Answers · People

How to Make AI Videos of People?

Published November 17, 2026

Quick answer: Describe invented people the way a casting note would — age and background first, then specific features — and draw each one once as a reference sheet. On SceneMixer every shot of that person is generated from the sheet, photo-like faces of real people can be refused, and video starts at 10 credits a second.
Still from “The Eagle's Vow”, an AI short drama generated end-to-end by SceneMixer (9 characters, 1 episode)
SampleThe Eagle's VowDigital HD1 episodeProduced end-to-end by SceneMixer from the scriptWatch the episodes

The hard part: the same person twice

Generating one convincing person is easy now. Getting the same person again, in a new outfit or a new scene, is where most attempts fail. One maker working on it day after day described the approach:

All i have done is prompted for the change of clothes and said use the same face in the image provided

That was a face-consistency question in r/generativeAI. About 300 posts in the community corpus ask how to turn a character sheet into a cinematic video, which is the same problem from the other side: a fixed person, moving.

Describe a person, not a type

"A beautiful woman" produces a generic face that varies every time. A casting note produces a person: a woman in her late thirties, Filipino, short black hair with a grey streak, a small scar on the chin, a habit of pushing her sleeves up. Put the age and background in the first sentence of the appearance description, because that is what the sheet is drawn from. If a character comes back looking the wrong age or background, fix that first sentence and regenerate the sheet.

One sheet per person, used in every shot

Each character gets a full-body reference sheet, and face and body crops from it are attached to every segment the person appears in. That is how the same face, hair and clothes survive across scenes and episodes. When the story changes a person — a new outfit, a bandage, a wedding dress — make it a state of the same character with its own image, used only in the segments where it applies.

Full-length reference sheet of a person in a grey blazer over a white shirt with a black belt and black loafers, standing on plain white
A reference sheet from one of SceneMixer's samples: an invented person drawn once, full length. Every shot of this person is generated with this sheet attached.

A live-action look and a believable world

For people who look real, pick a live-action look in Step 1, such as Digital HD or Shot on Film. The project's target region adapts names, cast backgrounds and settings for Western, East Asian, Southeast Asian or Latin American audiences, so a story set in Manila is cast like one. Give each shot a light source and keep two or three people in frame; faces hold up best in small groups with clear light.

Making invented people that look real and stay the sameDescriptionage and background first, then specific featuresSheetone full-body image per person, used in every shotStatesnew outfits or injuries as images of the same personLooka live-action style and a target region for the castShotsa light source and two or three people per frameReal faces are off the table; invented ones are the route.
How an invented person survives from the first shot to the last.

Faces that act

A person on screen is believed through small behaviour more than through skin detail. Write what the face does: she bites her lip before answering; he looks at the door, not at her; she laughs a second too late. Give each person one physical habit in the description and use it in the storyboard. Keep emotional moments in close-ups, where the face has room, and give the reaction its own shot with a silent beat. People in AI video look most real when they are doing something specific in a light you could point to. Ordinary detail helps too: a coffee cup going cold, a coat thrown over a chair, a phone face down on the table. Those details belong in the shot text, where they cost nothing to add, and they make a generated room feel lived in rather than staged.

Why real people are off the table

Reference photos whose faces look like a real person can be refused at generation, and well-known people are refused outright. That protects the people in question, and it keeps your project clear of likeness problems. On a paid plan you can upload a photo as a starting point for a character; the safest route is still an invented face described in words. When a reference image is flagged as a real person's face explains the check.

Limitations of AI videos of people

  • Real and famous people cannot be generated; the cast must be invented.
  • Hands, fine jewellery and crowded scenes still need more attempts.
  • A face described vaguely will drift; specific features hold.
  • Earlier clips keep the old look after a sheet is changed, so regenerate the segments that person appears in.
  • The sheet-to-video count is a rule-based match over a fixed window of public posts.

FAQ

Can I make an AI video of a real person?

No. Faces that look like a real person can be refused and famous people are refused outright. On SceneMixer describe an invented person instead.

How do I keep the same person in every AI shot?

Use one reference sheet per person and let SceneMixer attach it to every segment; outfits become states of the same character.

Why does my AI character look the wrong age?

The sheet is drawn from the first sentence of the description. Put age and background there, regenerate the sheet, then the affected segments.

Which look makes AI people look most real?

A live-action look such as Digital HD or Shot on Film, with a light source in each shot and small groups.

Related: realistic AI videos, fixing a character who looks the wrong age, and the collection method behind the count.

Write one casting note

Age and background first, specific features after. One sheet per person, used in every shot.

Sign up free

Per-second video rates are on the pricing page with the same numbers.