This article solves one problem: you have a single phone snapshot of your product, and you need a listing-ready main image out of it.
Pick the right tool first
| Tool | What it does | Images per run |
|---|---|---|
| White-Background Product Image | Strips the original background and drops the product onto clean white, light gray, or a soft contact shadow | 1 |
| Product Scene Image | Places the product into a home, outdoor, office, minimal-tabletop or holiday setting | 1 |
| Product Image Set | Runs pure-white hero, light-gray premium, lifestyle scene and detail close-up in parallel | 4 |
| Product Studio | Recognizes the product, builds a product profile, then generates a full main-image set to your chosen marketplace's spec | 5-7 |
Whichever you use, as long as your own photo is part of the input, a fidelity constraint is injected into every single request: the product's shape, color and packaging text must be preserved, and only the background, scene and model may change. In rare cases the model still drifts — just re-run it, and any job that fails releases the credits that were held. The long version is in How product fidelity works.
What kind of upload produces an accurate result
- Sharp, focused on the product. Reference images are compressed to a 1024px long edge before they reach the model, so a 40 MB original buys you nothing extra. What actually decides the outcome is how large the product sits in frame and how clean its edges are.
- One subject per photo. Line up three shade variants in a single shot and the model can't tell which one you want preserved — you'll get a hybrid of all of them.
- Avoid heavy occlusion. If a hand, a prop or a price tag covers half the packaging, the model has to guess at the hidden part. Packaging text goes wrong here more often than anywhere else.
- Shoot packaging text square-on. Text photographed at a steep angle tends to smear when it gets repainted.
Getting all 11 marketplace presets
Choose your marketplace in Product Studio and the image count, ordering, aspect ratio and hero-image rules are enforced server-side rather than left to the model's discretion:
| Marketplace | Images | Ratio | Min. edge |
|---|---|---|---|
| Taobao / Tmall | 5 | 1:1 | 800px |
| JD.com | 6 | 1:1 | 800px |
| Pinduoduo | 6 | 1:1 | 800px |
| Douyin E-commerce | 5 | 1:1 | 800px |
| Xiaohongshu | 6 | 3:4 | 800px |
| 1688 | 5 | 1:1 | 800px |
| Amazon | 7 | 1:1 | 1000px |
| AliExpress | 6 | 1:1 | 800px |
| Temu | 6 | 1:1 | 800px |
| Shopee | 6 | 1:1 | 800px |
| Lazada | 6 | 1:1 | 800px |
Every marketplace has its own hero-image rule, and the system prepends it to the hero prompt. Amazon, for instance, wants a pure white background, the product filling at least 85% of the frame, and no text, watermarks or props at all; Pinduoduo comes down hard on overlaid text. Text on feature and spec images follows the destination — English for overseas marketplaces, Chinese for domestic ones.
Bulk output, pricing and refunds
There are several ready-made routes for volume: Product Image Set fires four images in parallel, Pose Split returns four different angles in one run, Batch Background Removal accepts multiple uploads at once, and Product Studio produces a whole set once you pick a marketplace.
Billing runs on credits — CN¥1 = 10 credits, images from 2 credits each, and every new account gets 30 credits on signup, enough to try a few shots before deciding whether to top up. Full rules are in How credits are calculated and on the pricing page.
Jobs run in the background, so closing the tab won't lose them. Finished images land in My Works and can be saved back into Assets for reuse. A failed generation automatically releases the credits that were held, with the reason shown on the job. One caveat: an image that rendered fine but that you simply don't like — bad angle, busy background — is not a failure. In that case swapping the template or the source photo and re-running is the faster fix.