What is Midjourney V6, and what's different about it?

Midjourney, one of the most popular AI image generators, has released version 6 of its model as an early test. Founder David Holz announced it in the service's Discord server, which has more than 17 million members, calling it the most realistic model the team has released. It is the first major new version since V5 arrived in alpha in March.

Most listings date the release December 21. A report published that morning described it as arriving "last night," which in North America means the evening of December 20. Either way, V6 is available now to Midjourney subscribers, but you have to switch it on.

What Midjourney is, in brief

Midjourney is made by Midjourney, Inc., a small San Francisco company that calls itself an independent research lab, led by Holz, who earlier co-founded the company Leap Motion. Its service opened to the public in July 2022.

You use it mainly through Discord, a chat app. You type the command /imagine followed by a prompt, a text description of the picture you want, and the Midjourney bot replies with a grid of four images. Buttons under the grid let you upscale one (make a larger, more detailed version) or create variations on it. Optional settings called parameters, added at the end of a prompt with two dashes, control things like the shape of the image or the model version.

What's new in V6

Holz listed the main changes in his announcement:

Change	What it means for you
Much more accurate prompt following	The image is more likely to contain what you asked for, in the arrangement you described
Longer prompts	You can describe a scene in more detail without the model ignoring the end
Text in images	V6 can draw short text, such as a word on a sign, if you put the words in quotation marks
More realism and detail	Early examples show finer skin texture, lighting and reflections at the same resolution
New upscalers	Two modes, "subtle" and "creative," both doubling the image size

Holz described V6 as the third model Midjourney has trained from scratch on its own large computing clusters, and said it took nine months. Text was a notable gap: competing tools, among them Ideogram and OpenAI's DALL·E 3, could already write words into images, while legible text had been a weak spot for Midjourney since its launch. Holz called V6's ability "minor," and suggested using --style raw or a lower --stylize value to help.

Some features from V5.2 are missing for now, including panning left and right and zooming out. Holz said they would be added in later updates, and warned that because V6 is an alpha, it will change often and without notice before its full release.

How to turn it on

V6 is not the default. You have two options:

Type /settings in the Midjourney Discord, or in a direct message to the Midjourney bot, and choose V6 from the version menu at the top.
Or add --v 6 to the end of any prompt.

Why the new prompting matters

For more than a year, Midjourney users have shared prompting tricks: adding words like "award winning," "photorealistic," "8k," a camera brand or a film type to push the model toward polished, cinematic results.

V6 changes that. Holz said it is much more sensitive to the words in your prompt and advised users to drop that kind of "junk." The trade-off, in his words, is that results may feel "less vibey" unless you ask for a mood, but the model understands you far better when you are explicit. His message to users was direct: "You will need to re-learn how to prompt."

Two parameters help steer this:

--style raw gives more photographic, literal, less opinionated results.
--stylize sets how much Midjourney's own taste is applied. The default is 100; Holz suggested that lower values may follow your prompt more closely, and higher values, up to 1000, may look more beautiful.

Early reactions are mixed. Some users are posting strikingly detailed portraits and promising text results, while at least one reviewer who tested it found the difference from V5.2 smaller than expected.

Try it: Take a prompt you have used before with keyword padding. Run it once in V6 unchanged, then rewrite it as two or three plain sentences describing exactly what is in the scene, and run it again with --v 6. Compare which version matches what you pictured.

How should you rewrite your prompts for Midjourney V6?

Write like you are describing a photo to a person, not tagging it for a search engine. Say who or what is in the image, what they are doing, where, the lighting and the point of view. Drop quality words that don't describe anything visible.

Older keyword-style prompt	V6-style rewrite
old fisherman, portrait, award winning, 8k, ultra detailed, cinematic	Close-up portrait of an elderly fisherman with a gray beard, wearing a yellow raincoat, standing on a wooden dock at dawn, soft side light from the left
cozy cafe sign, cute, trending	A hand-painted wooden sign above a small cafe door that reads "Open Early," morning sun, brick wall

If you want text in the image, keep it short and put it in quotation marks, as in the second example. If the result looks too stylized, add --style raw or lower --stylize; if it looks flat, try a higher value or describe the mood you want in words.

Finally, expect change. V6 is an alpha, so results may shift from week to week. Keep a small set of test prompts and rerun them now and then; it is the quickest way to learn what the model rewards.