Nova Canvas IMAGE_VARIATION similarityStrength vs Sume n and reference

IMAGE_VARIATION is the Nova Canvas task for making new versions of an existing picture. Amazon's docs (read 2026-10-06) say it accepts 1 to 5 input images and a similarityStrength between 0.2 and 1.0. Sume has no equivalent number, so you control closeness with wording and with the model you pick.

Mapping the pieces

How to steer closeness on Sume

Say what must stay. A prompt such as "same product, same angle, new background" asks for a close variation; "same product, new angle and lighting" asks for a loose one. The earlier post on image-to-image strength on Sume goes through this.

Use n to get several takes at once. Each model has its own cap, listed as the n range in GET /v1/images/models. The post on n limits and cost per model has the table.

Four variations in one call

The request below asks for four takes and prints each URL plus the billed total.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={
        "model": "ideogram/ideogram-v4.5",
        "n": 4,
        "quality": "low",
        "prompt": "Same bottle, same angle, four different background colours.",
        "input_references": [{"type": "image_url", "image_url": {"url": "https://example.com/bottle.jpg"}}],
    },
    timeout=90,
)
body = r.json()
for item in body.get("data", []):
    print(item["url"])
print("billed:", body.get("usage", {}).get("cost"))

At the list prices in the Sume docs, an Ideogram 4.5 low image is $0.03 list and $0.0375 billed, so four takes cost about $0.15.