Categories
Artificial Intelligence

Experimenting with Midjourney’s Zoom

Much to my surprise, I’ve been experimenting with Midjourney’s Zoom function, which they just added. What got me started was the unexpected realization that people are doing animations with these images. I’d never thought of that.

Much to my surprise, I’ve been experimenting with Midjourney’s Zoom function, which they just added. What got me started was the unexpected realization that people are doing animations with these images. I’d never thought of that. But why not? If you essentially have infinite zoom, you could stitch those together in video editing software and zoom between them, and have animation. I’ve only just begun to consider the implications and possibilities.

I have good reason to consider these possibilities. Over the last few years I’ve been working on an album and a collection of short stories, which are intended to compliment each other. Each song on the album will correspond to a short story, and they’re meant to be taken in together (though each will also stand on their own). I first started dabbling in generative A.I. quite by accident. I’d been thinking about how I wanted to package the album, and decided I was going to draw an image to correspond with each song / story, and those would be included in a booklet inside the CD. I knew they would necessarily be rather simple images, because I wouldn’t have the time to do anything elaborate.

Then I stumbled onto Midjourney. And I started thinking… maybe I can do more elaborate images after all.

At this point I’ve been tinkering with Midjourney for about a year. I’ve generated thousands of images. The early stuff was rather crude, and mostly had a “wow” factor just because of what it was, a new thing to me, this magic of generating images out of thin air using text prompts. I eventually learned ways to make better images. Then I stumbled onto image prompts, and started using my old artwork and even family photographs as root images to create new things. And I eventually started re-doing old Midjourney images as they released new versions, each better than the last. I’ve been creating images on Midjourney since V3, on up through V4, and V5, to the latest, V5.2. Each iteration got better, and the images got more and more amazing.

Then recently I saw a curious new button on some of my images. It was the new “Zoom” button. Honestly, I didn’t fully appreciate it at first. I zoom’d out on a few of my images, and really liked it. But I mostly saw it as a variation of Dall-E’s out-painting features. That perception changed when I realized people were creating videos using Midjourney’s Zoom. Naturally, I started thinking of ways I could use it, too. And since I’ve been thinking about creating some videos for some of the songs on the album, I’m naturally excited about the possibilities this represents.

Anyway, I’ve been doing some experiments with this in mind. Here are some of the test images.

First, I made the following series based on an old prompt for the early days coupled with V5.2, and expanded using Zoom. But if you want an idea of how far Midjourney’s algorithms have come in a short period of time, just look at the image below. This is an image I did about a year ago using the exact same prompt as the images which follow it. Before, meet after.

Woman On The Moon - OG

This first image below was based on that old prompt, which referred to a woman trapped in the wreckage of a spacecraft on the Moon. Midjourney has come a long way in a year!

Woman in Wreckage on The Moon - 01
Original image

Once I had the new image I wanted to Zoom in on, the aim was to create a series of base images I could then use for making smooth videos that zoomed out seamlessly. From there I then made the following images, each with a more distant perspective.

Woman in Wreckage on The Moon - 02
2nd image, 2x zoom
Woman in Wreckage on The Moon - 03
3rd image, 2x zoom
Woman in Wreckage on The Moon - 04
4th image, 2x zoom

Of course, me being me, I started thinking of other ways to do things, and other possibilities. I tried using modified prompts to alter the result with each images, but didn’t really understand what that meant. I kept the original prompt but modified it. And while I got cool images, the subsequent images didn’t reflect my changes in the prompts. I would later learn that what I needed to do was abandon the original prompt altogether and simply describe what was happening with each subsequent image.

What’s listed below is a good example of that. I’ve experimented with using my old drawings and base images, or image prompts, to have Midjourney generate new but recognizable images on top of it. The following image is an original colored pencil drawing I made back in the late 1980s or so.

Original pencil drawing of woman in blue dress.
Original pencil drawing of woman in blue dress.

I had tinkered with creating new images before, but wanted to see what v5.2 would do with it.

Midjourney render using woman in blue dress drawing as an image prompt.
Midjourney render using woman in blue dress drawing as an image prompt.

Okay, color me impressed! And, of course, from there I wanted to experiment with changing the context of the image as I zoom’d out. I kept the original prompt this time, but put it in a different context, this time describing it as someone posing among photography equipment during a photo shoot in an English garden. That resulted in this…

Midjourney 2x zoom out using modified prompt
Midjourney 2x zoom out using modified prompt

I also wanted to try changing the ratio of the image, so I zoom’d out again using the same prompt as above, but with a different ratio (16:9 if you must know). That gave me this…

Midjourney 2x zoom out using modified ratio
Midjourney 2x zoom out using modified ratio

I also decided to try some others, and let Midjourney run on its own, just to see what it would come up with. The following is a test of v5.2, which was based upon a line drawing.

Then I started zooming out…

Woman wearing Día de Muertos makeup and a wedding dress
Woman wearing Día de Muertos makeup and a wedding dress

I wanted to see what Midjourney would come up with if I provided no additional prompts or modifications beyond 2x zoom.

Día de Muertos makeup, 2x zoom
Día de Muertos makeup, 2x zoom

Actually, I quite liked that. So… what would it do if we went a bit further, another 2x zoom?

Día de Muertos makeup, 2x zoom
Día de Muertos makeup, 2x zoom

I quite like that. Each image expanded on the idea of the original image. I did pick my favorite of each set of images Midjourney gave me, and some of the variations were quite wild. But the ones I picked and upscaled seemed to stay with the theme and feel of the first image.

Why stop there? I had another initial image I quite liked, again based on an old prompt, slightly modified. I’ll just post these and let them speak for themselves.

A beautiful Persian woman lounging on a chaise
A beautiful Persian woman lounging on a chaise
A beautiful Persian woman lounging on a chaise, 2x zoom
A beautiful Persian woman lounging on a chaise, 2x zoom
A beautiful Persian woman lounging on a chaise, 2x zoom
A beautiful Persian woman lounging on a chaise, 2x zoom

The power and adaptability of Midjourney’s Zoom function should be apparent from these images. I’m impressed. It’s a given I’ll be exploring the possibilities here.

But keep in mind, I generated these images to experiment with videos. I’m still going to do that. But I think we’ve crammed enough images into this post. I’ll include the results of the video experiments in another post.

Hopefully the images I’ve included here will give you a good reason to go experiment with Midjourney’s new Zoom feature. It really does open up a lot of possibilities.

Anyway, color me impressed.

0 0 votes
Article Rating
Subscribe
Notify of
guest
0 Comments
Inline Feedbacks
View all comments
0
Would love your thoughts, please comment.x
()
x
Close Bitnami banner
Bitnami