Portrait Editing Example: Original and Two Recorded Outputs

This is a visual example, not a controlled benchmark. The source image and prompt are available, but the exact model versions, seeds, sampling settings, and run logs were not retained.

Available

Downloadable source image

Available

Recorded editing prompt

Not retained

Exact model IDs and run parameters

What can be inspected

Input: download the original image.
Recorded prompt: “the Mixue Bingcheng mascot riding on a motorcycle”.
Outputs: the two image files preserved in this repository and shown below.

What is not known

The original record does not establish the exact API model IDs, dates, seeds, steps, guidance values, masks, output settings, or whether more than one result was generated before these examples were selected. Those details are therefore not reconstructed or guessed.

Visual comparison

The output labels below reproduce the project filenames; they are not a verified claim about a current model version.

Output stored under the Gemini project label — Output A — stored under the Gemini project label

Output stored under the Nano Banana project label — Output B — stored under the Nano Banana project label

How to interpret this page

Use it to inspect composition, text rendering, subject replacement, and obvious artifacts.
Do not use it to rank model quality, speed, cost, or consistency.
A future controlled benchmark should pin the model IDs, preserve raw responses, disclose every run, and publish the selection rule.