.^{14, 10} His written responses emphasise continuity without equivalence: spiritual and computational systems may prompt comparative questions, but he explicitly describes them as fundamentally different systems.

The significance of AI in these works therefore lies not in the sudden appearance of an entirely new artistic language, but in the way it transforms existing methods of assembling images, revisiting archives and constructing possible worlds.

Delphine Diallo, Into the Dream, 2026. Film still. Courtesy of the artist and Load Gallery.

From Tool to Relationship

For Delphine Diallo, reducing artificial intelligence to a device for producing images would flatten the meaning of her practice.¹

"AI is not simply a tool for creating images. It is a reflection on our relationship with existence, the universe and matter."¹

In the interview, Diallo traced her engagement with AI back through nearly two decades of photographic practice. She began with analogue photography, black-and-white film and the darkroom, developing an understanding of image-making as a relationship between the external subject and the interior world of the observer. After the publication of her book Divine in 2022, she felt that her relationship with the camera had reached the end of a cycle. AI did not cause that transformation; it arrived when the urgency she had previously experienced through photography was already changing.¹

In her conversation with OOA Gallery, Diallo explained that she initially worked with ChatGPT to establish a relationship and a framework before creating images. Through repeated exchanges, references, corrections and instructions, she developed what she calls a protocol: a particular space within the system where suggestion and unpredictability could coexist with artistic intention.¹

"At first, I worked with ChatGPT to create a relationship with the machine before even creating my images."¹

Diallo describes this exchange as a "real dialogue", but the term is an artistic account of her working relationship, not a technical claim that the system has become an independent author. In her description, she establishes references and a framework, judges what the system proposes, rejects results and leaves room for surprise.¹

Linda Dounia Rebeiz adopts a more reserved vocabulary. In a published interview with Le Random, she describes a GAN trained with her own material as an amplifier or a digital "griot", invoking a West African figure associated with oral history, storytelling and the transmission of collective memory: a system capable of reorganising information she has supplied and returning it in unfamiliar forms. Prompt-based software, by contrast, can feel closer to a query-and-response service. For Dounia, a more conversational process emerges when different media interact, control decreases, and the artist must repeatedly question, adjust and reassess unexpected translations between them. She is careful not to describe this straightforwardly as collaboration.⁵

This difference between Diallo and Dounia is productive. It shows that artists working with related technologies do not necessarily attribute the same role or agency to them. AI may be experienced as an interlocutor, an amplifier, an instrument, an archive, an investigative system or simply one stage within a larger process.

Archives, Classification and the Limits of Representation

The first major question concerns what precedes the generated image: the archives, labels, descriptions and visual conventions from which a model learns. The practices of Dounia, Diallo, Atairu and Tsegba show that representation cannot be separated from the systems through which images are recorded, named and retrieved.

Archives, memory and power

The relationship between image and memory is central here, although the artists approach it differently.

In her essay AI's Unseen Borders, Dounia calls AI models "memory machines". The memory is not their own: it is assembled from the ways people have recorded, classified and interpreted the world. For Dounia, images and data are not neutral reflections of reality. Both are shaped by context, intention and power. Decisions about what is recorded, how it is named, where it is stored and how it is retrieved influence what becomes visible, classifiable and potentially available for model training.⁴

She therefore treats the creation of archives from multiple perspectives as a resistance to disappearance. Her practice involves gathering photographs, scanning documents, digitising material that is not yet available online, identifying and tagging records, and assembling more locally sourced datasets where the available record is incomplete or distorted.⁴

Tsegba introduces a complementary problem: disappearance through excess. She rejects the assumption that putting images online guarantees their survival. "Digital abundance can be another form of disappearance."³ When millions of images circulate, she argues, important stories can become harder to find. Her collage practice therefore asks not only what an archive contains, but how images that become difficult to find within digital abundance might be brought back into view and recontextualised. This complicates any simple equation between more images and more memory.

Àsìkò adds a further limit to the idea of the archive. He argues that not all cultural knowledge exists in books or archives: much of it is embodied and carried through ritual, gesture, performance, objects, cloth, song and oral tradition, enduring through lived experience.¹⁰ Read in relation to Dounia and Tsegba, this introduces a different kind of absence. Even a richer digital archive cannot contain every form of cultural memory, because some knowledge depends on participation, embodiment and transmission between people.

Archival scholarship has long challenged the idea of the archive as a passive storehouse. Joan M. Schwartz and Terry Cook argue that decisions made during record creation, appraisal, description, preservation and access participate in shaping social memory. Their argument concerns archives rather than machine-learning datasets; the connection to AI drawn here is an analytical extension. This does not make every archive arbitrary; it means that what survives, how it is classified and how it becomes accessible are themselves historical acts.¹¹

Diallo uses a different but related word: rememorisation.

"When we decolonise, we rememorise what has been forgotten."¹

Diallo describes "recalibration" as the repeated recognition and correction of outputs she experiences as falling back into colonising patterns. She insists that this cannot be reduced to a technical procedure: in her account, the person working with the system must question their own inherited gaze and historical formation.¹

Her stated intellectual references include Frantz Fanon, Marcus Garvey, Amos N. Wilson, Octavia Butler, Toni Morrison and Malidoma Patrice Somé. These names should not be described as training data in a technical sense. They form part of the intellectual framework through which Diallo questions histories, evaluates responses and recognises narratives she wishes to challenge.¹

From OOA Gallery's curatorial perspective, a speculative image cannot replace an archive and should not be mistaken for historical evidence. But it can expose what an archive does not contain. It can make an absence visible.

When models mis-see

The limitations of generative AI become particularly clear when artists ask systems to represent places or cultural details that are insufficiently documented or repeatedly stereotyped.

Research on text-to-image systems indicates that such examples are not isolated anomalies. A study evaluating DALL-E 2, Stable Diffusion 1.4 and Stable Diffusion 2 found that all three systems consistently under-represented marginalised identities, to differing degrees, in generated depictions of professions. The finding concerns those evaluated systems rather than every model, but it supports treating representation as a structural question rather than a series of accidental errors.¹⁸

According to Dounia's published account, when she queried diffusion models, a common class of generative image systems, about Dakar, the results were dominated by dirty streets and low, dilapidated buildings that bore little resemblance to the city she knew. She observes that much of the publicly available archival material about Senegal was produced by people who neither came from nor lived there. In Dounia's interpretation, these outputs reflect not Dakar itself, but the dominance of a narrow image of Africa within the available visual record.⁴

Atairu’s project documentation and a published interview describe some of her most concrete investigations of this problem. Her Blonde Braids Study tested whether Midjourney (V4) could represent Black women with the requested blue-black or plum-black complexions wearing blonde braided extensions. Many outputs instead depicted caramel-complexioned subjects, while the hairstyles frequently appeared wavy or unnaturally silky.⁶

Minne Atairu, Blonde Braids Study IV, 2023. Text-to-Image. Courtesy of the artist.

The selected image presents two closely framed portraits against a dark ground, each in denim, their faces almost touching. The left figure has a markedly lighter, caramel complexion and long pale-blonde braids; the right figure is much darker-skinned and wears dark, blue-black braids. The juxtaposition makes the instability Atairu describes immediately visible: hair colour, hairstyle and skin tone are not held together by the system in the culturally specific combination she requested, and the paler blonde attaches to the lighter-skinned figure.

Atairu interprets these results as evidence that the model had not adequately learned the cultural relationship between Black hair, braided extensions and the word "blonde". Instead, it extrapolated from more heavily represented associations between blondeness, white women and natural hair.⁷

Atairu encountered another revealing failure while working with Imagen 3. When she entered "BOX BRAIDS" in uppercase, the system sometimes interpreted the word box literally, producing grid-like or tessellated structures. The result exposed the distance between a model's vocabulary and the lived cultural knowledge contained in the name of a hairstyle.⁷

A dataset may contain images of Black people without containing the distinctions, language and cultural context necessary to interpret them. Representation is therefore not solved merely by increasing the number of images. It also depends on classification, the relationship between words and images, and the knowledge of those designing the system.

In 2023, Dounia curated In/Visible for Feral File, a digital art exhibition that launched on 12 June 2023 and included Adaeze Okaro, Minne Atairu and Serwah Attafuah among its participating artists. The project framed AI's construction of Black realities as a field of visibility, omission and distortion rather than as a neutral technical process. That framing should be attributed to Dounia and Feral File rather than presented as a universal technical conclusion.¹⁹

Forms of intervention: prompting, testing and training

These failures do not lead the artists to the same form of intervention. Some work through repeated prompting and correction, others construct datasets or train smaller systems, and others use the model's errors as evidence.

In Diallo's account, "recalibration" takes place primarily through a conversational and artistic protocol. She introduces references, corrects responses, changes her language and rejects results that return to stereotypical patterns. The method is cumulative and depends on repeated exchanges, a particular vocabulary and years of artistic and intellectual research. Her interview with OOA Gallery does not, however, establish that she changes a foundational model's parameters or retrains an underlying large language model.¹

Dounia's practice includes a more directly technical engagement with model training. She began experimenting with GANs and has described training models with material she assembled herself. Working with project-specific datasets gives her greater control over the source material entering a particular experiment, although it does not amount to control over the wider infrastructures of generative AI.^{4, 5}

Atairu often uses existing generative systems as investigative instruments. Their errors become evidence of the cultural and structural limits within them. She has also described a growing interest in designing systems suited to communities that are too often treated only as end users.⁷

The expression "working with AI" can therefore describe very different activities: prompting a commercial model, constructing a long-term conversational framework, curating a dataset, fine-tuning a system, training a GAN, combining AI outputs with photography and collage, using AI during preliminary research or building a moving-image environment. Treating these activities as equivalent obscures the artistic decisions that give them meaning.

Bodies, Characters and Speculative Worlds

The second major question concerns what kind of figure enters the image. Across these practices, representation is not limited to visible diversity. It also involves interiority, cultural memory, speculative identity and the relationship between a work and the wider practice from which it emerges.

Giving characters an interior life

One of the most distinctive aspects of Diallo's process concerns how she writes characters. Many prompts begin with appearance: skin, clothes, facial features, pose, lighting or setting. Diallo begins with personality.

"When I describe a scene and a character, I describe the character's interior, their personality and their traits, more than their physical appearance."¹

She introduces memories, emotional qualities, modes of behaviour and ways of inhabiting space. The system proposes a physical form, but this form is expected to carry an interior life.¹

Here, the distinction matters because visible diversity alone cannot guarantee a culturally specific or psychologically convincing image. Diallo's method shifts attention from the catalogue of outward features towards the construction of interiority, while still leaving the generated physical form open to acceptance or refusal.¹

Presence, intention and cultural specificity

Okaro approaches these questions from the relationship between photography and AI. She argues that complexity belongs neither to the camera nor to the algorithm, but to the artist's attention to people as individuals rather than symbols. Cultural specificity, in her account, emerges through lived experience and through details such as gesture, colour, language, fashion and memory.²

"Technology can generate images, but it can't replace the responsibility artists have to understand the people and cultures they're representing."²

Okaro does not define authenticity by the medium used. She locates it in intention: why an image is made, which conversation it enters and what emotional or cultural truth it seeks to communicate. At the same time, she preserves a distinction between generated and photographic images.²

"Photography carries the weight of presence."²

For Okaro, even a carefully staged photograph retains traces of an encounter between the artist, the subject and a real moment in time. Read in relation to this Insight, her response complicates any simple opposition between technological experimentation and human presence: AI may visualise memory, imagination and alternate realities, but it does not remove the artist's responsibility towards the people and cultures represented.²

Adaeze Okaro, Drifting, 2019. Photography. Edition of 4. Courtesy of Load Gallery and Adaeze Okaro.

In Drifting, a single figure stands within a field of warm amber and deep shadow, draped in and surrounded by translucent tulle that extends beyond the body and softens the boundary between figure and space. The lighting is close and enveloping; the fabric catches an orange glow while the ground dissolves into darkness. For this Insight, the image offers a counterpoint to generative image-making: its significance lies not in any assumed use of AI, but in Okaro's insistence that photography retains the trace of an encounter between artist, subject and moment.

Speculative figures across media

Serwah Attafuah offers another approach to the construction of digital figures. Her broader practice encompasses digital image-making, sound and immersive technologies, giving form to speculative environments and Afrofuturistic avatars informed by mythology, science fiction and personal experience.⁹

In an interview published by Adobe, Attafuah described using generative AI to accelerate brainstorming and to create reference images for visions that did not yet exist. In this workflow, AI is not necessarily the final medium; it can help materialise an initial idea before the artist develops it through her wider digital practice. Because the interview was published by a company that develops and markets generative AI products, it is used here only as a source for Attafuah's account of her own working process, not as independent evidence of the wider benefits of AI.⁸

Generative AI should not, however, be assumed to run through all of Attafuah's digital work. ACMI states explicitly that no part of her 2025 moving-image installation The Darkness Between the Stars was made using generative AI. The fifteen-minute work combines video, projection, an original soundscape and e-waste sculptures made from discarded electronic material. Its "how it was made" documentation lists visual effects (VFX), animation, 3D modelling, character design, the real-time 3D platform Unreal Engine, rigging (creating a digital skeleton that allows a 3D character to move) and simulation; the physical e-waste frames were produced in collaboration with the artist's father, the metal sculptor Stephen Attafuah.⁹ Her practice therefore offers a useful reminder that digital art, immersive art and AI-generated art are not interchangeable categories.

Àsìkò similarly presents AI as one element within a wider practice. He explains that his process begins before image generation, with research into Yoruba cosmology, conversations, sketching, photography, editing and sustained engagement with cultural history. "AI is simply one tool within a much broader creative practice."¹⁰

His account of diasporic memory is equally specific. Parts of his heritage, he writes, arrive through fragments: family stories, photographs, objects, language, research and imagination. Digital image-making can bring those fragments into relation, but he cautions that digital images cannot replace lived cultural experience. He is less interested in reconstructing an exact past than in creating a space where memory, research and imagination can meet.¹⁰

Àsìkò also draws a deliberately qualified parallel between Yoruba cosmology and computational systems. He describes both as involving structures that are not immediately visible, while stressing that they are "fundamentally different systems". The significance of the comparison lies in the questions it opens about transmission, participation and what becomes visible, not in treating spiritual and computational knowledge as equivalents.¹⁰ Published material on his Orishas similarly documents images made with the assistance of AI in relation to ancestral planes and other dimensions, in his own words.¹⁴

Àsìkò, The Orisha Osanyin, 2024. Photography, AI. Framed lenticular print, 80 × 120 cm. Courtesy of Àsìkò.
Costume: Temitope Uduak Betiku. Costume Assistants: Styleforte, Ruthel Styling. Muse: Oghenekewen Elson. Line Producer: Bukie Garuba. Makeup: Onome Ezekiel. Makeup Assistant: Ifeoluwa Aduloju. Hair: Kehinde Are. Hair Assistants: Nike Folorunsho, Kemi Moyegun. Assistants: Thankgod Ezedibia, Olaoluwa Akinyoola. BTS: Tayo Opatayo, Chinedum Isichie.

The composition centres a frontal figure whose striped, earth-toned agbada, a flowing West African robe, and body are interwoven with dense green foliage, leaves and vines climbing from the ground to a tree-like crown of branches rising above the head. A large circular architectural opening frames the figure, and a misted forest of towering, moss-covered forms recedes on either side; a row of small stylised birds runs along the lower edge. The work is presented as a framed lenticular print; lenticular printing uses a lens-based surface to create shifts in depth or image as the viewing angle changes. In his own statement, Àsìkò explains that the work began with photography, a real body, real fabric and real adornment, while AI was used to construct the surrounding spiritual architecture, living forest and ancestral atmosphere. He describes his interest as lying in the space between "what the camera sees and what culture remembers".²⁰ Load Gallery notes that the work was originally created for the Yoruba Museum in Nigeria.¹² The composition holds photographic presence and constructed environment in the same image without treating one as a substitute for the other.

Read together, these practices suggest that the future is not merely a new surface. It is a question of who inhabits the image, which memories accompany them and what forms of consciousness they are allowed to possess.

Beyond the label "AI artist"

Atairu puts the problem succinctly: "Generative AI is just the tool I choose to express myself as an artist."⁷ Her wider practice makes the distinction concrete: alongside generative systems, she has engaged histories and subjects including the Benin Bronzes, the Black Madonna, Black hair, photography and gaps in Black historical archives.⁷ AI is therefore one medium within a broader investigation rather than the subject that defines the practice.

The same caution applies differently across this group. Attafuah's generative experiments sit within a much wider digital practice; Àsìkò describes AI as one tool among research, photography and editing; Tsegba's selected Dambe is documented as digital collage rather than AI-generated work; and Okaro moves between photography and AI without locating authenticity in either medium alone.^{8, 9, 2, 3, 10, 12}

Okaro extends this caution to the category of African and diasporic art itself. She notes that artists are still expected to fit familiar narratives or aesthetics despite practices that move across different media, methods and positions.² Her point reinforces the need to treat "African AI art" as a field of distinct practices rather than as a single visual language or cultural position.

Diallo's experience adds another reason for resisting the label as a shortcut. She describes three years of sustained research, experimentation and financial investment, while also recounting online attacks that equated her use of AI with the loss of artistic legitimacy.¹ The rendering speed of a system says little about the time required to build a practice, a body of references or criteria for accepting and rejecting results.

For galleries and collectors, the relevant question is therefore not whether a work can be reduced to the label "AI art", but what the artist actually did, what is fixed or variable in the work, how authorship is documented and how the piece belongs to a longer practice. The medium remains important, but it should clarify the work rather than substitute for interpretation.

Material Conditions, Ownership and Autonomy

The third major question concerns the material and institutional conditions of AI. The artists considered here do not treat it as a neutral system existing outside political and economic structures.

In AI's Unseen Borders, Dounia connects visual representation to the broader AI value chain, noting that the African continent is prominent within it partly because raw materials used for technological hardware are sourced there. These observations form part of her own critical account of AI.⁴

The infrastructure is also human. Research published in the Weizenbaum Journal of the Digital Society, drawing on ILO surveys conducted in India and Kenya, argues that AI depends throughout its lifecycle on human labour, distinguishing the algorithmic workers who code and fine-tune models from the often less visible data workers who label, clean and expand datasets. The study places the recognition of this labour within the ethics of AI itself, without measuring any particular commercial model.²¹

Material costs also continue after a model has been trained. A 2024 study presented at the ACM Conference on Fairness, Accountability, and Transparency found that multi-purpose generative AI systems were orders of magnitude more expensive per inference, meaning each use of a trained model to produce an output, than task-specific systems across a range of tasks, even after controlling for model size. The study does not provide one universal footprint for an AI-generated artwork, and it concerns deployment, when trained models are actually used, rather than training, but it is sufficient to reject the idea that generation is immaterial.²²

Diallo similarly argues that extraction did not begin with generative AI. Photography, archives, museums and digital platforms have long displaced images and cultural material from the people to whom they were connected.¹

Diallo describes her response not as an abandonment of AI, but as a search for greater autonomy within it. She described an ambition to install a model locally, outside the major commercial platforms, so that she could protect her research and work in a more independent environment. At the time of the conversation, this remained an objective requiring funding rather than an already completed technical system.¹

Tsegba adds a different dimension to the question of materiality. She describes the digital realm as a space in which worlds can be invented, inhabited, dismantled and rebuilt with extraordinary precision, while also recognising that this precision can reveal a desire for control. Her increasing return to making by hand reintroduces another kind of knowledge: "The digital opens the possibility of imagining; material practice reminds me that uncertainty and touch are equally important."³

From OOA Gallery's perspective, the question of autonomy therefore concerns more than authorship of a final image. It includes control over the working environment:

• Who stores the conversations and files?
• Who can access the artist's research?
• What happens when a company changes its model or withdraws a service?
• How is a time-based or software-dependent work conserved?
• What does a collector acquire: a file, an edition, a certificate, a display system or a right to present the work?

Museums have developed dedicated conservation methods because computer-based works are vulnerable to software change, hardware obsolescence and the loss of documentation. The Guggenheim's Conserving Computer-Based Art initiative treats acquisition, preservation, maintenance and display as linked problems requiring collaboration between conservators and computer scientists, while MoMA's media conservation practice spans software, video, film, sound and performance-based works and stresses managing technological change while respecting the artist's intent.^{23, 24}

For galleries and collectors, these institutional practices suggest that an acquisition should document not only the visible file, but also the authorised edition, technical dependencies, display conditions and artist-approved strategies for future migration, emulation (recreating an older technical environment on newer systems) or replacement. This is a practical inference from museum conservation practice, not a universal rule for every digital artwork, and a static digital still does not carry the same dependencies as a software installation or multi-channel video.^{23, 24}

These questions do not diminish the artistic significance of AI-based work. They form part of its material reality.

What Art Can, and Cannot, Change

The final argument concerns the scale of artistic agency. These works can alter the images, archives and debates circulating around AI, but their capacity to transform commercial models, ownership structures or technical infrastructures remains limited.

Forms of artistic agency

The presence of African and diasporic artists within generative AI does not automatically transform the models themselves. Publication online does not establish that a work will enter a future training dataset. Nothing in the sources considered here allows us to assume that, if it does, the artist will be informed, credited or compensated.

Yet artistic participation can still act on several levels. Dounia constructs project-specific archives and datasets in response to absences she identifies. Atairu turns representational failures into concrete objects of investigation. Attafuah uses speculative digital figures and environments without allowing one technology to define her entire practice. Àsìkò brings research, photographic presence and constructed environments into relation while insisting that digital images cannot replace lived cultural experience.¹⁰ Diallo develops characters through interiority and sustained dialogue. Okaro insists that cultural specificity and responsibility cannot be delegated to a medium.² Tsegba asks what kinds of memories remain encounterable under conditions of digital excess.³

In each case the machine never appears alone. It arrives with an archive, a vocabulary, a body of images, an infrastructure and a history of classification, and it meets an artist who brings another history: photographic experience, cultural knowledge, a collage practice, a research method, a set of refusals. Read together, the works suggest that artificial intelligence is neither a neutral tool nor an autonomous creator, but a contested visual territory. Historical absences may be repeated inside it, or exposed. Existing images can disappear into abundance, or be made newly encounterable through collage. Generative systems can offer speculative bodies, while photography insists on the particularity of an encounter.

Their work can change the images circulating in contemporary culture even when it does not change the parameters of a commercial model. It can influence viewers, students, curators, developers and other artists, and establish a record against which future systems may be judged. The objective is not to replace one fixed visual canon with another, but to make the field more plural, more situated and more aware of the histories embedded within its technologies. The most important contribution of these artists may therefore be neither their acceptance nor their rejection of AI, but their refusal to approach it passively. They ask what the system has inherited, identify what it cannot see, and construct new archives, protocols and worlds from within its limitations.

Diallo condenses this position in a single sentence:

"AI does not change my relationship with the world. It enlarges it."¹

The distinction matters because the system can extend a field of imagination without becoming the source of the artist's lived experience, cultural knowledge or ethical responsibility. That is what allows critical awareness and imaginative openness to coexist. Near the end of the interview, Diallo returns to the metaphor of the mirror. The image is deliberately personal rather than universal: the machine reflects what she brings into the exchange and extends the field in which she can examine it.¹

"You create the mirror of the beauty of your own life, and the machine extends the lyrical field of that profound beauty. The reason I do all this is for the love of life."¹