Machines can now generate photorealistic faces, abstract compositions, and stylistic pastiches of named painters in under a minute, using tools such as Midjourney, Stable Diffusion, or DALL-E. This capability has prompted a recurring question in galleries, classrooms, and online forums: if a model can produce an image that looks finished, skilled, and even emotionally resonant, what work is left for art to do? The question assumes that art and image production are the same activity measured on the same scale of competence.
They are not the same activity, and the distinction matters for anyone trying to evaluate, collect, teach, or make work in a period when image generation has become cheap and fast. This article separates the two terms, reviews what art historians mean when they describe art's social function, and examines two competing explanations for why art retains value: the scarcity argument, which is weakening, and the intentionality argument, which currently holds. It closes with a practical checklist for assessing any given image.
Separating Image Production From Art-Making
Image production refers to the technical act of generating a visual artifact: pigment applied to a surface, light recorded onto film, pixels assembled by a trained model. Art-making refers to a practice embedded in intention, context, reception, and a historical lineage of prior work that the new object responds to, consciously or not. A photocopier produces images. A printing press produces images. Neither development abolished painting, because neither altered the social and intentional conditions that make an object function as art.
Marcel Duchamp's Fountain (1917), a urinal submitted to an exhibition under a pseudonym, demonstrated this separation explicitly: the object required no manual skill to produce, yet its status as art derived entirely from the framing decision, the submission, and the argument it made about authorship and selection. A diffusion model generating an image from a text prompt is, categorically, performing image production. Whether the result also functions as art depends on conditions external to the generation process itself.
What Historians Mean by the Function of Art
Art historians typically describe several overlapping functions that artworks have served across different periods: commemorative, documentary, devotional, status-marking, and ideological. The cave paintings at Chauvet, dated to roughly 30,000 BCE, likely served ritual or documentary purposes tied to specific communities and practices now only partially understood. Dutch guild portraits of the 17th century marked social status and recorded commissioning relationships between sitter, patron, and painter. Soviet agitprop posters of the 1920s served an explicit ideological function, designed for mass reproduction and persuasion rather than private contemplation.
These functions depend on human social context: who made the object, under what commission or constraint, for which audience, and within what set of accountable relationships. A visually similar image produced without any of that context does not automatically inherit the function. This is the core reason historians resist collapsing "art" into "well-made image": the function was never located solely in the image's visual properties, but in the network of human relationships the image was embedded in.
The Scarcity Argument and Why It Is Weakening
One common explanation for art's value is scarcity of skill. Grinding pigments, mastering linear perspective, or achieving anatomically convincing figure drawing required years of training, and the resulting rarity of competent practitioners supported both the price and the prestige of finished work. Michelangelo's Sistine Chapel ceiling, painted between 1508 and 1512, represents an extreme case of time investment: roughly four years of sustained physical and technical labor that very few people on earth were capable of executing.
Machine generation collapses this particular scarcity. A trained illustrator might spend 8 to 20 hours producing a detailed photorealistic portrait; a text-to-image model can produce a visually comparable image in under a minute, at near-zero marginal cost per additional image. If scarcity of technical competence were the sole source of art's value, this collapse would indeed threaten art's status. The scarcity argument, however, conflates "art" with "skilled image production," which is precisely the conflation the historical record does not support, since plenty of historically significant works, including folk art, outsider art, and readymades, were never valued primarily for technical scarcity.
The Intentionality Argument and Why It Still Holds
The intentionality argument locates art's distinctiveness not in rarity of skill but in the presence of a directed human purpose that is revisable through judgment and exercised under some form of accountability. Three terms matter here: intention, a formulated communicative purpose; judgment, the iterative decision-making that responds to that purpose during the making process; and constraint, the material or conceptual limits the maker works within and against.
Alice Neel, for instance, would sometimes leave backgrounds unfinished in her portraits, a judgment call signaling unresolved psychological material in the sitter's presence. A generative model has no comparable stake in its output. It has no commitment to the subject matter and no accountability for the claim an image might make about a real person or event. It can be prompted to simulate the visual appearance of an unfinished background, but it cannot be said to have judged that the background should remain unresolved for a reason tied to the subject.
| Dimension | Human art-making | Machine image generation (current) |
|---|---|---|
| Revisability tied to purpose | Present; choices respond to an evolving intention | Limited to prompt or parameter edits, not internal judgment |
| Accountability | Traceable to a named maker or workshop | Diffuse; model, dataset, and prompter share unclear responsibility |
| Context of production | Embedded in commission, tradition, or social stakes | Typically absent unless supplied by a human user |
Where Machine Output Currently Falls Short
Beyond the conceptual argument, there are concrete technical gaps as of 2024. Diffusion models still produce inconsistent hand and anatomical rendering without additional correction passes such as inpainting. Maintaining a single consistent character across a multi-image project typically requires supplementary tools, such as LoRA fine-tuning or ControlNet conditioning, and even then a human operator often generates a large batch, sometimes 50 to 200 variations, to select a usable set of 5 to 10 images.
Training data provenance also remains legally unresolved. Getty Images filed suit against Stability AI in 2023 over alleged unauthorized use of copyrighted images in training data, and the outcome of such cases may affect how machine-generated work is classified for commercial or exhibition purposes. Readers evaluating a specific project's legal standing should consult an intellectual property attorney rather than rely on general commentary, since the law in this area is actively being litigated and varies by jurisdiction.
Where This Distinction May Shift in the Future
The distinction between image production and art-making is not fixed for all time; it depends on evolving institutional and legal definitions of authorship. The United States Copyright Office issued guidance in 2023 stating that purely AI-generated material, without sufficient human authorship, cannot be registered for copyright, while human-selected or human-arranged combinations of AI elements may qualify under certain conditions. This guidance may be revised, and artists working with generative tools should confirm current requirements with a copyright professional before relying on it for a specific project.
Some artists already treat generative models as collaborators rather than tools, where the authorial act resides in iterative curation. Mario Klingemann has used generative adversarial networks since roughly 2016, with his role centered on selecting, training, and framing outputs rather than generating a single image. If institutions eventually formalize collaborative human-machine authorship as its own category, the line discussed in this article would shift accordingly, without necessarily eliminating it.
Common mistakes
- Treating visual polish as proof of artistic intention, when polish is a property of the generation process, not evidence of a communicative purpose.
- Assuming increased technical complexity in an image automatically increases its artistic content.
- Relying on scarcity of skill as the sole justification for art's value, which ignores devotional, documentary, and ideological functions historians have long documented.
- Assuming any machine assistance disqualifies human authorship entirely, which overlooks that brushes, cameras, and printing presses have always mediated art-making without eliminating it.
A Practical Summary for Discussion
When evaluating whether a specific image functions as art rather than as mere image production, ask five questions: Is there a stated or inferable communicative intention behind it? Is there evidence of revision tied to that intention rather than to technical polish alone? Who is accountable for the claims the image makes, particularly if it depicts a real person or event? What context, commission, or audience was it made for? And how is it being used after creation, as documentation, commentary, decoration, or something else?
These questions will not resolve every borderline case, and formal attribution or valuation decisions, particularly for sale, exhibition, or legal purposes, should involve a curator, appraiser, or attorney with relevant expertise rather than rely on general criteria alone. The checklist is meant for discussion and self-assessment, not as a substitute for professional evaluation in contexts where the distinction carries financial, legal, or reputational weight.
