Midjourney tests a 'Thinking Mode' for sharper prompts

Midjourney is testing a new "Thinking Mode" to see what happens when an image generator takes a moment to process before it paints.

The feature is currently live on its Alpha website. Users can try it by opening the lightbox of an existing generation and hitting "Rerun (Thinking)" in the lower right corner.

Giving diffusion a second thought

According to the team, internal tests show that letting the model think yields noticeable upgrades in prompt accuracy, typography, and visual coherence.

Text rendering and complex instructions have long been notorious weak spots for image models. Midjourney is now collecting community feedback on where it helps and where it falls flat.

If the experiment pans out, the team says it might roll out widely as a dedicated mode or as a post-generation option to "add more thinking" to an existing image.

Why it matters

Reasoning-time compute is no longer just for chatbots like OpenAI o-series; it is coming for creative visual workflows.

For designers struggling with botched lettering or ignored prompt details, trading a few extra seconds of compute for a usable first draft is an easy bargain. We think offering this as a post-run reroll rather than replacing the fast default is the right move, preserving the addictive speed of casual generation while giving pros a precision tool when details actually matter.