One gpt-image-2 call, 9 hairstyle variants: prompt engineering for grid layouts

A hairstyle preview tool moved from eight gpt-image-2 calls to one prompt that generates a labeled 3x3 grid.

The new single-call approach kept the same face across all variants, avoiding identity drift in the previews.

It cut preview cost from $0.32 to $0.04 and reduced wait time from about 40 seconds to about 6 seconds.

The result shows how prompt engineering can replace repeated generations with a faster, cheaper multi-variant workflow.