People in architecture videos: AI casts, stock entourage or real people, and how to choose

A building with nobody in it can look like a model, and a building full of the wrong people can look like an ad for something else. Architects have argued about entourage in renderings for decades; video makes the choice harder, because people move, faces draw the eye and anyone recognizable needs permission. This guide covers why people in architecture videos help, the three places they can come from (an AI-generated fictional cast, stock or renderer entourage, and real people you film or photograph), what a model release covers, how to cast for the place, and how to keep faces from turning uncanny.

Key takeaways

• People give a building scale and show how it's used; in a short vertical film, a few figures at mid-distance usually do more than a crowd.
• Real, recognizable people need a signed model release for commercial use; Shutterstock, for one, treats even a partially visible face as recognizable.
• Stock cutouts pasted into a still don't animate well, so remove them from images you plan to turn into AI video.
• An AI-generated cast avoids releases for real people, but New York has required ads to disclose AI "synthetic performers" since June 2026, and platforms may add their own AI labels.
• Cast for the place and its users, and review every figure for uncanny faces and for anything that suggests who is or isn't welcome.

Should you put people in an architecture video?

Yes, in most cases. People give a building scale, show how a space is used and make a short film feel inhabited rather than modeled. The real decision is where they come from: an AI-generated fictional cast suits AI films and concept work, stock or renderer entourage suits rendered animations, and real people suit films of finished projects, provided everyone recognizable has signed a release. Keep them few, at mid-distance and doing ordinary things, and cast them to reflect the place.

Figures do three jobs an empty frame can't. They set scale: a person beside a 12-foot window tells you it's tall without a dimension line. They show use: someone reading in the window seat explains the window seat. And they set the hour and the mood, from a morning coffee to a crowded evening courtyard.

They also cost attention. Viewers look at faces before facades, so every person on screen competes with the architecture. That's why people in architecture videos work best small, in profile or turned away, and busy with something calm.

Three ways to get people into the frame

Most studios end up using more than one: real people in the film of a finished café, a fictional cast in the concept film for the next one.

Real people: what a model release covers

A model release is the signed permission a person gives for their likeness to be used, typically for commercial purposes such as marketing your studio. If a client, resident, employee or passerby can be recognized in your project video and the video promotes your business, assume you need one.

Recognizable is broader than most people expect. Shutterstock, whose rules for contributors are a useful benchmark for commercial use, says "a release is needed for every identifiable person in the image," that "a partially visible face is still considered recognizable" and that "even partial silhouettes can require a release" (Shutterstock contributor support, as of October 2026). Those are an agency's licensing rules, not a statute, but they show where professionals draw the line.

Practical habits for a finished-project shoot:

1. Ask before the shoot, not after. Owners who appear in their own home, staff in a restaurant and employees in an office all need to agree in writing.
2. Keep the release with the project files. Note which images and videos it covers, so you can reuse them in a new film later.
3. Treat children with extra care. A parent or guardian signs, and many studios simply keep children out of marketing images.
4. Shoot empty when you can't get releases. An empty space at the right hour often reads better than a busy one.

Privacy rules for people photographed in public vary by state, and how a release must be worded is a legal question; this is general information, not legal advice. For the wider question of who owns project photos, see photo licensing for video and AI. For films that put a real client on camera, client testimonial videos covers how to ask.

Stock entourage and 3D people from your renderer

Entourage is the classic answer for renderings: cutout photographs of people placed into the scene, or 3D people from your renderer's asset library. Both work well in stills. In video, they behave differently.

Cutouts in a still you plan to animate. An image-to-video model sees a cutout as part of the picture. It may animate a flat figure strangely, merge it with the background or redraw its face. If you're turning renderings into AI video, render a version without cutouts.

3D people in a rendered animation. Many real-time renderers ship animated people that walk, sit or talk on loops inside your model. They're consistent from frame to frame, because they're geometry, but they can look stiff up close. Keep them at mid-distance, and read the asset library's license for commercial use; terms differ between tools and between free and paid content.

Stock photos of people. Licensed stock imagery of people normally comes with releases already obtained by the agency, but licenses usually restrict sensitive uses and implied endorsement. Read the license before a stock figure ends up in a client's housing ad.

An AI-generated fictional cast

The newest option is a cast the video model creates: people who look real but aren't based on anyone. For a film made from renderings of an unbuilt project, that's often the only practical way to show the spaces in use.

The appeal is real. No one to schedule or release, and the people fit the generated light and motion because they're made by the same model. Arch2Video works this way: every film includes a small cast of original, fictional adults in at least one scene, and people already in your photos aren't animated; they're replaced by that cast. There is no option for a film without people, so if you need an empty sequence, use a rendered animation instead.

Two things to watch.

Disclosure in advertising

A fictional person in an ad is a regulated thing in at least one state. New York's synthetic performer law, signed December 11, 2025 as Chapter 617, defines a synthetic performer as a digitally created asset intended to look like a human performer "who is not recognizable as any identifiable natural performer," and requires whoever produces an ad for a commercial purpose to "conspicuously disclose" that a synthetic performer appears in it, where that person has actual knowledge (New York State Senate, S8420A, as of October 2026). It took effect 180 days after signing, in June 2026, with civil penalties of USD 1,000 for a first violation and USD 5,000 after that. If you run a project film as a paid ad that reaches New York, add a plain line such as "People shown are AI-generated." Platforms may also apply their own label; see Meta's AI info label on ads. This is not legal advice.

Accuracy and expectations

An AI cast shows how a space could be used, not how it is. On a concept film that's honest. On a film of a finished building, a fictional crowd can imply a level of use that isn't true, so caption it accordingly.

Casting: who appears, and why it matters

Whoever appears in the film is a statement about who the building is for. A school film with only one kind of family, or an apartment film where every resident is the same age, sends a message whether you meant it or not.

Cast the way a good brief would: reflect the people who live or work in the place and the people the client hopes to reach. That usually means a mix of ages, backgrounds and, where it fits the building, people with visible disabilities using the space as designed.

For US residential marketing this goes beyond taste. The Fair Housing Act's advertising rules cover the people shown in ads for housing, including AI-generated ones; what the Fair Housing Act means for rendering videos and AI casts explains the principles and what to do when you can't control who the model generates. With any AI cast, review it the way you'd review a casting call, and hold back a film that suggests a preference before it goes out.

Avoiding uncanny faces

Faces are where AI video, and stiff 3D entourage, most often look wrong, because viewers are experts at faces. Most problems can be prevented:

• Keep people at mid-distance. A figure crossing a courtyard reads as life; a face filling a third of the frame invites scrutiny.
• Favor profile and back views. Someone looking out of the window shows the view; someone staring into the lens breaks the spell.
• Give them calm, simple actions. Walking, sitting, reading and pouring coffee hold up; handshakes, children playing and groups crossing paths are where hands merge and limbs multiply.
• Remove figures from your source images. Small or cropped people in an image give the model something to distort.
• Watch at full screen before publishing. Check faces, hands and feet on the floor, then drop the scene if a large face is wrong (fixing AI video artifacts covers when to drop).

Frequently asked questions

Do I need a model release for people in the background of my project video?

If they can be recognized and the video promotes your business, plan on it. Shutterstock, for example, treats even partially visible faces and some silhouettes as recognizable. People who are too small, blurred or turned away to be identified are usually treated differently, but the safest options are an empty frame, a released person or a fictional one. Rules vary by state; this is not legal advice.

Can I make an Arch2Video film without any people in it?

No. Every Arch2Video film includes a small cast of original, fictional adults in at least one scene, and there is no option to turn it off. If a project needs a film with nobody in it, a rendered animation from your 3D model is the better tool.

Will the AI animate the people who are already in my photos?

In Arch2Video, no: people in your uploaded photos aren't animated and are replaced by the fictional cast. Other image-to-video tools may try to animate them, often with distorted faces, so remove figures from source images unless you want that.

Do I have to label AI-generated people in an ad?

In New York, yes in most cases: since June 2026, an ad produced for a commercial purpose must conspicuously disclose a synthetic performer when the advertiser knows one is there. Other states may follow, and ad platforms have their own AI labeling. A short line in the caption is an easy habit; this is not legal advice.