Studios that once worried only about naming a human assistant, a printmaker, or a studio technician now face a different question: how does one acknowledge a piece of software, or a trained model, without implying it holds artistic intent it does not possess? The matter is not merely etiquette. It touches on contracts, gallery wall text, grant applications, and the reputations of practitioners who wish to be neither evasive nor grandiose about their methods.

This article sets out a working approach for Ochre & Circuit readers: how to think about the distinction between a collaborator and a tool, how to phrase credit lines for varying degrees of involvement, and how to adjust that language depending on whether the audience is a curator, a licensing officer, or a casual visitor. Readers with contractual or copyright concerns should consult a qualified intellectual property solicitor, as the legal status of AI-assisted work remains unsettled in most jurisdictions as of this writing.

Why Crediting Language Has Become Contested

Disputes over AI attribution tend to arise from three directions at once. Collectors and juries want to know how much human decision-making shaped a finished work, since this affects both valuation and eligibility for certain prizes. Funding bodies, including several regional arts councils in the United Kingdom and parts of the European Union, have begun asking applicants to disclose AI involvement in a declared percentage or descriptive band. And peer practitioners, who have watched the term "collaborator" applied loosely to software, have grown wary of credit lines that appear to borrow the language of partnership to add novelty or technical mystique to a piece.

A further complication is that the tools themselves vary enormously in what they contribute. A text-to-image generator that produces a finished composition from a single prompt operates quite differently from a code-assisted generative system that a studio has trained for months on its own archive of marks and then steers through hundreds of iterative adjustments. Treating both scenarios with identical language, such as calling each one "AI-assisted," flattens a distinction that juries and collectors increasingly want preserved.

The 2023 case of a Colorado state fair submission, in which a Midjourney-generated image won a digital art category before the method became widely known, is often cited in studio discussions as the moment credit language became a public concern rather than a private courtesy. Whatever one's view of that outcome, it demonstrated that vague or absent disclosure invites retroactive scrutiny that is considerably harder to manage than a clear statement made at the outset.

What Counts as a Collaborator Versus a Tool

A tool executes a defined function without exercising discretion over the work's direction. A collaborator, by contrast, is generally understood to introduce choices that the artist did not fully specify in advance and could not have predicted. Under this working definition, a camera is a tool even though it mediates light in complex ways, because the photographer retains authority over framing, timing, and selection. The question with AI systems is whether their outputs introduce comparably unpredictable choices, and whether the artist's role becomes closer to selection and editing than to origination.

Several factors help a studio decide where a given use falls:

• Degree of prompt specificity. A highly detailed prompt that constrains composition, palette, and subject leaves less room for the system to introduce unplanned decisions, which leans the description toward "tool."
• Volume of iteration and curation. An artist who generates forty variations and selects one is exercising editorial judgment, but the generative step itself may still have contributed substantial unplanned content.
• Post-generation alteration. Work that is substantially repainted, recomposed, or recoded after an AI-generated starting point shifts authorship back toward the human maker.
• Custom training versus off-the-shelf use. A model trained specifically on an artist's own archive, over weeks or months, functions differently from a general-purpose public model used for a single session.

None of these factors settles the matter alone. A studio in Bristol that trained a model for fourteen months on its own ceramics archive, then used it to suggest glaze patterns it later hand-applied, reasonably described the system as a "generative collaborator" in its exhibition notes, since the sustained training and the unpredictability of its suggestions warranted stronger language than a single prompt session would.

Sample Credit Lines for Different Levels of AI Involvement

Rather than relying on a single catch-all phrase such as "AI-assisted," studios are better served by matching the credit line to the actual degree of involvement. The table below offers starting templates that can be adapted to house style.

Naming the specific model or software, such as Midjourney, DALL-E 3, Stable Diffusion, or a bespoke in-house system, is increasingly expected by exhibition organisers and is good practice regardless of expectation, since it allows viewers and researchers to understand the technical lineage of the work rather than treating "AI" as a single undifferentiated category.

Avoiding Common Overstatements and Understatements

Overstatement usually takes the form of language borrowed from human partnership, such as describing a prompt-based image generator as a "co-artist" or "creative partner" when the artist's total involvement was a short text prompt and a single selection from a handful of outputs. This phrasing can mislead collectors about the labour and judgment embedded in the piece, and may also invite unwelcome scrutiny if a juror later asks the artist to describe the "collaboration" in detail and finds the account thin.

Understatement is the opposite failure: describing extensive AI generation as merely "digital techniques" or omitting mention of AI entirely in a context where disclosure is expected, such as a grant application or a juried exhibition with explicit AI disclosure requirements. The Royal Photographic Society and several print biennials now require explicit disclosure forms, and a failure to disclose, even if unintentional, can result in disqualification after the fact rather than a simple correction.

A middle failure, less discussed, is vague hedging: phrases such as "created using emerging technologies" or "explores new tools of image-making" that avoid naming AI at all. This language tends to read as evasive once a viewer suspects AI involvement, and it rarely satisfies disclosure requirements that ask for specific methodology.

A practical check that several studios have adopted is to write the credit line, then ask whether a technically informed stranger, reading only that line, could reasonably guess the approximate proportion of the work generated by the system versus shaped by hand. If the answer is no, the line likely needs more specificity rather than more enthusiasm.

Adjusting Credit Language for Different Audiences

Wall text for a gallery visitor and a declaration for a funding body serve different purposes and can reasonably differ in tone, though not in substance. Wall text might read: "This piece was developed with the assistance of a generative image model, guided and selected by the artist across approximately sixty iterations." This gives a general audience enough context without technical jargon.

A grant or competition declaration typically requires more precision, often in a standardised field such as "percentage of final content generated by AI" or "description of AI tool and version." Arts Council England's guidance, updated in 2024, asks applicants to describe both the tool used and the nature of the artist's creative control, which means a one-line credit is insufficient for that context and a fuller paragraph, sometimes two hundred words or more, should be prepared separately.

Licensing and commercial contexts raise a third set of concerns, since a buyer may need to know whether the work can be reproduced without infringing a model provider's terms of service, a question that depends on which company's tool was used and under what licence. This is a matter for a solicitor familiar with intellectual property and technology licensing, not something a credit line alone can resolve, and studios selling AI-assisted work commercially would do well to seek that advice before finalising sale terms.

Common Mistakes

Studios most often err by reusing a single generic phrase, such as "AI-assisted," across every piece in a show regardless of how differently each work was made, which erases distinctions that juries and collectors have begun to expect. A second frequent error is updating credit language only after a challenge arises, rather than documenting the AI workflow, including prompts and iteration counts, at the time of making, which leaves the artist reconstructing details from memory under pressure.

Practical Next Steps

Studios wishing to formalise their approach might begin by drafting a short internal style sheet, modelled on the table above, that matches credit phrasing to specific workflows already in use. Keeping a simple log of prompts, model names, and iteration counts at the time of creation, even a basic spreadsheet, makes later disclosure far easier than reconstruction after the fact. Finally, where a work is intended for sale, competition entry, or grant application, it is worth drafting the fuller disclosure paragraph before the shorter wall-text version, since trimming a precise statement is simpler than expanding a vague one under deadline.