A grid turns a pile of photos into a checklist with pictures in it. Define four views once, front, back, label and the suspected defect, and every item produces the same 2×2 board. A model asked to compare a batch of these boards across a day of arrivals is doing something quite different from a model asked to describe a loose collection of photographs.

The comparison case is the one that matters. When the question is whether this scratch is worse than the one accepted yesterday, the two boards have to be built the same way or the comparison is meaningless. Fixing the layout is what makes successive inspections comparable, and it costs nothing. The batching logic behind a series of boards is described in batch images for AI.

Choosing the cells: front and back for the general state, a close-up of the label or the reference for identification, and a close-up of the defect or the wear point. Four cells is usually enough. If the item is small or the defect is fine, spend the fourth cell on a tighter crop at higher magnification rather than adding a fifth view.

Resolution is where inspection grids differ from social grids. A defect visible at 2048px can disappear at 1024px, so export at 3000px or more when the whole judgement rests on a scratch, a seam or a print quality. Keep a clear gap between cells so a defect near an edge is not read as belonging to the neighbouring view.

Keep the discipline in the file names too, not only in the image: one board per item, named after the reference, so a batch stays readable after a hundred exports. The free plan allows 2 exports per day, which is the real constraint on a large batch; the 24-hour Pass at $3 removes it for a day and is the cheaper option for a one-off inspection run.

Because everything is composited in the browser, photos of a customer return or an unreleased product are never uploaded while you build the board. The free plan includes 12 images per grid and PNG or JPG output up to 4000px, with no account and no watermark.