Two image variants may both render numbers at large size but differ when arranging several independent values on a grid. Test exact order and readability separately from overall layout. Oakgen lists Nano Banana 2 and Nano Banana Pro in its public model directory, and variants may expose different controls.
Build the test input
Prepare a fictional tournament graphic with a two-by-three grid and the values 09, 12, 08, 13, 16, and 20 in that order. Use a simple header and obvious spacing rules. Keep the expected sequence in a copy file. Ask neither model to invent other number labels.
Run the proposed comparison
Use the same output shape and exact text brief for both operations. If reference upload is used, give both the same unbranded empty grid. Review the originals, then reduce previews to the dimensions of the intended card. Save stage-by-stage readings rather than judging the six items in one glance.
Decide what passes
Require the full sequence plus the correct first-item leading zero and no unexpected repeats. Note a spelling failure even if the mere idea of a numbered grid looks competent. Distinguish errors that appear only at small scale from genuine model substitutions visible at large scale.
Act on the outcome
A result with precisely positioned cards can be useful if the editorial team sets the numbers as live text in a layout editor. Choose that approach when several values are updated together. For a one-time illustration with the full sequence correct, compare readability after display resizing before approving it.

