AI Photo & Video Guidelines

Using AI To Create Photos and Videos

AI tools can be useful early in a project: a placeholder for a wireframe, a rough visual to pitch a concept, a quick animatic before a shoot. They are not a substitute for real photography, video, or illustration in anything that gets published. Before an AI-assisted image or video goes live, it needs to meet the same brand and accessibility standards as any other asset on the site, and a person needs to review it.

Do

Use AI for early drafts and exploration, understanding the version that gets published will be real photography, real footage, or brand-approved artwork.
Use AI for cosmetic cleanup on real photography or video: removing stray hairs, reducing shine, or taking out a blemish or a piece of trash in the background.
Use AI for practical edits: cropping, resizing, color correction, lighting adjustment.
Use AI to prototype a layout, infographic, or storyboard concept as a starting point for your own design work, not as the final asset.
Label an AI-generated image or video when it is published, the same way you would credit a photographer. Example: Image created with AI tool name.
Have a person review any AI-assisted image or video before it goes live, same as any other content on the site.

Don't

Don't publish an AI-generated image or video that could mislead, misinform, or misrepresent an event, a place, or a person.
Don't generate anything photorealistic that depicts an event or scene that did not happen. See the table below for the specific subjects that are off-limits to generate, including students, faculty, and staff who do not exist on our campus.
Don't create, modify, or imitate the W&M wordmark, logo, seal, coat of arms, or any other trademarked mark.
Don't depict campus buildings, structures, art, or grounds that do not exist yet as if they are already built, and don't add invented features to a real building, such as an extra floor, a statue, or different signage.
Don't publish AI slop: low-effort, low-quality images or video with no human oversight and no real value to the person looking at it.

What Can and Cannot Be Generated

Breakdown of what can and cannot be generated by AI
Subject	Ruling	Where to go instead
The Wren Building, historic buildings, campus landmarks	Do not generate	These are real, documented places and a generated version reads as wrong to anyone who knows campus. Canto provides real photography of campus.
Students, faculty, staff, alumni	Do not generate	AI-generated images can misrepresent the W&M community. Canto provides these images.
Campus grounds and identifiable campus scenes	Do not generate	A generated scene can invent signage, maps, or directional markers that read as real, which misinforms visitors and works against the wayfinding project underway across campus. Canto provides real photography of campus.
W&M logo, cypher, seal, coat of arms	Do not generate	AI almost always gets this wrong. Canto provides these brand marks.
Abstract textures, patterns, background gradients	Acceptable to generate with conditions	Use brand colors.
The Griffin mascot	Acceptable to generate with conditions	AI-generated images of the Griffin must include an Eagle's head, lion's body, lion's tail, wings, and must be green and white. Only use in athletics and spirit contexts. Canto provides the official Griffin logos.

Common AI Mistakes To Look For

AI-generated images, video, and brand marks can look convincing at first glance and fall apart on a closer look. William & Mary do not allow AI-generated images of campus buildings, students, faculty, staff, alumni or any W&M trademarked graphic. These are common issues to spot as you come across AI-generated material.

In Images

Hands and fingers: extra or missing fingers, fingers that fuse together, joints that bend the wrong way.
Text and signage: garbled, misspelled, or nonsensical lettering on signs, books, screens, or clothing.
Eyes and teeth: pupils that do not match between eyes, asymmetric eyes, too many or uneven teeth.
Repeating patterns: brick, fabric, or fencing that warps, repeats unnaturally, or dissolves into noise partway across the image.
Lighting and shadows: shadows falling in different directions than the light source, or reflections that do not match what they are reflecting.
Object counts: extra limbs, duplicated background objects, or a crowd with the same face repeating.
Edges: haloing or blur where a subject meets the background, especially around hair.
Wrong buildings: AI often renders different colonial buildings instead of those on campus, most often the Governor's Palace or the Capitol building in Colonial Williamsburg, since both are colonial-era brick in the same town.
The cupola: a frequent tell on the Wren Building. AI-generated cupolas are often the wrong shape, sit in the wrong spot on the roofline, or are borrowed from a different building's tower entirely.
Invented text: cornerstones, historical markers, or building plaques with lettering that is garbled or does not match what is actually inscribed.

In Video

Flicker and morphing: textures that shimmer, or objects that subtly change shape from one frame to the next.
Unnatural motion: movement that looks floaty or weightless, or limbs that bend in ways a body cannot.
Lip sync: mouth movement that drifts out of sync with the audio.
Shot-to-shot consistency: a background detail that changes between two shots meant to be the same location.

In Logos and Marks

A mark that is close to the W&M logo but slightly off: wrong typeface, a warped cypher or seal, an extra flourish, or a green or gold that does not match the approved values.
Malformed or garbled lettering in any generated wordmark.
Asymmetry or an incorrect element count in a seal or crest, such as the wrong number of shield sections.

If you catch yourself checking whether something is real, that is the signal to stop and go to Canto instead of trying to fix it. When in doubt, ask: could this be mistaken for a real photo or video of campus, a real person, or the official logo? If yes, it does not get published.