Most Midjourney users fix a bad image the same way: they rewrite the prompt, add a few more adjectives, and roll the dice again. Most people do not know that, since 7 October, there is a second option on Midjourney's alpha site. You can ask the model to look at its own image, work out where it missed your prompt, and try again.
It is called Thinking Mode, and it lives behind a small "Rerun (Thinking)" button. It is early, it is slower, and Midjourney has published no numbers for it. But for the most common Midjourney failures (the wrong number of objects, a broken layout, mangled text on a sign) it changes what the right next click is. Here is how it works, when to use it instead of rewriting, how to write prompts it can check against, and how the new shared folders fit a team workflow.
What is Midjourney Thinking Mode?
Midjourney Thinking Mode is an experimental rerun option on alpha.midjourney.com for images made with V8.2, including edits. When you click "Rerun (Thinking)" in the lower right of the lightbox, Midjourney compares the image with your prompt, finds where it missed (wrong objects, layout, anatomy or text) and generates again.
The details come from Midjourney's own alpha changelog of 7 October 2026 and its "Testing Thinking Mode" post of 8 October:
- Where: the alpha website only, not Discord and not the main site yet.
- Which images: jobs made with V8.2, both standard generations and edits.
- What it targets: Midjourney says tests show it helps with "prompt accuracy, typography, and coherence."
- Status: an early version. Midjourney says it "might" make it available broadly as a thinking mode, or as a way to "add more thinking" after a job.
In plain terms, it is a second pass that reads your prompt like a checklist. That is useful, but only if your prompt is written so it can be checked.
When should you rerun with Thinking instead of rewriting the prompt?
Use Thinking when the image ignored something your prompt clearly said: a missing object, the wrong count, a broken layout, warped hands or misspelt text. Rewrite the prompt when the image did what you asked but you asked for the wrong thing, such as the wrong mood or style. Use Edit when only one small area is wrong.
A quick way to decide is to ask one question: "Did Midjourney disobey me, or did I describe it badly?"
- It disobeyed you (three cups when you said two, "SALE" spelt "SLAE", the logo on the wrong side): click Rerun (Thinking). The prompt is fine, the execution missed.
- You described it badly (it looks cold when you wanted warm, too busy, wrong era): rewrite the prompt, change the style reference or adjust --stylize. Thinking cannot fix a target you never set.
- One region is off (a hand, a label, a background object): use the V8.2 editor and change only the selected pixels. Our Midjourney Edit Model guide covers that workflow.
The habit to break is rewriting a prompt that was already right. Every rewrite changes the target, so you lose the parts that were working.
There is also a cost argument. A rewrite usually means a fresh grid of four images, a new round of comparing them, and often a second rewrite when the new grid fixes one problem and creates another. A Thinking rerun starts from the image you already liked and aims at the specific miss. That should mean fewer rounds, as long as the first image was close. If none of the four images in a grid is close, skip Thinking and go back to the prompt, because there is nothing worth correcting yet.
How do you write a prompt Thinking Mode can check?
A checkable prompt states the facts of the image as separate, testable items: how many subjects, where they sit, any exact words in quotes, and the details that must appear. Thinking Mode looks for places where the image missed the prompt, so the more concrete each item is, the more precisely it can find and fix the miss.
Compare two prompts for the same café promotion. "Cosy café flat lay with coffee and pastries and a sale sign" gives Thinking almost nothing to check. Was there coffee? Yes. Done. The version below gives it six things to verify.
Try this prompt:
Top-down flat lay on a pale oak table. Exactly two white ceramic cups of latte, one on the left third and one on the right third. Three croissants on a round grey plate in the centre. A small kraft-paper sign in the bottom centre with the exact text "AUTUMN SALE" in black capital letters. Soft morning window light from the top left, warm neutral tones, lots of empty table space in the top third for a headline. --ar 16:9 --v 8.2
Four rules make a prompt checkable:
- Use numbers, not plurals. "Exactly two cups" can be checked. "Some cups" cannot.
- Give positions. Left third, centre, bottom centre. Layout is one of the things Thinking is built to fix.
- Put exact text in quotes and keep it short. Two or three words render far better than a sentence.
- Separate style from facts. Put lighting and mood after the facts, so the facts are not buried in adjectives.
How do shared folders fit a team image workflow?
Shared folders, launched in the same 7 October alpha update, let you invite people by username as Collaborators, who can generate straight into the folder, or Viewers, who can only browse. You can keep a folder private to invited people, let anyone with the link view it, or open it to any Midjourney member.
For a small marketing team, that turns Midjourney from a solo tool into a shared campaign board. A simple setup:
- One folder per campaign. Name it after the campaign and pin it to the top of the sidebar.
- Designers as Collaborators, account managers or clients as Viewers.
- Set folder defaults. Midjourney's alpha now lets a folder keep its own default parameters, so the aspect ratio and version stay consistent for everyone.
- Pin the checkable prompt in your team notes, so every collaborator starts from the same facts and only varies the style.
One privacy point matters here. Midjourney says sharing "doesn't change image privacy": non-stealth images can still appear on Explore and on profiles, even inside a private folder. If a campaign is confidential, work in Stealth Mode, which Midjourney offers on its higher-tier plans, or keep the work off Midjourney until launch. If you also need on-brand images in other tools, our Gemini brand style guide covers a different approach.
Where does Thinking Mode fall short?
Thinking Mode is an alpha experiment with no published results. It only works on V8.2 jobs on the alpha site, it is slower than a normal rerun, and an early hands-on review reports that reruns use Fast GPU time. It also cannot rescue a vague prompt, because it can only check what you actually wrote.
Specific limits to plan around:
- No official numbers. Midjourney says only that tests show it "helps". One headline online claims an 80% cut in prompt failures, but we could not find that figure in Midjourney's own posts, so do not plan around it.
- Cost. A Japanese review on AIpicks (9 October) says reruns consume Fast GPU time but need no extra plan. Watch your hours if you run it on every image.
- Not for taste. It fixes misses against the prompt. It will not make a dull concept interesting.
- It may change. Alpha features move fast. The button, the name or the behaviour may be different by the time you read this.
- Long text is still risky. Thinking targets typography, but a full sentence on a poster is still better added later in a design tool.
How can you test Thinking Mode in 20 minutes?
Take one prompt that failed you recently, rewrite it as a checkable prompt with counts, positions and quoted text, and generate four images on the alpha site. Pick the image with the clearest miss, click Rerun (Thinking), and count how many of your stated facts are correct before and after.
- Is the object count right?
- Is each item where you placed it?
- Is the quoted text spelt correctly?
- Are hands, faces and edges clean?
- Did the style you liked survive the rerun?
If Thinking fixes two or more facts without losing the style, make it your first move whenever an image disobeys a clear prompt, and save rewriting for when the idea itself is wrong. Most of the gain, though, comes from the checkable prompt. It helps every rerun, with or without Thinking, and it makes your prompts easier for teammates to reuse.
The best Midjourney users are not the ones with the longest prompts. They are the ones who can tell a bad prompt from a bad roll. We understand AI. We understand you better. With UD by your side, AI doesn't feel cold.
Reviewed by the UD AI team. Feature details are from Midjourney's alpha changelog (7 October 2026) and "Testing Thinking Mode" update (8 October 2026), with cost notes from AIpicks (9 October 2026). Thinking Mode is an alpha experiment and may change without notice.
How does your team's AI skill really rank?
Now that you have the technique, the next step is building it into a workflow that runs reliably every time. Start with UD's free AI Rank to see where your business stands on AI adoption, and we'll walk you through every step, from tool setup to workflow design and deployment.