Product Updates··3 min read

Pixel-Perfect Sizes for Marketplaces

Set exact width and height per instruction in flat-to-model so outputs match Amazon, Shopify, Zalando, and ASOS specs without post-crop.

By On-Model Team

On-Model AI fashion model framed by overlapping precision rectangles representing different marketplace image dimensions

Every fashion retailer has its own image guidelines, and most of them won't accept anything close. Shopify product listings run at 2048×2048. Amazon Fashion main images at 2000×2000. Zalando catalog masters sit around 1500×2000. Luxury platforms like Farfetch lean tall portrait at 2:3 or 3:4.

Common sizes for on-model fashion imagery (always check each retailer's seller portal — exact requirements vary):

RetailerCommon sizeAspect ratio
Shopify (DTC product)2048×20481:1
Amazon Fashion (main)2000×20001:1
Zalando (catalog master)1500×20003:4
ASOS (PDP master)1500×1900~4:5
Farfetch (PDP)1000×15002:3

None of these line up cleanly with the existing 1K / 2K / 4K presets crossed with the standard aspect-ratio list. So until now you'd generate the closest preset and crop manually to spec. Awkward, and never quite the master file you actually wanted.

What's new

Flat-to-model now accepts an optional width × height per instruction. When set, they override size and ar, and the output is delivered at exactly those pixel dimensions, no post-crop. Available on Pro 1 and above.

In the app, each instruction's Output section has a new Dimensions toggle: Preset (the existing Output Size + Aspect Ratio dropdowns) or Custom (two number inputs, Width and Height). Pick the spec you need and submit.

In the API, it's a pair of optional fields inside options:

POST /flat-2-model
{
"identity_code": "default-pro-...",
"project_id": "proj_...",
"images": ["sweater-uuid", "chinos-uuid"],
"instructions": [
  {
    "prompt": "Editorial fashion catalog shot, neutral studio backdrop",
    "pose": "standing three-quarter",
    "options": {
      "width": 2000,
      "height": 2800,
      "format": "jpg"
    }
  }
]
}

The two input assets

We started with a pair of standard e-commerce flat-lays — a folded navy crew-neck sweater and a hung olive chino — both shot against a plain light backdrop, the sort of master imagery any catalog already has on hand:

Flat-lay of a folded navy crew-neck cashmere sweater on a light grey backdrop
Input 1 — Navy sweater (flat-lay)
Flat-lay of a pair of olive green tailored chinos on a light grey backdrop
Input 2 — Olive chinos (flat-lay)

One scene, three sizes

The same flat-lays, paired with the same model and the same preset, rendered at three different specs by changing only width and height (click any image to zoom):

On-Model — custom dimensions
2000×2800 — 5:7 portrait master
2048×2048 — 1:1 (Shopify, Amazon)
2704×2024 — 4:3 landscape hero

Three native masters, one preset's worth of clicks. No follow-up crop, no resampling: the file you download is the file you upload to the marketplace.

Constraints worth knowing

  • Range: 256-7000 pixels per side, in multiples of 8.
  • Aspect ratio: between 1:4 and 4:1. Outside that band, output quality starts to degrade.
  • Credits: billed at the implicit tier — max(W, H) ≤ 1024 charges as 1K, ≤ 2048 as 2K, anything larger as 4K. So 2000×2800 is a 2K-tier job; 3000×4500 is 4K-tier.
  • Tier: Pro 1 and above (see pricing). Free and Basic users keep the preset flow.

The frontend validates each input live as you type (whole numbers, in range, multiples of 8, ratio inside the bounds). The same rules are enforced server-side.

Free and Basic users can still author presets that include custom width and height — the dimensions just stay dormant until the account upgrades. No re-edit later.

flat-to-modelmarketplacesprocustom-dimensionsamazonshopifyzalandoasosfarfetchworkflow