What's new in Midjourney V6.1, and how do you use it?

On Tuesday, July 30, 2024, Midjourney released V6.1, a new version of the AI model that turns written prompts into pictures. It is a point release rather than a new generation, but the list of changes is long, and Midjourney has made it the default for all users right away.

The improvements target the details that most often give away an AI image: tangled fingers, awkward limbs, waxy skin and garbled lettering. Tom's Guide, which tested it on release, says skin now looks more natural and text is easier to read, and calls it a bigger step than the small version number suggests.

What changed in V6.1

Here are the main changes from Midjourney's announcement, in plain terms:

Change	What it means for your images
More coherent images	Hands, limbs and whole bodies, as well as animals and plants, hold together better
Better image quality	Fewer stray pixel artifacts; richer textures and skin
Sharper small features	Eyes, small faces and distant hands come out more precise
New 2x upscalers	Enlarged images keep more texture and detail
About 25% faster	Standard jobs finish sooner
Better text	Words you put in "quotation marks" in a prompt are drawn more accurately
New personalization model	Your personal style profile is applied with more nuance and accuracy
Personalization code versions	An old code keeps the style data from the job where you first used it
New `--q 2` mode	Takes 25% longer and sometimes adds texture, at the cost of some coherence

Midjourney sums it up by saying images should look "generally more beautiful" across the board.

A few tools have not caught up. Inpainting and outpainting, which let you repaint part of an image or extend it beyond its edges, still use V6.0 for now. That covers zoom, reframe, repaint and vary region.

When and where it arrived

Midjourney posted the update on its announcement page on July 30 and wrote on X that V6.1 is its best model yet. You can use it in two places: on Discord, the chat app where Midjourney started and where you type prompts to its bot, and on the Midjourney website, which has its own settings panel. Tom's Guide confirms that the version switch works on both.

Some people had expected a bigger "V6.5" this summer. Instead, Midjourney chose smaller, more frequent steps and says a V6.2 will likely follow in a month or so. It is making V6.1 the default partly so it can gather as much usage data as possible before that next release. The company also says this update was shaped by the priorities its community submitted and ranked on its ideas page.

How an image model like this works

Midjourney does not publish the details of its models, but the general idea behind today's image generators is well known. The model has learned, from a very large number of pictures paired with descriptions, how words relate to shapes, colors, light and texture. When you send a prompt, it starts from random visual noise and refines it step by step toward an image that fits your description.

An analogy helps. Imagine a sculptor who starts with a rough block and removes a little material at each pass, checking the description each time. A better model is a sculptor with steadier hands: it makes fewer mistakes in the hard places, like the joints of fingers, and finishes faster.

An upscaler is a second step that enlarges the finished picture. Simply stretching an image makes it blurry, so an upscaler adds plausible detail as it grows the image. V6.1's new 2x upscalers are designed to keep textures crisper during that enlargement.

Personalization is a Midjourney feature that steers results toward a personal style, identified by a code you can add to prompts. With versioning, reusing a code from an old job brings back that job's personalization model and style data, so the look stays consistent.

Why it matters this week

For designers, illustrators and hobbyists, V6.1 reduces cleanup work. Better hands and bodies mean fewer rerolls, and more reliable lettering makes posters, mock-ups and greeting cards more practical. In Tom's Guide's test, a movie-poster prompt with a quoted title and subtitle produced four options; two matched the requested style, and one of those spelled both lines correctly. That is progress, but it also shows that text still needs checking.

The update also shows how this field moves: steady refinements in short cycles, guided by what users vote for.

Try it: Write one prompt with a person holding a sign that shows a short phrase in quotation marks. Run it with V6.1, then add --v 6 to run it on the older model. Compare the hands and the spelling side by side.

How do you get the best results from Midjourney V6.1?

You do not need to do anything to start: V6.1 is the default. To be sure, add --v 6.1 at the end of your prompt, or pick it in /settings on Discord or in the website's settings panel. To go back, choose V6 or add --v 6.

Be specific and concrete. Describe the subject, the setting, the light and the style, and put any words you want drawn inside quotation marks:

Prompt

A hand-painted wooden sign reading "Fresh Bread" hanging outside a small bakery at sunrise, warm light, watercolor style --v 6.1

Try --q 2 when texture matters more than perfect structure, such as for fabric, stone or foliage, and compare it with the standard result. If you use personalization, keep a note of the codes you like, since each one now remembers the style it produced. Finally, look closely at hands, text and small faces before you use an image; they are much better, but not flawless.