Testing Thinking Mode

Midjourney has begun testing a feature it calls "Thinking Mode," an experimental image generation option accessible through its Alpha platform at alpha.midjourney.com. The mode is not yet part of the standard generation flow - users need to open the lightbox view of an existing job and select "Rerun (Thinking)" in the lower right corner to try it out.

The name suggests an approach similar to reasoning-style inference seen in large language models, where the system spends additional compute time working through a problem before producing output. Applied to image generation, this could mean the model takes more deliberate steps to interpret a prompt, resolve ambiguities, or plan compositional elements before committing to a result - though Midjourney has not yet detailed the underlying mechanics.

Thinking Mode joins a broader industry trend of applying extended inference techniques beyond text. Several AI labs have explored using chain-of-thought or search-based reasoning to improve outputs in domains like code and math, and the approach is now being adapted for creative and visual tasks. Whether the additional processing translates to meaningfully better images - in terms of prompt adherence, composition, or detail - is precisely what Midjourney is looking to learn from this test.

The rollout is limited to the Alpha site for now, keeping the experiment separate from the main product while feedback is gathered. Users who try the feature are encouraged to share their impressions, making this as much a structured community evaluation as a technical test. If results are promising, a wider release to the standard Midjourney platform would be a natural next step.