How to Generate Consistent Scenes in Midjourney

*This blog uses affiliate links which may earn a commission at no additional cost to you. Thank you for your support!

There is a video that I saw AGES ago, and still quote into my thirties. It is a constant inside joke between my partner and I, and now I’d like to let you all in on it.

Any time there is a slip-up in a tv show, a movie, an internet clip, etc, we always say “Ope, (shut up, we’re northern) your editing lacks continuity.”

It’s to the point that those real-life matrix glitches prompt one or both of us to utter that line. Where’d we get this ridiculous habit?

Ugh, what a gem.

My point is–continuity. Consistency. Consistent scenes in Midjourney. Elisa did an excellent job in her blog about Midjourney seeds (–seed) and how to use them to prompt in the original style or vibe of the image you generated previously to help create consistent images. So why reinvent the wheel?

Y’know, my mom (who is my #1 fan and loves me to pieces, but is also the reason why I’m as blunt, cynical, and dark as I am) has told me a few times “If there is an easy way, and an easier way, you’ll find the hard way.” I’m not saying she’s right, but she definitely isn’t wrong.

As my lovely friends at Pink Horn have seen, I went just a little bit nuts when I was introduced to making videos with AI. Just a little bit.

And because I’m a disgusting perfectionist in all things that don’t really matter to anyone but me and my dopamine, I get tenacious about things. I’ll get every platinum trophy on PS5. I’ll always die to save you from a hook in Dead by Daylight. I’ll use a machete to get double points on every zombie in Call of Duty (and to hear Ana call me a monster as she revives me for the 87451920 time). I will memorize every lyric to any song I hear and I will use them to annoy the party chat.

And my videos WILL not lack continuity–I WILL GENERATE CONSISTENT SCENES IN MIDJOURNEY!!!–(as much as I can manage, anyway).

Consistent SCENES in Midjourney:

So in editing together the above Stephen Gammell-style (all time fav) inspired short, I had an idea for a woman to walk through multiple hallways and comment on all her “perfect” qualities–her brains, her grace, her beauty, etc. But in the background, I wanted there to be pieces of those she’d stolen those abilities from. Therefore, I needed this woman to stay as consistent-looking as possible.

Sure, I can use seeds. That’s what they’re for. And I did for awhile….but, you see….it was too easy. And not perfect enough.

Naturally I found the hard way.

We’re gonna skip ahead to the clip, repaint, and input part. If you want to know more about prompting photos and art styles, read the damn blog.

I did my prompting and remixing and came up with this photo:

consistent scenes in Midjourney

Style: Check.

Creepy Woman: Check.

Gammell style: Checkety check check.

I love the way the woman looks in this shot, I’ll mess around with movement in Pika Labs in a few, but first I want to prompt the other images I have planned out for this horror short. And I need the woman to look as consistent as possible to the original in each frame.

I prompted the next scene I wanted, using the seed from the original and it wasn’t at all what I wanted. A few more remixes and several reprompts did nothing for me.

consistent scenes in Midjourney

Then I started to think (never a good idea) about repainting the area, but with a photo instead of a prompt.

I took this lovely image to Picsart and cropped it, then uploaded it to my personal discord server and copied the image URL.

I plugged that URL right in front of the prompt, then I varied the region of the woman and tried to input the cropped photo as a seed, which was unsuccessful.

Then I tried vary: region, repaint the woman, paste the photo link, give minor directions (woman, Stephen Gammell style. Long dress) and voila, look at these beauties:

Of course, if you found those directions tough to follow, I made a video for you:

Consistency with Filmora 12 Smart Cutout

But I ran into another problem while using Pika Labs to generate my video footage. I wanted the figures to dance behind the woman, and for the woman to just keep walking. Pika disagreed. If Pika had the chance, it’d ask the world to dance (name the song and you get a cookie).

I was making things difficult again. But after two days of frustration, fear, and a general hatred for life, I figured out how to merge/splice/stick together two different videos to create the one consistent scene.

Successfully editing them together was all thanks to Filmora 12’s AI Smart Cutout tools. Check out my working session below where it actually looks like I know what I’m doing:

A few audio files, some music, and some more Filmora magic, this article has deets on all that good stuff, the Pink Horn team received the full video of me finally nailing the shot. The rest is history.

Just make sure to never let your editing lack that continuity.

Stay chaotic my friends.

Scroll to Top