Merging images for identification has one condition that comparison does not need: the objects must not touch each other visually. If two leaves overlap across a cell boundary, the answer blurs with them. A gap between cells is not cosmetic here, it is what keeps each answer attached to the right object.

The layout that works best is the one that matches how you will read the answer. Six items in a 2×3 grid, read left to right and top to bottom, come back as six lines you can number the same way in the prompt. Four items in a 2×2 do the same. The Strip of 3 preset is useful when the three objects are meant to be compared rather than identified one by one. For the wider picture of how layout steers an assistant, see combine images for AI.

Photo quality matters as much as the grid. For identification, one object per cell, centred and filling most of it, with the background as plain as you can make it. A wide shot with the object in a corner spends most of its pixels on context the model does not need. If you cannot crop, at least keep the object whole: an object cut by the cell edge is a common reason for a wrong answer.

Short captions are optional and often useful. A small label with a location, a date or a variant name gives the model a piece of context it cannot read from the image, and it gives you a way to check afterwards that the answers line up with the cells you think they do. Keep the labels short so they do not compete with the object itself.

Expect identification answers to be tentative, and treat them that way. A model identifies from what it can see, so a partial view, poor light or a blurred detail can produce a confident answer that is wrong. The useful pattern is to merge the images that show the identifying feature, ask once, then verify the answer against a reference before acting on it.

MergeFrame assembles the grid in your browser, so nothing is uploaded while you prepare it. The free plan gives you 12 images per grid and 2 exports per day, in PNG or JPG up to 4000px, with no account and no watermark. If you are working through a long list of items, the 24-hour Pass at $3 removes the daily export limit for 24 hours and adds PDF and WebP output.