The question here is: "I uploaded the photo and the AI reads it — do I really need to type anything?" You do, but only once. Below is what recognition can see, what it can't, and where the few lines you add end up getting used.
What photo recognition sees, and what it misses
From a single photo, recognition can read shape, color, rough category, packaging structure and the text printed on the packaging. What it usually can't read falls into these buckets:
- material and composition — faux leather and real leather, acrylic and glass often look identical;
- exact specs — capacity, net weight, dimensions, the size range you actually stock;
- performance figures — wattage, battery life, water resistance rating, device compatibility;
- certifications and compliance information;
- your selling points and target audience, which are business judgments and simply aren't in the picture.
Beyond recognition, Product Studio also searches public information about comparable products to fill in specs — but "comparable" is not "yours." The system is honest about this: every field carries a source tag — AI-inferred, web-verified or confirmed — and purely inferred values are marked down in confidence. The merge rules explicitly forbid inventing brand names, certifications or patents, and prefer leaving a field blank over guessing. So when you see AI-inferred, give it a glance; the blanks are yours to fill.
Fill it in and the whole downstream chain benefits
A product profile isn't a static record — it gets injected into the generations that follow:
- Image and video runs push the product description into the prompt, adding one more hard constraint for the model. Concretely: write only "face cream" and the AI might hand you a metal pump bottle; write "50ml frosted glass jar, wooden screw cap" and what comes out is much closer to the jar on your shelf.
- Detail pages are generated from the real specs in the profile, so spec graphics don't carry invented data.
- Selling points and audience are captured into the merchant knowledge base, then retrieved and reused later when writing copy or presenter scripts.
- Drama ads, seeding videos, video remakes, canvas workflows and launch campaigns all have an Import from product library option, so name, selling points and product photo are filled in once and reused everywhere.
Fill it once, reuse it everywhere, and re-scans won't overwrite you
This is the part most people underestimate: any field you edited by hand is marked confirmed and is preserved when you later hit re-scan, with the newly recognized value for that field yielding to yours. So you never get the "I fixed the net weight and the AI changed it back" experience.
Put differently, building the profile is a one-time cost: one set of photos plus one set of specs go into the product library, and after that every tool pulls them in — no describing the product again.
The least-effort way to actually do it
- In Product Studio, shoot 1-4 sample photos and let the system run a recognition pass (15 credits by default; the figure shown in the tool is what governs). You get a draft profile back.
- Edit only two kinds of things: the key specs tagged
AI-inferred, and the fields the AI left empty — material, net content and compatibility range go missing most often. - Write 3-5 selling points and one line on the audience. Write them as facts, not ad copy; the copywriting tools will use them as raw material later.
- Save. From then on, every tool imports from the product library.
The division of labor in one sentence: the AI's job is to extract everything visible in the photo, and yours is to supply what isn't. Do this part properly and the accuracy of every asset downstream moves up with it. For what the system enforces to keep your product unchanged, see Product fidelity.