--- title: "How to keep product details in AI images" canonical: https://withicarus.com/blog/keep-product-details-in-ai-images site: Icarus description: "Why AI tools change your logo, label and stitching, and a short test with your hardest product to check that the real thing shows up in every shot." --- # How to keep product details in AI images Why AI tools change your logo, label and stitching, and a short test with your hardest product to check that the real thing shows up in every shot. _October 7, 2026. Written with AI from the sources listed below, and checked automatically before it went live._ ![A seller at a wooden table holds a printed photo next to a leather tote bag to compare stitching and colour](https://icarus.b-cdn.net/blog/keep-product-details-in-ai-images/comparing-tote-bag-with-photo.png) _Made with Icarus_ AI changes your logo and label when it draws your product from a description instead of working from a photo of it. To keep product details in AI images, start every shot from one clear photo of the real product, then check each result against the real thing before you publish. This post explains why the details drift and why a close match can still cost you a listing. Then it walks you through a short test you can run with your hardest product, plus a checklist for the fine print. ## Why AI changes your logo and label An image model doesn't copy your product. It paints a new picture to match the words you gave it. To the model, your logo is just more shapes to paint. Text is where you see it first. OpenAI's own guide says: ["Although significantly improved, the model can still struggle with precise text placement and clarity."](https://developers.openai.com/api/docs/guides/image-generation) The same guide warns that [recurring brand elements may not stay consistent](https://developers.openai.com/api/docs/guides/image-generation) from one image to the next. It also says the model [may have trouble placing elements exactly](https://developers.openai.com/api/docs/guides/image-generation) in compositions where layout matters. So say you type "a navy tote with our gold logo on the front." The tool makes up a navy tote and makes up a gold logo. It might come close. It won't be yours. These are the details that usually drift: - The letters on a label, especially small print - The shape and spacing of a logo - Stitch lines, seams, zips and hardware - The exact shade of a colour - How a pattern repeats across the fabric One bad shot is easy to spot. The real trouble is a set where the logo is right in most frames and a little off in a few. ## Why close enough can cost you a listing Buyers check the details that matter to them: the label, the colour, the hardware. If the photo shows a different bottle from the one that arrives, you've made a return and lost a customer's trust. Shopping platforms care about this too. Google Merchant Center tells sellers to make sure the photo [reflects the exact color, pattern, finish, and material](https://support.google.com/merchants/answer/6324350) of the product being sold. It also bars [logos or store names superimposed in post-production](https://support.google.com/merchants/answer/6324350) unless they're a physical part of the product. So pasting your real logo over a garbled one isn't a clean fix there. There's a disclosure side as well. Google says images made with generative AI [must carry metadata saying they were AI-generated](https://support.google.com/merchants/answer/6324350). If you're not sure what each place you sell asks for, the [AI label checker](https://withicarus.com/tools/ai-image-disclosure) gives a plain answer for each one. No tool can promise a platform will accept your image. You still check the result and the platform's current rules yourself. ## Start from a photo, not a description The fix is to stop asking the AI to imagine your product. Give it the product. ![A ceramic jar on a white sheet by a window being photographed with a phone on a small tripod](https://icarus.b-cdn.net/blog/keep-product-details-in-ai-images/photographing-jar-by-window.png) _Made with Icarus_ In Icarus this is called [product pinning](https://withicarus.com/features/exact-product). You upload one clear photo of your product and save it. From then on you tag it in any shot and describe the scene in plain words. Icarus keeps your real colour, logo and stitching in every image. Use one photo from one angle. Several angles of the same product reduce consistency, so pick the view that shows the details buyers check. A good source photo makes everything after it easier: - Plain background, even light, no harsh shadows - The label or logo facing the camera - Sharp focus, with no motion blur - The whole product in frame, nothing cropped - No hands, props or other products touching it Save the product once and you don't have to upload it again for the next shoot. The same saved product works in a studio scene, a lifestyle shot, an on-model photo or a short video. ## Take the test with your hardest product Long checklists and scorecards are fine for big teams. You can learn most of what you need in a few minutes with one product. ![A tan leather wallet with cream contrast stitching and brass corners on a plain white background](https://icarus.b-cdn.net/blog/keep-product-details-in-ai-images/leather-wallet-contrast-stitching.png) _Made with Icarus_ 1. **Pick your hardest product.** Choose the one with the smallest print, the busiest pattern, contrast stitching or a detailed logo. If a tool gets that one right, the easy ones will be fine. 2. **Take one clear photo.** One angle, label facing the camera, plain background. 3. **Save it as a product in Icarus.** Then tag it in a handful of very different scenes: on a bathroom shelf, held in a hand, on an outdoor café table, in a tight close-up. 4. **Make the set.** Keep the default size so the test stays cheap. 5. **Put each result next to the real product.** Look at full size. Read the label letter by letter. Count the stitches on one seam. Hold the colour up in daylight. 6. **Mark each detail pass or fail.** If the logo changes in even one frame, you've found out what you needed to know. The test costs very little. An image at 1K costs $0.03 when it holds a saved product, so ten shots at $0.03 come to $0.30. Every price is on the [pricing page](https://withicarus.com/pricing), and you see what a shot costs before you make it. Run the same test with the same photo on any tool you're considering, so you compare like with like. The [guide to choosing an AI product photography tool](https://withicarus.com/guides/ai-product-photography-tool) has more checks to run after this one. ## Check the fine print before you publish Product pinning keeps your real colour and logo. Tiny text still deserves a human look, the same as you'd give the proofs from any photo shoot. Ingredient lists, sizes, batch codes and small legal lines are where you need to look closest. Run through this before a shot goes live: - [ ] Logo shape, spacing and colour match the real product - [ ] Every word on the label reads correctly at full size - [ ] Stitching, seams and hardware sit where they do on the real piece - [ ] Colour matches the product in daylight, not just on your screen - [ ] Pattern lines up and repeats the way it does on the fabric - [ ] Nothing was added that isn't on the product - [ ] The file carries any AI label the platform asks for If one frame is off, don't patch it by pasting a logo on top. Make the shot again from your saved product, or describe the change you want in plain words and let Icarus edit it. An image edit costs $0.05. Then run the checklist again. ## Questions ### Why does AI misspell the text on my product? The model draws letters as shapes instead of copying them from your label. OpenAI's guide says its model [can still struggle with precise text placement and clarity](https://developers.openai.com/api/docs/guides/image-generation). Starting from a photo of the real product, then reading the result at full size, is the safest way to work. ### How many product photos should I upload to keep details right? In Icarus, one clear photo from one angle. Several angles of the same product reduce consistency, so choose the view that shows the label and logo best. ### Can I just add my real logo on top of an AI image? That can break platform rules. Google Merchant Center bars [logos superimposed in post-production](https://support.google.com/merchants/answer/6324350) unless they're physically part of the product. It's better to make the shot again from a photo of the real product. ### Do AI product images need a label? It depends on where they run. Google Merchant Center says AI-generated images [must contain metadata saying so](https://support.google.com/merchants/answer/6324350). Check each platform's current rules before you publish. ## Sources - [Google Merchant Center Help: Image requirements](https://support.google.com/merchants/answer/6324350) - [OpenAI image generation guide](https://developers.openai.com/api/docs/guides/image-generation) - [Gemini API image generation](https://ai.google.dev/gemini-api/docs/image-generation) - [How AI-generated images in ads are identified and labelled on Meta](https://www.meta.com/en-gb/help/artificial-intelligence/355108217670024) --- Site: Icarus (https://withicarus.com) Canonical: https://withicarus.com/blog/keep-product-details-in-ai-images Contact: tanmay@onai.studio