What stayed the same in all three edits?
We ran GPT Image 2.5 Flare in Mazee.ai Studio with one reference image, the same prompt, a square frame and the 2K setting. Each run returned a 2048 × 2048 PNG and cost 10 creation credits, for a total of 30. We kept every output in submission order.
- Label text. Each result read VARELIN, CEDAR / IRIS, EAU DE PARFUM and 50 mL, in the original order. Spelling, casing and the slash matched.
- Bottle shape. All three kept the front-facing view, the rectangular glass bottle, the cylindrical silver cap and approximately the same liquid level by visual inspection.
- Surface details. Highlights shifted on the cap and along the glass edges. Some letters gained a thin light edge and the paper looked smoother or rougher depending on the run.


See the three outputs and label close-ups
Run 1. Look at the thin light edge around the black lettering and the more pronounced paper grain. The new scene adds a strong reflection under the bottle. Open the 2048 × 2048 original.

Run 2. All four lines still read correctly. The paper looks smoother than in run 1 and the rainbow is concentrated to the right of the bottle. .





