When clients see the first version of an AI video, the most common question is "why doesn't this person look like the one in the previous shot?" It isn't a mistake; it's how AI image generation currently works. Rather than explain afterwards, we say it up front and design the script around it.
Why they look different
AI videos are usually generated one shot at a time, and every generation is "drawn from scratch". Even with an identical description, the face, the details of the clothing and the shape of a product will vary. Real filming is different: the same person and the same object are recorded by the camera.
What's possible now
- Within one shot: people and objects are broadly stable; short shots of a few seconds are more stable than long ones.
- Across shots: reference images can keep a character roughly similar (hairstyle, skin tone, clothing colour), but details still change.
- Products: anything with a fixed appearance (a logo, a specific shape) is hard for AI to get right every time. We usually recommend compositing real photos; see can AI videos show real products and logos.
- Text: text inside AI images is often garbled, so all on-screen text is added in post.
How to script around it
- Don't run one character through the whole video. Use "different people in different scenes", for example a new user in each segment; viewers won't expect them to be the same person.
- Show characters in profile, from a distance, from behind, or hands only. When the face isn't the point, the differences aren't obvious.
- Make every shot a standalone picture, and tie the story together with narration or subtitles rather than continuous action.
- Anything that must stay consistent (logo, product, a real person) uses real material; AI handles the scenes and the mood.
Which videos suit AI, and which don't
Concept explanations, illustrative scenarios, brand mood and abstract visuals suit AI well. A story with one continuous protagonist, videos needing precise product detail, or testimonials with real people are still better filmed. See what AI video production is.
How we handle it
When we receive a script, we first point out the segments likely to have consistency problems, suggest adjustments, and show you the visual direction for each segment at the storyboard stage. Confirming before generating saves far more time than changing afterwards. See how AI video revisions differ from editing.
Frequently asked questions
Can our staff's photos be used as characters in the AI video?
They can serve as references, but the resemblance is limited and written consent from the person is required. For real people we recommend filming.
Will consistency improve in future?
The technology keeps improving. With today's tools, we promise only what can be done and say so before production.
Is a cartoon or animated style more stable?
Stylised visuals are more forgiving of detail differences, and viewers are less critical of them, so consistency problems are smaller than in a photorealistic style.
