Separate the image model from the video tool
Nano Banana is Google's AI image generator for creating and editing still images through Gemini. Use it to prepare a character reference before planning motion in a separate video workflow. Nano Banana is not selectable in MakeStory, whose current image generation uses GPT Image 2.
Nano Banana is not selectable in MakeStory. MakeStory's current image generation uses GPT Image 2. You can also use an image you own or have permission to use as the source in the clip workflow. Check the current upload requirements before continuing.
Design a reference for its next use
A teapot detective needs a silhouette that survives a wider shot. Give it a short curved spout, one blue handle and a small copper badge. Leave out a page of decorative detail. The reference should make these choices visible so you can compare later scenes against them.
Build and review the image
- Write a character brief with fixed appearance details and a separate list of scene actions.
- In your chosen image service, generate a clear view with simple light and an uncluttered background.
- Change one detail per revision. Check that an edit to the badge does not silently change the handle or body.
- Save the accepted image and brief. Use an approved source image in a supported clip workflow, then describe the motion separately.
A reference brief you can adapt
Character: small cream ceramic teapot detective. Fixed details: curved spout, one blue handle, copper badge. Pose: standing still, full silhouette visible. Light: soft window light on a plain tabletop. Exclude: lettering, crowds and moving background objects.
Keep a record of accepted details
Check the number of handles, badge position, spout shape and visible material. Zoom in, then view the image at phone size. A detail that disappears at normal viewing size cannot carry character recognition. Keep the accepted version instead of repeatedly regenerating an entire cast.
Will an image reference guarantee continuity in video? No. Motion can change shapes or introduce extra parts. Review the moving output against the still, especially during turns. The interface language also does not guarantee correct writing inside an image, so inspect any requested text yourself.