Getting good results from your product photos
Generated images aren’t invented from a text description. The AI works from your reference photos, and it can only draw what it can see. Good references produce images of your product. Thin references produce images of something that looks a lot like your product, and buyers notice the difference before you do.
What a good reference set looks like
- Several angles. Front, back, side, and a three-quarter view. One angle means everything else is a guess.
- Sharp and well lit. Daylight or bright even lighting, no motion blur. The model inherits whatever murkiness you feed it.
- True colors. Skip the moody filters. If the product is forest green, the photo should show forest green.
- The product alone. Clean, uncluttered shots give the AI a clear read on shape and proportion.
Close-ups of anything your copy mentions
This is the rule that separates good results from frustrating ones. If your listing talks about a zipper pocket, a charging port, a locking lid, or a stitched logo, include a close-up photo of that exact feature. When a mechanism isn’t visible in any reference, the AI guesses where it goes and what it looks like, and guessed hardware reads wrong instantly. We’ve watched a plush toy’s zipper migrate to three different body parts across generations because no photo ever showed it clearly.
Tip: Quick audit: read your bullets, and for every physical feature they name, check you have a photo where that feature is clearly visible. Missing one? Take it with your phone, it takes a minute.
File limits
- JPEG, PNG, or WebP.
- Big files are compressed automatically. If a photo is still over 8 MB after compression it’ll be rejected, which in practice only happens with enormous originals.
- Batches cap at 50 MB total, which is plenty for a full reference set.
What to avoid
- Watermarked or heavily branded stock-style shots. The watermark can leak into generations.
- Collages or images with text overlays. The AI reads them as part of the product.
- 3D renders that don’t match the shipped product. Renders are fine as references if they’re accurate, but mixing an idealized render with real photos gives the AI two conflicting products.
Upload once, reuse everywhere: the photos live in your project’s library and every image slot in modular A+ or seamless can draw from the same set, which also keeps your page visually consistent.