Midjourney tests 'Thinking Mode' for images
Midjourney is alpha-testing a "Thinking Mode" that reruns image jobs with extra reasoning, which it says improves prompt accuracy, typography and coherence, so reasoning-style compute is now being tried in a major image generator.
Midjourney said in a post to users that it is testing a new "Thinking Mode" for image generation on its Alpha website, alpha.midjourney.com, and asked for feedback.
The feature is not a standalone generation option yet. To try it, users open a finished job in the lightbox and click "Rerun (Thinking)" in the lower right corner. That regenerates the image with the thinking step applied.
The company says its internal tests show the mode "helps make the model better at prompt accuracy, typography, and coherence." It did not publish benchmarks, example comparisons or any measured results to support that. It also gave no details on how the mode works, how much it adds to generation time or cost, or which model versions it supports.
Midjourney called the release an exploration. It said it wants "to both improve performance and just generally investigate this direction." It has not committed to a launch. The company said it "might" eventually offer the feature more broadly, either as a dedicated thinking mode or as a way to "add more thinking" to a job after it has run.
For now, Midjourney is asking testers to find where the mode helps, where it doesn't and what other problems come up. It wants examples and opinions posted in its #ideas-and-features channel.