Midjourney Thinking Mode: Is a Rerun Worth Keeping?

A Thinking rerun earns its place when it fixes a requirement that the original image missed. A prettier result can still be the wrong replacement: a label may become readable while a product changes shape, or a scene may gain detail while losing the empty space needed for a headline. Decide what must improve before clicking the button.

Midjourney's October 8 announcement, checked on October 12, 2026, describes an Alpha website test. The entry point is Rerun (Thinking) beneath a job's lightbox, in the lower-right corner. Midjourney reports improvements in prompt accuracy, typography, and coherence in its tests and asks users for examples where it helps or fails. Wider availability remains a possible future direction in that announcement. This article offers a review method, not an account-level test or a measured success rate.

Choose a failure that you can name

Start with an existing image that is close to useful. Write one sentence explaining why it cannot yet be delivered. “The headline is misspelled” is actionable. “It needs more impact” leaves too much room for taste to overwhelm the comparison.

For a product composition, the defect might be a missing handle, an extra object, or an incorrect spatial relationship. For a poster, it might be a broken word or text placed across the subject. For a scene, it might be a shadow that contradicts the light source. Keep the defect specific enough that another reviewer could recognize it without hearing your explanation twice.

Then name what already works. The camera angle, background color, subject silhouette, and space reserved for a caption may all be essential. These are preservation requirements. They matter because a successful repair should not quietly create a larger production problem elsewhere in the image.

Keep the comparison tied to one prompt

Save the original job, exact prompt, references, and visible settings before rerunning it. If you have been experimenting with folder defaults or style references, check them first. The Alpha folder and style preview workflow explains how to keep that setup understandable.

For the first comparison, use the same source job. Do not simultaneously rewrite the prompt, swap a reference image, and change the aspect ratio. If several inputs change, you can still choose your favorite result, but you cannot reasonably attribute the difference to Thinking alone.

This is a proposed practice brief, not an observed output:

A cream ceramic teapot on a pale blue table, handle on the right and spout on the left. One small cup sits in front of the teapot. A paper card behind the cup reads “TEA AT FOUR”. Soft window light enters from the left. Leave the upper third of the composition uncluttered.

The brief contains a count, a relationship, a short phrase, and a layout requirement. Those give the reviewer concrete things to check. Use your own production brief if it has equally clear requirements; there is no benefit in testing a complicated scene that you would never need.

Start Creating

Read the image in three passes

First, check literal compliance. Count the objects. Confirm which object is in front. Read every character in the requested phrase, including spacing and punctuation. Do this at the size at which the image will actually appear. Text that looks convincing in a thumbnail can contain obvious errors when enlarged.

Second, inspect relationships. A handle and spout may both exist yet belong to an impossible object. A cup can overlap a card in a way that makes the text unusable. Look at contact points, perspective, repeated shapes, reflections, and shadows. The goal is to find contradictions that a quick aesthetic judgment might miss.

Third, check preservation. Compare the silhouette, visual hierarchy, crop, and usable negative space with the original. If the new image fixes the card but removes the area needed for the campaign headline, record both outcomes. Do not combine them into a vague score that hides the tradeoff.

A compact review note can use three entries: requirement, observation, and decision. For example, “Card wording: all characters readable at final size; pass” is stronger than “text looks better.” If a detail is ambiguous, mark it unresolved and inspect it directly instead of treating confidence as evidence.

Decide whether to keep, edit, or try a different brief

Keep the rerun when the named defect is fixed and the preservation requirements still hold. Save the accepted file separately from the source so that a later experiment cannot accidentally replace it.

Choose a targeted edit when most of the frame is approved and the remaining defect occupies a small area. A complete rerun creates another whole-image review task. For that situation, the V8.2 inpainting and outpainting guide offers a different workflow. A Thinking rerun and a local edit answer different production needs.

Rewrite the brief when it contains conflicting instructions. A tiny card cannot carry a long paragraph and remain naturally readable in a wide room shot. Decide whether the text or the room is the priority, then make a separate comparison with the revised prompt. Label that experiment clearly so it does not get mixed into the original comparison.

For exact campaign wording, consider adding the final text in your layout software after approving the illustration. That is a production choice, not a claim that the model cannot generate text. It gives you direct control over spelling, font, line breaks, and last-minute copy changes.

Stop when the comparison stops answering a question

Repeated reruns can become a search for novelty rather than a repair process. Set a small experiment budget that suits the project, then review the results together. This is an editorial recommendation, not a statement about Midjourney's pricing or how the test consumes account resources.

If successive candidates keep trading one defect for another, change the approach: simplify the scene, separate typography from image generation, or edit the approved frame. Preserve unsuccessful candidates that reveal a repeatable problem. They can help a teammate understand why a visually attractive version was rejected.

If the Thinking entry is absent, check the current Alpha interface and official announcement. The October post does not establish a Discord command, an API parameter, universal access, or a particular subscription entitlement. Avoid copying an invented flag into a production workflow.

Hand off the decision with the image

Deliver the chosen image together with its source job, prompt, intended crop, and a short note about the defect that was repaired. Record any unresolved wording or shape issue before the asset reaches layout or animation. A reviewer should be able to understand why this file was selected without comparing a folder of unnamed downloads.

For a campaign, apply the same acceptance rule to each asset. One successful poster does not establish that every later prompt will behave the same way. Thinking is most useful to evaluate as an additional attempt with a clear purpose: repair a named problem, protect the parts already approved, and keep the version that meets the brief.