Midjourney News 10/08/2026 AI Rating: Medium

Midjourney Tests 'Thinking Mode' and Launches Shared Folders in Alpha

#Midjourney#Alpha#Thinking Mode#Shared Folders#Collaboration

Midjourney published two updates on October 8, 2026 for its alpha website (alpha.midjourney.com): a weekly changelog introducing shared folders as its headline feature, and a separate dedicated post detailing an early “Thinking Mode” that the changelog had also flagged as new.

Details

  • Thinking Mode: lets users rerun a job made with V8.2 (standard or edit) “with thinking” — Midjourney looks at the resulting image, identifies where it missed the prompt (wrong objects, layout, anatomy, or text), and generates again to fix it; it’s accessed from a job’s lightbox via “Rerun (Thinking)” and is explicitly an early version the team wants feedback on
  • Shared Folders: folders can now be shared with other users, who can be invited by username as either Collaborators (who can generate directly into the folder and build on others’ work) or Viewers (who can only browse); folders can also be shared by link with private, “anyone with the link,” or open-to-any-member access levels, and the owner can delete a shared folder for everyone
  • Privacy caveat: Midjourney stressed that sharing a folder does not change individual image privacy — non-stealth images inside a shared folder can still show up on Explore and on creators’ profiles
  • Other changes bundled into the same changelog: stalled job submissions now fail cleanly within two minutes instead of hanging, the inline editor no longer crashes after an edit, users can pin a folder to the top of their sidebar, folder search/filters moved into the main feed toolbar, and Live Style Previews now work with custom styles

What happened next

Both features remain explicitly experimental and alpha-only, with Midjourney asking for feedback through its Discord #alpha-ideas-and-bugs and #ideas-and-features channels before any broader rollout. Shared Folders is Midjourney’s most direct move yet toward multiplayer-style collaboration on image generation, while Thinking Mode points toward using a self-review pass — checking a generated image against its own prompt and re-generating to fix mismatches — as a way to improve accuracy without training an entirely new model.