You can turn Remix mode on by activating its button in /settings. After that, all Variation buttons (V1–4, Variations) will summon a text window where you can adjust or completely change your prompts.

To illustrate how it works, I’ll start with a story. To put the then-recent Remix mode to the test, I started with a generation of a beetle and then used Remix to obtain a beetle-theme head-piece/mask.

Remix mode is one of the most impactful features in Midjourney. It gives us the next level of control over our generations.

With Remix, it is possible to develop one generation throughout multiple iterations to be as close as possible to our initial idea. Here is another fashion experiment: using Remix, I aimed to obtain an iridescent Yeezy unibody coat inspired by the shape of a tardigrade.

The idea is to start with the reference shape and then step by step approximate its features to the wanted result. For instance, we started with a general picture of floating space tardigrade and then added iridescence. Next step: transforming into a piece of clothing.

On the last step I added more details and referenced a specific style (Yeezy).

It's time for a Remastering and Variations!

Remix is terrific that way. It virtually lets us build very complex prompts by not keeping all the prompt parts in one line. Instead of generating ABCD, you first generate A, then add B, then add C, then finish it with D.

Could we start with a complex prompt from the beginning? Yes, and we could have succeeded, or not. And we definitely have less control this way.

And yes, V4 can do, too!

By the way, speaking of V4...

Let's conduct the same experiment but use --v 4 instead of --v 3.
SPOILER: It didn't go as I expected...
(ﾉ◕ヮ◕)ﾉ*:･ﾟ✧

So first thing, I tried the exact same setup. Start with a space tardigrade, then move to the unibody.

As you can see, it kind of broke half-way, and "Yeezying" it out didn't help much. I also tried less abrupt transitions: instead of going straight to oversized iridescent puffed up unibody I first went to just unibody costume, but weirdly enough, --v 4 didn't bring back good results, only bizarre ones.

I felt Remix mode in V4 was more difficult than V3 and less responsive. However, I am sure it has great potential and is a must-try for any dedicated Midjourney explorer.

As you have seen, you can change the style of a particular generation on the go.

And not just that. In some cases, you can swap specific details, even the subject of your scene!

This archbishop has a remix history of his own, and it's a good example of how style AND subject swap work in V3. The last one is our guy before Remaster!

What about V4? Let's check with roughly same ingredients. I upscaled three of the first variants of the prompt and remixed different elements for each of them.

I'd say, pretty impressive. What about style transfer? It works, too.

I decided to put V4 Remix mode to another test—to compare it to V3 experience. Remember the Beetle to Headpiece case? Here the premise is close: I wanted to travel from a drawing of a beetle to a Napoleon Army samurai. V4 gave me hard time...

I started with a blueprint of a beetle, and then went on to give the beetle some volume and make it more realistic.

Happy with the last beetle, I proceeded to turn it into a samurai.

Didn’t really work: I wanted a full body portrait, so I came back to square "Beetle" and started again from there.

Once again, third step, an attempt to add Napoleon vibes, didn't work. I came one step back and tried again. And then it broke again. :/

I mean, Napoleon guys are technically what I was asking for, but I almost lost the shape of the beetle there. So I went once step back once again and tried yet again from my previous U1.

That was close enough, yet somehow V4's Remix experience was less smooth, often leading to dead-ends. So my personal preference stays with V3, at least until the next iteration of V4 is out. ಠﭛಠ

Conclusion? Remix is an amazing tool to build up very complex prompts or to drive a generation through a series of changes to achieve a completely different result based on the initial prompt.

Happy midjourneys!

— Andrei Kovalev