Ochre & Circuit
Contact
Foundations Updated 2026-10-15 8 min read

You will learn what a well-written process statement should disclose about tools, prompts, and human editing. By the end, you will be able to judge whether a claim of authorship matches the working method described.

How to Read an AI-Generated Artwork's Process Statement
Thomas Aldridge
Written by Thomas Aldridge Research & Fact-Checking Lead
Key points
  • Process statements should name specific software or models, not just the general term AI.
  • Look for the degree of human selection, editing, and curation described in the statement.
  • A vague statement is itself useful information about a work's transparency.

A process statement is often the only document separating an artwork from a mystery. Readers of this publication will encounter such statements appended to catalogue entries, gallery wall text, and online listings, and the quality of these statements varies considerably. Some are written with care and precision; others amount to little more than a gesture toward transparency. This article explains how to read these statements with discernment, what their common phrases actually mean, and what questions a careful reader ought to ask when the wording leaves too much unsaid.

Our purpose here is not to pass judgment on any particular method of making art, nor to suggest that one approach is inherently superior to another. Rather, it is to equip readers and critics with the vocabulary and the skepticism required to evaluate claims that are, at present, rarely standardised and often poorly defined. A well-informed reader will find that most process statements reward close attention, and that the gaps in a statement often tell us as much as its content.

What a Process Statement Is and Why It Matters

A process statement is a short written account, usually between fifty and four hundred words, describing how a given work was produced. In the context of AI-assisted or AI-generated art, this typically includes some combination of the following: the tools or models used, the nature and extent of human input, the number of iterations or generations involved, and any post-processing applied after the image or object was generated. The statement may appear as a wall label, a footnote in a catalogue, or a paragraph on an artist's website.

These statements matter for several practical reasons. First, they affect valuation and provenance; a collector purchasing a print will want to know whether it is one of three variants or one of three thousand. Second, they affect critical interpretation; a work described as "a single unedited generation" invites different questions than one described as "composited from forty generations across six months." Third, they matter for institutional policy, since many galleries and competitions now require disclosure of AI involvement as a condition of exhibition or entry.

It is worth noting that a process statement is a claim made by the artist or the gallery representing the artist, not an independently verified fact. Unless a statement is accompanied by supporting material, such as version histories, exported prompt logs, or timestamped files, readers should treat it as testimony rather than proof. This distinction becomes important later in this article when we discuss vague statements.

Common Phrases and What They Actually Disclose

Certain phrases recur across process statements with enough frequency that readers benefit from knowing their typical, though not universal, meaning. The table below lists common wording alongside what it generally discloses and what it tends to leave unstated.

PhraseWhat It Usually DisclosesWhat It Often Omits
"Generated using a diffusion model"The general family of technology employedWhich specific model, version, or fine-tune
"Extensively prompted and curated"Human selection occurred among multiple outputsHow many outputs were generated, and how selection criteria were set
"Hand-finished" or "retouched"Some manual editing followed generationThe proportion of the final image attributable to that editing
"No post-processing"The output was not altered after generationWhether upscaling, cropping, or format conversion occurred, since these are sometimes excluded from the artist's definition of "processing"
"Collaborative process with AI"Iteration between human and model took placeWho made the final selection decisions, and how many rounds occurred

Readers should also note the difference between a statement describing the tool and one describing the method. "Made with Midjourney" names a tool but says nothing about method; it could describe a single prompt or a six-month iterative project involving hundreds of generations. By contrast, "forty generations refined through inpainting over three editing sessions" describes a method regardless of which tool was used. Statements that name only the tool, without method, generally disclose less than they appear to.

A further phrase worth attention is "trained on my own archive" or similar wording suggesting a custom or fine-tuned model. This claim, when accurate, is significant, since it implies the artist built a dataset from personal work rather than relying solely on a general-purpose model. However, this claim is difficult for a reader to verify without technical documentation, and some statements use this language loosely to describe curation of prompts rather than actual model training.

Questions to Ask When the Statement Is Vague

Many process statements, particularly those written for general audiences rather than specialist ones, are vague by necessity or by choice. When a reader encounters a statement that does not specify method, the following questions can help determine how much weight the statement should bear in forming a judgment about the work.

  • How many outputs were generated before this one was selected? A statement silent on this point may describe either a first attempt or a selection from thousands; the difference affects how one reads claims of artistic intention.
  • What counts as "editing" in this statement? Cropping and color correction are different in kind from inpainting or compositing multiple generations together, yet both are sometimes grouped under "minor adjustments."
  • Is the described process repeatable, or unique to this piece? A workflow applied consistently across a series suggests a developed method; a one-off description suggests the statement may have been written after the fact to fit a single result.
  • Does the statement specify a model version or date? Models change over time, sometimes substantially, and a statement that omits version information makes the work difficult to situate historically or to compare with later pieces made using updated tools.
  • Who wrote the statement? A statement drafted by a gallery's marketing staff may emphasize different details than one written directly by the artist, and readers should ask whether the voice matches other documented accounts from the same artist.

When a statement cannot answer these questions, even informally, readers and critics should feel free to say so in their own writing. Noting "the artist's process statement does not specify the number of generations involved" is a legitimate and useful observation, not an accusation. It simply marks the limit of what can responsibly be concluded from the available documentation.

Comparing Process Statements Across Different Media

Process statements differ meaningfully depending on the medium of the final work, and readers should calibrate their expectations accordingly. A statement accompanying a printed, framed image carries different implications than one accompanying a generative video, an interactive installation, or a piece involving physical fabrication such as a 3D-printed sculpture.

For static images, the central questions tend to concern iteration count, selection criteria, and post-processing, as discussed above. For video and animation, an additional layer of disclosure becomes relevant: whether frames were generated individually and assembled, whether a single temporal model produced continuous motion, and how long the raw output was before editing. A statement for a ninety-second video that says only "AI-generated" leaves open whether this took one generation pass or required dozens of discarded attempts stitched together.

For works involving physical fabrication, process statements should ideally address the handoff between digital generation and physical production. A sculpture described as "AI-generated form, 3D printed in resin" raises questions about whether the digital model required manual repair or simplification before printing, since generated 3D meshes frequently contain errors that require a technician's intervention. Readers interested in fabrication details, and those considering a purchase of a physical piece, may wish to consult a conservator or framing professional regarding material longevity, since this falls outside the scope of a process statement and outside the expertise most galleries can offer.

For sound and music works generated or assisted by AI, process statements should ideally specify whether the audio was generated end-to-end or whether AI was used for a discrete component, such as generating a melodic line later arranged by a human composer. The comparison across media makes clear that no single checklist item applies universally; the reader's questions must adapt to what the medium makes possible and what it makes difficult to verify.

Common Mistakes

Readers new to this subject sometimes assume that a long process statement is automatically more honest than a short one, when in fact length often reflects writing style rather than disclosure depth. A second common mistake is treating the presence of technical vocabulary, such as naming a specific model or sampler, as proof of rigor, when such terms can be included without any corresponding detail about iteration or selection. A third mistake is expecting process statements to be independently audited; at present, almost none are, and readers should not assume institutional vetting beyond what the statement itself claims.

Practical Next Steps for Readers and Critics

Readers approaching a process statement for the first time will benefit from a brief, repeatable checklist rather than an exhaustive investigation each time. The following steps are offered as a starting point, to be adapted to the medium and context at hand.

  1. Read the statement once for tone, then again specifically to identify which of the five elements, tool, method, iteration count, human input, and post-processing, are named and which are absent.
  2. Note whether the statement specifies a model name and version or date; if absent, treat claims about technique with appropriate caution.
  3. Compare the statement against other works by the same artist, where available, to see whether the described method is consistent or appears tailored to a single piece.
  4. Where the work is physical, check whether the statement addresses the digital-to-physical handoff, and consult a conservator or fabrication specialist if material questions arise that the statement does not answer.
  5. When writing criticism or cataloguing notes, distinguish clearly between what the statement claims and what has been independently verified, using phrases such as "the artist states" rather than presenting claims as settled fact.

A process statement, read carefully, offers a genuine window into how a work came to exist, but it remains a document written by an interested party. Approaching it with the same careful, courteous skepticism one would bring to any provenance record will serve readers well, and will, in time, encourage artists and institutions to write statements that reward such attention.

This article is offered for general information and critical discussion; readers facing specific legal, contractual, or valuation decisions should consult a qualified professional. Disclaimer

Related Articles

Generative Art Tools Compared: Diffusion, GANs, and Rule-Based
Foundations 8 min read

Generative Art Tools Compared: Diffusion, GANs, and Rule-Based

You will see how diffusion models, generative adversarial networks, and rule-based systems differ in how they produce images, and what each approach implies for an artist's control over the result. This comparison will help you choose a method that matches your intentions rather than following trends.

Read article