Midjourney V8.1 draft mode makes 24 images; Sume caps n at 10 per call

A Midjourney V8.1 draft-mode job makes 24 lower-resolution images and uses half the fast hours of a standard V8.1 SD job, per Midjourney's June 16, 2026 post. On Sume, one POST /v1/images call returns at most 10 images, and each model's own cap is often lower, so 24 images means 3 calls at a cap of 10 and 6 calls at a cap of 4. Midjourney is not in the Sume image catalog.

The two limits

The Midjourney figures are from its update post (read 2026-10-05) and are dated June 16, 2026. The Sume figures are from the Image API docs.

Calls needed for 24 images

Divide 24 by the model's n maximum and round up. The docs' own catalog example shows a model with n max of 4, so check the model you plan to use.

• Cost does not change with the split: the docs say you pay cost_usd x n, so 24 images cost 24 x cost_usd however many calls you make.
• Fewer, larger calls may fall back to a 202 job if they pass the 30-second wait, so read the status code.

Fan out in code

This sends 24 images as batches of 4, in parallel, and counts the images. Change CAP to the model's real maximum from the catalog. It uses asyncio.gather, so the whole set depends on the slowest call; check each response status as shown.

import asyncio, os, httpx

CAP, TOTAL = 4, 24

async def one(c, headers, n):
    body = {"model": "google/nano-banana-2", "prompt": "logo concept, geometric fox", "n": n}
    r = await c.post("https://api.sume.com/v1/images", headers=headers, json=body)
    return len(r.json()["data"]) if r.status_code == 200 else 0

async def main():
    headers = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
    sizes = [min(CAP, TOTAL - i) for i in range(0, TOTAL, CAP)]
    async with httpx.AsyncClient(timeout=90) as c:
        got = await asyncio.gather(*(one(c, headers, n) for n in sizes))
    print(sum(got), "images in", len(sizes), "calls")

asyncio.run(main())

What drafts are for

Draft mode is a way to look at many compositions cheaply and pick one. On Sume the same idea is a cheap model at a low tier for the sweep and a better model for the winner. Check the catalog pricing line for each before you decide. For another view of Midjourney against the API route, see whether Sume offers a Midjourney API.

Cost of a 24-image sweep

Sume's price is per image, so a 24-image sweep costs 24 x the model's cost_usd. As an illustration, a row billed at $0.05 per image would cost 24 x $0.05 = $1.20 for the set. Read the real line from the endpoints call, and set the first sweep at the smallest size and quality the model allows, then regenerate only the winner at full quality.