How to Keep the Same Character in Every AI Video Shot

You render a great first shot: a woman in a yellow raincoat at a bus stop. In the second shot she has a different face, and by the third the raincoat is green. This is the most common complaint about AI video, and it has a plain cause.

Why characters change between shots

Each AI video shot is generated on its own. If the model only gets words ("a woman in a yellow raincoat"), it invents a new woman every time who fits the words. Words describe a type of person. They don't pin down one person.

To keep a character, the model has to be shown them, not just told about them. That means pictures: a portrait, or a frame from an earlier shot.

The workflow that works

1. Describe each character precisely

Give each character a name and specific details: age, face shape, hair, a scar or freckles, one distinctive piece of clothing. "Meera, 30s, short curly hair, round glasses, mustard cardigan" is far easier to hold than "a young woman".

2. Make a portrait and approve it

In Voxmin FloStudio, each character in the cast gets a portrait. Regenerate it until it's the person you imagined. This portrait is the character from now on. You can also use your own picture of a character you designed.

3. Pick the video model for the job

FloStudio has two video models, and they hold characters differently:

• Flash Video (the default, faster and cheaper) starts each shot from a first frame. Make that first frame from the cast, and the shot begins with the right person.
• Ultra Video also takes reference pictures for each shot: the cast's portraits and frames from earlier shots showing the same people and places. For a story with several recurring characters, or close-ups where the face matters, choose Ultra Video for those shots.

You can set the model for the whole project or shot by shot.

4. Name the cast in every shot

The shot list says who is in each shot. That's how FloStudio knows which portraits to hand over. A shot that doesn't name Meera won't be given Meera's portrait.

5. Fix the one that slipped, not the whole film

If a face drifts in one shot, give that shot a better reference (an earlier frame where the face is clear) and render a new take. Earlier takes stay in the project, so you can compare and keep the best one.

Tips that make a visible difference

• Keep wardrobe fixed. Change clothes only when the story does, and say so in the shot.
• Light the face. Silhouettes and heavy shadow give the model less to hold.
• Fewer people per shot. Two named characters hold better than five.
• Medium shots over extreme close-ups when continuity matters more than drama.

What still goes wrong

No AI video model is perfectly consistent yet. Small details (a pattern on a shirt, an earring, the exact shape of glasses) can still shift. Plan a retake or two for the shots that matter most, and keep the most demanding close-ups short.

Next steps

• See how the cast, portraits and continuity work in the editor on the consistent character page.
• Making a series? The AI cartoon video maker keeps one cast across episodes.