Why AI Video Doubles Props and Swaps Outfits

What goes wrong with props and outfits in AI video?

In one of our own test clips, an actor carries a yoga mat into a room. She steps out of one flip-flop, and a whole pair appears on the floor while her other foot still wears one. Then she lays the mat down and hops over it. Her hair comes out tied back like the model in the product photo, though her own photos show it down. Between two cuts, the room changes too.

A clip gives no reasons, so we read back the brief it was made from. The causes below are likely ones, not proven. They come from our own tests, not a study of every video tool.

Why do props double or vanish?

Three things in that brief stood out.

• A change nobody asked for. The brief had her step out of her flip-flops. The person who asked for the video never asked for that.
• Two lines that pulled against each other. One line said to keep everything on for the whole clip. Another said to step out of the flip-flops. One flip-flop still on and a pair on the floor obeys both.
• One item, several names. The brief called the same footwear slides and sandals. The person called them flip-flops. The words and the pictures may have described two different pairs.

We planned the same video again with those three things fixed, along with other fixes on our side. Every item had one role and one name. The flip-flops stayed on, as asked, and no extra one appeared in our re-runs.

Why does the actor hop or jump?

The brief said where she ended up: standing in the middle of the mat. It never said how she got there. With no step written down, the clip found its own way: a hop.

When we wrote a step out, such as "she steps onto the near end of the mat", the hop did not come back in our re-runs. So write a move for any change of place or height: she steps, kneels, sits or rises.

Why does the outfit change?

Tools that keep a face steady usually send reference photos of the person, and those photos show clothes too. Those clothes can pull against the outfit the words ask for.

Product photos pull too. A photo of clothes on a model also shows that model's hair and build. The tied-back hair in our clip most likely came from there.

So we stopped leaving a new outfit to words alone.

How should you brief each item?

Give every item one role for the scene:

• Worn the whole time. It stays on from the first frame to the last.
• Taken off. Say when, and where it ends up. "The jacket comes off and ends folded on the bench."
• Put on. Say where it starts and when it goes on.
• Held. Say which hand.
• Carried, then set down. Say where it lands.
• Shown to the camera. One still moment, facing the lens.

Then count. "One jacket, one mat, two flip-flops, one on each foot." For a pair, name the side in the wearer's terms: her right foot bare, her left still in its flip-flop.

Pick one name per item and keep it. If you call them flip-flops, never call them sandals. And ask only for the changes you want. If you did not ask for the shoes to come off, say they stay on.

Why give the outfit and the room their own pictures?

A picture of your actor in the new outfit gives the clip something to copy. Make it from the actor's own photos, so the face stays theirs, and check it before you film. In one of our tests, the actor wore a dress described only in words. Filmed with a picture of her in that dress, she kept it for the whole take. We have not measured how much the picture adds on its own.

Only make one when the clothes really change. If the actor wears their usual clothes, their usual photos already do the job.

The room is the same. In our clip it changed between cuts. A picture of the empty room, sent with the scene, gives every angle the same room to start from.

How do you check a finished clip?

Watch it more than once.

• Count the items in every shot. A second shoe or a missing bottle is easy to miss at full speed.
• Watch the feet and hands. In our clip, both mistakes happened at her feet.
• Compare each side of a cut. Frames either side of a cut should agree on the room, the clothes and the props.
• Check the face against the actor's photos. Side by side, not from memory.

If one scene fails, redo that scene, not the whole video.

How does PersonaFix handle props and outfits?

PersonaFix builds these habits into the storyboard. You cast your actor and your product, and say what the video is for. The chat plans the scenes, and its reply names what happens to each item: worn, taken off, held or set down. If it adds a change you did not ask for, the reply says so, and you can say no in the chat.

When a scene needs pictures first, the chat adds one priced card for the whole board, called Pictures before filming. It can hold a look of your actor in the scene's outfit, a picture of an item, and a background when the scene needs one. Nothing is made until you press it. If you start filming without it, the scenes go ahead without those pictures. By default, a new look waits beside your actor's usual photos until you check it.

You can also turn on a paid check in Settings. After each clip is filmed, it looks for doubled or vanishing items and jumps. If it finds one, it shows the frames in doubt and offers a priced redo. It never films again by itself.

You still watch each scene. The steps are in how to put your product in an AI video, and the basics are in how to make an AI video.