What counts as generated imagery
Generated imagery here means pictures, and sometimes video or audio, produced by software that creates new content from a text description or from patterns learned from many existing examples. The category is broad and changing quickly. It includes fully synthetic images, edits that add or remove elements of a photograph, and enhancements such as colorization or upscaling of old pictures.
These tools can be useful, but a gallery has a special responsibility. Visitors usually assume that what an institution shows is documented and checked, so a synthetic image can mislead simply by appearing in that setting.
The central risk: confusing a picture with evidence
In exhibits, images often serve as evidence. A photograph of a place or person is taken to show what existed. A generated image does not record anything; it produces something plausible. That can be a problem in several ways:
- Invented detail. A generated scene of a historical event or a past city will include clothing, buildings, tools and faces that the software supplies by guessing. Details may be wrong for the period or place in ways that only a specialist notices.
- Stereotypes. Image tools tend to reproduce common visual patterns from their training material, which can include stereotypes about groups of people or cultures. Be especially careful when depicting communities.
- False authority. A realistic picture can be read as a photograph. Visitors may remember it as a genuine record, and later repeat it.
- Distorted reconstruction. Where a missing portion of an object or a damaged picture is “filled in”, the result mixes evidence with invention, and the boundary is hard to see afterward.
The more realistic the output, the more important it is to say what it is.
Label it clearly
If you use generated imagery, disclose it where visitors can see it, close to the image, and in plain words.
- State that it is computer-generated or that it is an artistic impression, not “a photograph” or “a record”.
- Say what it is based on, if you can do so honestly: for instance, “based on written descriptions and surviving remains”, and note that details are uncertain.
- Give the same disclosure in every format: the screen, any printed material, captions and transcripts, and the alternative text for people using screen readers. Writing captions and transcripts for exhibit media is a good moment to check this.
- Keep the label visible if the image is shared or captured, such as in a downloadable or social media version. A small watermark or credit line can help, though it can be cropped out, so treat it as additional protection only.
- Avoid ambiguous words such as “enhanced” without explanation.
Disclosure does not remove the need for judgment, but it lets visitors adjust their trust.
Questions to ask before using it
- Is the image needed? Could a real archive photograph, a drawing by a named artist, a diagram or a plain description do the job?
- Is it presented as evidence or illustration? Illustration for mood or explanation is easier to defend than a picture implying a documented scene.
- Has an expert reviewed it? A historian, archaeologist or curator with relevant knowledge should check generated historical or cultural images for errors.
- Does it depict real people or communities? Generating the likeness of a real person, living or dead, raises questions of consent and respect. Images of cultural subjects should be reviewed with the people concerned; see consulting source communities and cultural safety.
- What are the rights? Rules about who owns generated images and about the material tools were trained on differ and are unsettled in many places. See copyright and image permissions, and check the terms of any tool you use. If the position is uncertain for something prominent, consider another source.
- Does it replace a person’s work? Commissioning illustrators and designers supports the creative community and often produces more accurate, accountable work.
Restoration and enhancement
Using software to sharpen, colorize or fill in damaged historical images is a special case. These changes can look appealing but add guesses: the color of a dress, the face behind a blur, the text on a sign. If you show an altered version, display the original alongside or make it available, and label the changes. Keep unaltered originals preserved. Where accuracy matters, as in documentary or evidential material, think hard before altering at all.
Interactive and live generation
Some exhibits invite visitors to create images with a generation tool. Additional issues arise: the tool may produce offensive or inaccurate content, may store what visitors type, and may use their prompts or images for other purposes. Check how the tool handles data, tell visitors what happens to their input and add filters and staff plans for problems. Treat this as a privacy matter too, as outlined in visitor privacy. If visitors’ creations appear on a shared display, apply the same moderation thinking you would for any public contribution. Conversational exhibits raise similar questions about accuracy and disclosure.
Accessibility and honesty in the text
Alternative text and audio description for a generated image should describe it honestly as an image, and not narrate it as if it recorded real events. Plain language helps visitors understand a distinction that can be subtle: “an artist’s impression made with computer software” is clearer than technical terms.
A cautious position
A reasonable stance for a gallery is to use generated imagery sparingly, label it plainly, have knowledgeable people review it and avoid it where the picture would be taken as evidence or where it depicts real people and sensitive subjects. The technology will keep changing, so write your policy as principles rather than tool-specific rules, and revisit it regularly. The aim is the same as for every other kind of exhibit content: visitors should be able to tell what is documented, what is interpreted and what is imagined.