How to Activate Remix Mode

The best way to do this is through your direct messages with the Midjourney bot. Simply type the following command:

/settings … and hit enter (sometimes twice). Once the command is accepted, you will see:

You can experiment with any of the available controls. In green, you can see my default settings. Keep in mind that some options, like “Stealth mode,” are exclusive to premium subscribers, and I’m currently on the $10 monthly plan.

Activate the “Remix mode” button, and it will turn green:

Let’s Dive into Remix!

I’ve rewritten this section multiple times as I wanted to refine it. Imagine you need an illustration for a blog post featuring an apple pie recipe. You know you want an image of an apple, but your vision isn't clear. You might start with an initial prompt like this:

a close-up shot of an apple sitting on a chopping board Given that you lack a specific aesthetic, this is an ideal moment to utilize the chaos parameter. The chaos setting dictates how varied the resulting images will be. According to Midjourney, “Higher chaos values will yield more unconventional and unexpected results.” This is perfect for exploration, so let’s add a high chaos value:

a close-up shot of an apple sitting on a chopping board --c 100 (The “c” signifies “chaos.”) Now, let’s see what we generate:

I personally prefer the second option, so I’ll use that for some remixing.

First, I’m curious about altering the aspect ratio. The images will be square by default, but since this is for a blog, a 16:9 aspect ratio might be more suitable. So, starting with the second option, let’s hit the “V2” button for a remix prompt:

Upon clicking "V2," I’m prompted to refine my request.

The first attempt will be to remove everything and just insert an aspect ratio command, like so:

In theory, this should work. But did it? Let’s check:

It seems that simply inserting an aspect ratio command does not yield the desired outcome. Let’s try keeping the initial prompt while adding the aspect ratio command:

Will this approach work? Let’s find out:

This attempt seems to have merely stretched the image (although there are slight variations among them).

It appears that altering the aspect ratio may not be something Midjourney accommodates using remix. So, let’s restart with a fresh prompt that includes the 16:9 command, which should produce the correct aspect ratio, albeit with a different image. Again, fresh prompt:

a close-up shot of an apple sitting on a chopping board --c 100 --v 5 --s 250 --ar 16:9 --v 5 --s 250

Here’s what we have now:

The aspect ratio looks good. However, I’m not particularly impressed with any of these images. Still, option three stands out because it could allow for enhancements through remixing. It’s beneficial to maintain the overall composition in mind, even if there are flaws (like poor lighting), as improvements might be achievable directly within AI before any post-processing. So let’s give that a shot:

As you can see, I simply added “bright interior lighting” to the remix prompt. Initially, we started with this image:

… and here’s the result we received:

While this isn’t a complete success, the images are slightly brighter. Perhaps an outdoor setting would yield better results? Let’s explore that (again using option three from the original set):

Here, I included “outdoor picnic table scene.” Starting from:

… here’s what we got:

While the results are not groundbreaking, I anticipated that the outdoor context would naturally imply brightness. Let’s really enhance it this time (again remixing from the original):

This time, I specified “outdoor picnic table setting on a bright sunny day.” Will this make a difference? Starting from this image:

… we now have:

These images are indeed an improvement. Although they aren’t as bright as I had hoped (and keep in mind that results may vary!), they are certainly more usable. Notice how the overall composition has largely stayed the same throughout each remix. Let’s reflect on the original results grid:

The dark area beneath each remix (refer to two images above) was actually a part of the original third option, which is an interesting aspect for Midjourney to retain. For a quick comparison, let’s apply the exact same remix to the second option, which does not have that dark area at the bottom. Here’s the potential outcome:

This is intriguing, as option four provides a notably better apple (though, interestingly, it isn’t significantly brighter).

Returning to our original project, I find option three appealing because it incorporates some light and a dash of red:

For the blog banner, I might be inclined to edit it in Photoshop and call it a day. For instance, here’s a quick 2-minute edit where I brightened the image, boosted saturation, and added a title:

This quick edit serves as a reminder that post-processing your AI-generated images is always an option, just like any other artwork (meaning, you can often achieve a close result and then refine it further). Therefore, perfection doesn’t always require rerolling or remixing; aim to get as close as possible to a usable starting point!

Yet, we haven’t fully explored remix’s potential. Remember that Midjourney’s documentation showcases several possibilities we haven’t yet examined. These include:

Let’s try one last experiment before concluding this article. We’ll attempt a medium change. Our previous image appeared as a photograph, but let’s aim for something more imaginative. Here’s my remix (again remixing from the original):

I’m unsure what to expect, as I requested a change in medium and included some stylistic prompts. Will it resemble our source image below?

Take a look at the results:

Option four is quite fascinating. I’ll give it a quick edit in Photoshop, place it atop this piece, and call it a day. I’m certain that I’ve only scratched the surface of what remix can accomplish, but the images themselves reveal the nuances of what was retained and altered. If you create anything remarkable with remix, please share your experiences, as I find it intriguing and, for me, a bit more practical than the standard “V” (variation) functions.