Lamp product photos, lit and unlit, from one reference

A product page for a lamp needs two states. Off shows the shape in daylight. On shows the glow in an evening room. Shooting both takes a studio and a dark room; generating both takes one reference and two matched prompts.

With Sume's Image API the reference is the lamp, and the pair differs in one line of the prompt.

Two prompts, one difference

Write the scene once and change only the lighting sentence. Matching everything else makes it easier to line up the pair on the page.

import os, requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
SCENE = "Place this exact table lamp on a bedside table next to a book. Keep the shade, stem and base as in the reference."
STATES = {"off": "Daylight from a window, the lamp is switched off.",
          "on": "Evening, the lamp is switched on with a warm glow, the room is dim."}
for name, light in STATES.items():
    r = requests.post("https://api.sume.com/v1/images", headers=H, timeout=90, json={
        "model": "openai/gpt-image-2.5",
        "prompt": f"{SCENE} {light}",
        "input_references": [{"type": "image_url",
            "image_url": {"url": "https://example.com/lamp.jpg"}}],
        "aspect_ratio": "auto",
    })
    r.raise_for_status()
    print(name, r.json()["data"][0]["url"])

What to compare

• Shade shape, material and colour are the same in both images.
• The unlit lamp shows no glow; the lit lamp shows light through the shade, not a glowing base.
• Cord and switch are in the same place in both.
• Room lighting in the lit image looks plausible, with the shade as the brightest object.

Keep the real studio photo as the first image on the page. The generated pair shows the mood, and the original shows the product. Log usage.cost per call so the two states have a known price.

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.