Missing sections of 360 imagery can be filled by generative AI to improve the viewer experience. This example makes use of Google’s Gemini 2.5 Flash Image (Nano Banana) to generate imagery not captured in a 360 panorama.

Original Image

The previous article produced an equirectagular 360 image from a video:

The black borders are uncaptured imagery. They are not the intended area of interest, but you may find these areas distracting in the 360 viewer.

(Rotate horizontally or vertically to experience the original 360 scene.)

Can GenAI restore the missing imagery?

GenAI Prompt

Yes! Gemini’s Nano Banana can work with equirectangular imagery. Provide the previous image and following prompt to complete the full sphere.

Fill in the black border at the top and bottom of this equirectangular image. Do not change the dimensions of the equirectangular image. The scene should seamlessly extend into the black border. This should complete the scene without introducing anything of interest in the generated part of the scene. None of the pixels at the top or bottom should remain black.

With GenAI Completion

This works but failed to retain the original 2048x1024 image dimensions. No instruction in the prompt resulted in anything other than 1472x704 output. I’ve resized and resampled below to ensure the 2:1 ratio of an equirectangular image. The image quality was degraded before this alteration. Regardless, a-frame renders it the same.

Completing the scene with AI allows a more immersive experience. The added imagery is not a focal point and draws less attention than the missing imagery.

(Rotate horizontally or vertically to experience the completed 360 scene.)

The results are satisfactory for output from a free service tier. The image quality and dimension shortcomings were disappointing. Despite these flaws, it does improve the viewer experience.