Why label legibility is a conversion problem, not just a quality problem
Packaged goods buyers do a specific thing that apparel buyers do not: they zoom in to read. Supplement buyers check the serving size. Food buyers check allergens. Skincare buyers check the INCI list. Parents check age grading. When the panel is unreadable, the shopper does not assume the camera was soft — they assume you are hiding something, and they leave to go read the label on a competitor's listing.
That behavior shows up as an unusually high rate of image-gallery engagement followed by exit. If your analytics show shoppers opening image three or four and bouncing, and image three or four is your back panel, the panel is the problem.
Test your own gallery at the size customers actually see it. Shrink the main image to 200px wide. If the brand name and the primary claim are not readable, the front-of-pack shot needs to be reshot tighter — not sharpened in post.
Camera settings that keep type sharp edge to edge
Most unreadable label shots are not focus misses. They are depth-of-field misses on a curved surface. A cylindrical bottle puts the left and right edges of a wraparound label several centimeters behind the center — well outside the depth of field of a fast lens shot wide open.
| Setting | Recommendation | Why |
|---|---|---|
| Aperture | f/8–f/13 | Covers label curvature; stops before visible diffraction softening on most sensors |
| ISO | Base (64–200) | Noise reduction is what eats small type first |
| Focus point | One third into the curve | Splits the depth budget across the readable arc |
| Lens | 85–105mm macro | Flat field, minimal barrel distortion bending straight label edges |
| Capture | Tethered RAW | Lets you check type at 100% before the product leaves the set |
When one aperture cannot cover the whole panel — common on wide jars and squeeze tubes — focus stack. Five to nine frames stepped across the label, merged, will beat any amount of post-hoc sharpening. Sharpening does not add letterform information that the sensor never recorded; it just adds contrast to the mush.
Aggressive noise reduction and heavy JPEG compression are the two silent killers of small print. Both smear the low-contrast edges of six-point type while leaving large graphics looking fine, so the image passes a casual review and fails at zoom.
Lighting reflective, foiled, and shrink-wrapped packaging
Packaging materials are engineered to catch light on a shelf, which makes them hostile in a studio. Foil stamping, spot UV, metallized films, and gloss laminate all throw specular hits exactly where the brand name sits.
Three setups solve most of it:
- Large, close, and off-axis. A big softbox brought close and angled roughly 45 degrees off the panel gives an even wash without dropping a hotspot on center-front type. Distance is what creates hotspots — a small source far away is a specular point source.
- Polarize the pair. A polarizing filter on the lens plus polarizing gel on the key light kills glare on laminated cartons and shrink wrap. Expect to lose one and a half to two stops.
- Light the box, then relight the panel. Shoot the hero lighting pass, then a second frame lit specifically so the ingredient panel is clean, and composite. This is standard practice, and it is honest — you are not changing what the label says.
For shrink-wrapped multipacks, a strip of white card just outside frame, angled to fill the wrap, removes the crinkle glare that otherwise scatters across the whole front face.
Where AI helps on packaging shots — and where it must not touch
The useful split is between context and content. AI is genuinely strong at everything around the product and genuinely unreliable at the words printed on it.
Safe for AI
- Background replacement and seamless white cleanup
- Building lifestyle scenes around a photographed pack
- Shadow and reflection generation under the product
- Dust, fingerprint, and scuff removal on blank areas
- Straightening, perspective correction, and consistent framing across a catalog
Never generate
- Ingredient panels, nutrition facts, dosage tables
- Barcodes, lot codes, expiry fields
- Regulatory marks, certification logos, warning icons
- Brand wordmarks and any typography at all
- Net weight, volume, or count claims
The failure is subtle, which is what makes it dangerous. A model asked to clean up a supplement bottle will often re-draw the panel into something that looks like a nutrition table — correct grid, plausible typeface, invented numbers. Nobody notices at 400px. A regulator or a competitor filing a complaint notices at 100%.
Tools built for e-commerce catalogs, Retouchable included, are designed around preserving the photographed product while changing the scene around it — which is exactly the boundary packaging work needs. Whatever tool you use, the acceptance test is the same: the label pixels in the output must match the label pixels you shot.
Marketplace compliance for packaged goods imagery
Marketplaces treat images as product claims. If the image says one thing and the listing detail page says another — different net weight, different count, a certification mark that is not on the real carton — that is a discrepancy, and discrepancies get listings suppressed rather than politely queried.
| Requirement | What it means for label shots |
|---|---|
| Image must match the product sold | No regenerated or edited on-pack text, ever |
| Main image on pure white | Front of pack, no added badges or burst graphics |
| Minimum long side 1600px (zoom) | Shoot larger; zoom is how panels get read |
| No promotional overlays on main | Save claims for secondary and A+ images |
| Fills roughly 85% of the frame | Tighter crop directly improves thumbnail legibility |
Two regulatory dates are worth having on your calendar: the EU AI Act transparency obligations and California's AI transparency law both take effect on 2 August 2026, and both push toward machine-readable marking of AI-generated content. If your workflow generates backgrounds or scenes, start recording which assets were AI-assisted now, at the asset level, rather than reconstructing it later from memory.
A QC checklist that catches label problems before publish
Run every packaging image through this before it goes live. It takes about ninety seconds per SKU and it is the difference between a clean catalog and a suppression email.
- Read the panel out loud at 100%. If you cannot read a word, the customer cannot either.
- Compare against the physical pack. Word for word on claims, net weight, and count. Keep a reference pack at the desk.
- Scan the barcode from the screen. If a phone scanner reads it and returns the right GTIN, the barcode survived the edit.
- Check every logo and certification mark. Organic, recycling, allergen, and safety marks are the ones most often subtly deformed.
- Verify colors against the pack under neutral light. Brand colors on packaging are contractual for many licensed products.
- Shrink to thumbnail. Brand name and primary claim must survive.
- Log the provenance. Which images were AI-assisted, and for what — background, shadow, cleanup.
Photograph the words, generate the world — and never the reverse.