Why AI videos get hands and text wrong

Updated October 21, 2026 · Reviewed by The Rift team

Short answer

AI video learns patterns from lots of footage, and hands and writing are hard patterns: many fingers, tiny shapes, exact letters that must stay still. So a model may add a finger or turn a sign into nearly-letters. You can avoid most of it: keep hands out of focus, avoid signs, and ask for another version. In Rift, Spark, Rift's video agent, can plan the clip so these parts stay out of the frame.

Hands and text are where a clip most often gives itself away. The good news is you can usually steer around them.

What you need

A scene idea where hands and signs aren’t the point.
Willingness to ask for a second version.

Step by step

Tell Spark the scene and mention what to keep out of frame.
Check the Draft for any shot that shows hands or signs up close.
Tap Make.
Look at hands and any writing in the result.
Ask for another version if something looks off.

Lines to try with Spark

“Me walking down a street, hands in my pockets.”
“My dog on a beach, no signs or text in view.”
“A candle on a table, wide shot.”
“A car driving by at dusk, plain background.”

Tips

Hide hands in pockets, behind objects or out of frame.
Avoid shop signs and labels in the background.
Use a wide shot, where small details matter less.

Questions people ask

Why are fingers so hard?

They're small, they overlap and they move in many ways, so the shape is easy to get slightly wrong.

Why is writing wrong?

Letters have to be exact and still, while the video is changing every frame.

Can I add text over the video myself?

Text over video isn't in the app yet. Add it in your posting app.

Will this get fixed?

It gets better over time, but it's wise to plan around it for now.