Midjourney Rerun (Thinking) is alpha-only: the Sume workaround

Midjourney's Rerun (Thinking) is a button in the lightbox of a finished job on the alpha website. Midjourney's announcement, dated October 8, 2026 and read 2026-10-12, says it reruns the job so the model can look at where it missed the prompt, and that early tests help with prompt accuracy, typography and coherence. It is an early public test with no stated cost or wider rollout date, so nothing about it is callable from code today.

Sume has no Midjourney model and no thinking rerun. What it does have is a quality ladder on ChatGPT Image 2.5, which is the closest retry loop you can script. This post shows what that loop is and what it is not.

What Midjourney says, and what it does not

Only the facts on the announcement page are used here.

The nearest retry loop on Sume

Sume's Images API lists ChatGPT Image 2.5 as openai/gpt-image-2.5, with quality set to auto, low, medium, high, xhigh or max, and a default of high when you omit it. A scripted retry is: render, check the text, and render again at a higher tier when it is wrong.

This is a plain re-render, not a reasoning pass. The seed field exists in the schema but no model serves it in v1, so a retry is a new image, as covered in Midjourney Run as HD.

import os
import requests

prompt = "A poster that reads OPEN LATE, bold sans-serif, teal background"
for quality in ("medium", "xhigh"):
    r = requests.post(
        "https://api.sume.com/v1/images",
        headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
        json={"model": "openai/gpt-image-2.5", "prompt": prompt,
              "quality": quality, "aspect_ratio": "16:9"},
        timeout=60,
    )
    r.raise_for_status()
    print(quality, r.json()["data"][0]["url"])

When one region is wrong

If only the text is off, re-rendering the whole image throws away the parts that were right. ChatGPT Image 2.5 also takes mask_url, so you can re-render just the text zone; see fixing text placement with a mask edit. For the tiers and what each costs, read GPT Image 2.5 quality levels.

Choosing

Use Midjourney's button if you are already in the alpha and want its own model to reconsider the prompt. Use the Sume loop if you need an API, a hard cap on spend that you set yourself.

• Count attempts in your own code; Sume does not retry for you.
• Start at medium and step up, because each tier costs more.
• Switch to a mask edit once the layout is right and only the words are wrong.