Midjourney Adds Thinking Mode for V8.2 and Shared Folders in Alpha

Midjourney has released its latest weekly update to the alpha version of its site, alpha.midjourney.com, dated October 7, 2026. There are two major changes: a thinking mode that reruns jobs on the 8.2 model with a self-check of the result, and shared folders with two access roles, Collaborators and Viewers. The company calls both features early versions and asks users to send feedback. The rest of the changelog covers bug fixes in the editor, library organization, moodboards and the prompt bar.

Thinking Mode: the model checks its own work

The mode works with jobs created on version 8.2 in standard and edit modes. The user reruns a finished generation with the "thinking" option. Midjourney looks at the resulting image, finds the places where it diverged from the prompt, and generates it again. The company lists four types of errors the check targets: wrong objects, composition, anatomy and text.

The post gives no details on how it works. It is unknown how much time and how many GPU minutes such a rerun takes, how it is billed, or how many iterations the model performs. For a service where users have spent years tuning wording by hand, the idea is notable: some routine repeat runs can be handed off to the model. Anatomy and text inside images are exactly where generators miss most often.

Shared folders: collaboration with a privacy caveat

Folders can now be opened to other people. Invitations go out by username, and you pick one of two roles. Collaborators generate images directly in the folder and build on each other's work. Viewers can only browse the contents. There is also link access with three modes: only people who were added, view-only for anyone with the link, or the ability to add work for any Midjourney member. The owner can delete a shared folder for everyone, but the images stay in each author's personal history.

One separate warning: sharing does not change the privacy settings of the images themselves. If a work was not created in stealth mode, it may appear in Explore and on profile pages, even if it was generated in a private folder. For teams producing client materials, this is the key detail.

Fixes and minor improvements

Most of the changelog came out of discussions in the #alpha-ideas-and-bugs channel and office hours. The main items:

• Stuck edit submissions ("Submitting...") now end with an error within 2 minutes, and jobs that actually completed appear in the feed.
• The built-in editor no longer crashes to an error screen after an edit; older Firefox works again; long histories load all the way through.
• Hotkeys do not fire while a dialog or an enlarged image is open.
• A folder can be pinned to the top of the sidebar and renamed in place. Search and filters in Organize have moved to the top row of the feed.
• Deleting a personalization profile now gives an Undo button instead of a confirmation dialog.
• Moodboards: large uploads now go in batches of several images with a progress indicator.
• Prompts: --tile with an attached image returns an incompatibility message, --pv turns into a "pill" like other parameters, and Extend and Loop in video report the length limit.
• Draft mode on V8.1 and V8.2 always produces a grid of 24 images, and this is now stated in the tooltips and on the Try buttons.
• Live Style Previews work on custom styles.

Messages about account restrictions have been reworked separately. A temporary block now shows an Appeal button, and on restricted accounts the controls are dimmed with an explanation instead of a moderator error.

Planned next: an edit history that persists when you leave and come back, and new ways to search for and create styles. If you work in Midjourney with a team, check stealth mode before setting up a shared folder with client materials. Thinking Mode is worth trying on prompts where hands, composition or lettering usually break, and comparing it with manual reruns on time and GPU-hour spend.