My mother-in-law is ruthless about packaging. If a box does not tell her how to wash the blanket inside it, she puts it back on the shelf and buys the one that does. She is not unusual. For textiles, the care label is often the deciding factor at the point of sale, and it is almost always the ugliest thing on the packaging.
That makes the care label a good test case for AI image work. You need a flat, legible panel in a fixed grid, with tiny symbols that must stay recognisable, sitting inside a larger packaging photo or render. It is a narrow, unglamorous problem, and the tools handle it better than you would expect if you set the job up correctly. I did this recently in an AI image conversion tool, starting from a product shot that already existed, and the workflow is worth writing down.
Decide what you are actually producing
There are three different deliverables here and they do not share a workflow:
A packaging render where the care panel appears at an angle, partially visible, as part of the box. A flat label that will be printed and sewn in. A reference image for a designer to trace in a vector tool.
The conversion approach works for the first and third. For the second, you will fight the model forever, because printed labels need exact symbol geometry and no AI model will reliably give you that. Be honest about which one you need before you start.
Start from a photo, not from nothing
Converting an existing image is much more controllable than generating one from a blank prompt. Photograph or render the box flat-on, fill the frame, and keep the light even. Then use the conversion to change the treatment rather than to invent the object — push it toward a clean product render, a soft studio look, or a printed-catalogue aesthetic.
Style strength is the setting that matters most. Turn it up and the panel geometry starts to warp, the box corners soften, and the symbols drift. Keep it moderate and the model restyles the surface while holding the shape. If your tool exposes style versus content weight separately, lean toward content and let the style be a light touch.
Get the symbol grid right in the source
Care symbols — the washtub, the triangle, the iron, the circle — are simple enough that a model can reproduce their general impression but rarely their exact form. So do not ask it to. Build a plain grid of placeholder shapes in the source image at the correct size and spacing, convert the whole thing to get the panel's look, then drop the real ISO symbols back in on top.
This sounds like cheating. It is the only reliable method. Real symbols, real geometry, model-supplied surface and lighting around them.
Keep the type small and let it be small
Tiny text is where AI images fall apart. Asking for "care instructions in small print" tends to produce convincing-looking grey lines that are not words, which is fine at thumbnail size and embarrassing at full resolution. Two options: keep the text genuinely unreadable and treat it as texture, or reserve a clean strip and set the type yourself afterwards.
If you must render text in the image, keep it to a handful of large words — fibre content, a temperature, a country of origin. Anything longer will not survive.
Colour accuracy is the hidden trap
If the care label sits on a coloured box, conversion will shift that colour slightly. On a neutral kraft or white carton the drift is invisible. On a deep navy or a saturated red it is obvious the moment the converted image sits next to the original photo on the same page.
Fix this by comparing swatches rather than trusting your eye in isolation. Sample the box colour in both images and nudge the converted version back toward the original. If the tool has a colour-preservation or palette-lock option, use it; that exists precisely for this problem.
Match the material to the product
A care panel on a linen throw should look printed on textured uncoated stock with slightly soft edges. The same panel on a sportswear box should look sharp on a gloss laminate. Describe the stock, not the ink: "matte uncoated board, fine fibre texture, slightly absorbent print edges" versus "gloss laminate, crisp print, faint specular highlight across the panel".
This one sentence does more for realism than any amount of lighting description, because material is what the eye uses to judge whether a package photo is real.
A short checklist before you call it done
Box corners still square. Panel edges straight, not bowed. Symbols in a consistent baseline grid. Box colour within a shade of the original. Contact shadow present. No invented words anywhere a shopper will read them.
If all six hold, the conversion has done its job. Anything less and you are better off converting again from a cleaner source than trying to patch the output. You can run the whole pass in the converter, and honestly, once the reference photo is right the rest takes about ten minutes — the preparation is the work, and the label is just where it shows.