Handbag product photos: hardware close-up as a second reference

A handbag's value sits in small parts: the clasp, the zip pull, the stitching and the strap anchor. A generated lifestyle photo that gets the shape right and the hardware wrong sends customers to the returns desk. The remedy is cheap: give the model a second reference of the part.

Sume's Image API runs openai/gpt-image-2.5 with up to 16 input references, so a full-bag shot and a close-up travel in the same request.

Two references, two jobs

Image 1 gives silhouette, colour and proportions. Image 2 gives the hardware. Say so, and describe the scene you want around the bag, not the bag.

import os, requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
resp = requests.post("https://api.sume.com/v1/images", headers=H, timeout=90, json={
    "model": "openai/gpt-image-2.5",
    "prompt": "Image 1 is a leather tote bag. Image 2 is a close-up of its gold clasp and stitching. Photograph the same bag on a cafe table beside a coffee cup, soft window light. Keep the shape, colour, clasp and stitching exactly as in the references.",
    "input_references": [
        {"type": "image_url", "image_url": {"url": "https://example.com/tote-full.jpg"}},
        {"type": "image_url", "image_url": {"url": "https://example.com/tote-clasp.jpg"}}
    ],
    "aspect_ratio": "auto",
})
resp.raise_for_status()
if resp.status_code == 202:
    raise SystemExit("job envelope: poll data.status_url")
body = resp.json()
print([d["url"] for d in body["data"]], body.get("usage"))

A QC list for bags

Scale the scene, not the bag

To get five scenes from one pair of references, use the n parameter: Sume allows up to 10 per call, with lower ceilings on some models, so read the model's n range from its descriptor first. Log usage.cost per call. Keep the original photographs as the product page's anchor shots, and use the generated scenes for ads and social, where the bag is part of a mood and not the only evidence.

Before you run a catalog

Run one SKU end to end first. POST /v1/images returns the images directly when it finishes within 30 seconds; past that it returns a 202 envelope with status_url and result_url, which the jobs and results guide explains. Handle that branch before you loop over a catalog, and write each result's URL and usage.cost to a file keyed by SKU, so a failed run restarts where it stopped and nothing is paid for twice.