TLDW logo

Consistent Characters in Midjourney, Finally Solved!

By Glibatree

Summary

## Key takeaways - **Three-Step Asset-to-Scene Workflow**: Step one builds assets by grabbing still frames from AI video of your character's face and full body moving and performing actions. Step two designs rough compositions using those assets, and step three makes edits to scenes for interactions without sacrificing consistency. [00:20], [00:45] - **Dual Reference Beats OmniRef**: Generating face and full body together solves prominence problem: close-up faces lose detail when zoomed out, full bodies distort faces, but dual images provide high-quality cohesive references for both. [02:04], [03:18] - **Video Turns Character into Actor**: Use video prompts to direct your character from starting frame to any pose or expression, more controllable and accurate than OmniRef since it works from ground truth reference and counts details like leaves consistently. [05:18], [05:46] - **Batch Edits via Stitched Grids**: Stitch multiple frames side-by-side to edit and upscale all at once in Midjourney editor; provides context so fixes stay character-accurate and speeds up processing over one-by-one. [09:51], [10:27] - **Layers Enable Drag-and-Drop Scenes**: Midjourney layers let you drag transparent character assets into scenes, scale/rotate with alt key, and erase per layer; use close-up heads for best expressions and keep editing layer on top. [17:30], [18:49] - **Second-Order Videos Fix Interactions**: For complex scenes, build rough composition with props, generate video to settle into natural interactions like grabbing a hat, stacking motion for dynamic poses beyond basic assets. [21:53], [23:39]

Topics Covered

  • Prominence dictates reference quality
  • Video prompts turn you into director
  • Stitch grids preserve batch consistency
  • Second-order assets enable interactions

Full Transcript

Omniref just can't get results like this. My new workflow will not only make

this. My new workflow will not only make your midjourney characters more consistent and controllable, but today I'll give you a reliable way to build scenes with any number of consistent

characters in them and teach you to guide and refine those scenes into exactly what you want them to be. My

method involves three steps. Step one is building assets by grabbing still frames from your AI video of your character's face and full body moving and performing actions. This step will enable you to

actions. This step will enable you to turn one reference into all the assets you need to create the poses and expressions for your story or project.

Step two is designing your rough compositions using those assets to create any scene your character needs to be in. And for step three, I'll teach

be in. And for step three, I'll teach you how to make edits to those scenes and get your characters interacting with the world and looking as good as possible. This might also involve

possible. This might also involve cleaning up any of the wonkiness that sometimes comes from photo bashing AI images together. The point will be doing

images together. The point will be doing all of that without sacrificing consistency whatsoever.

This means instead of relying on something like Omni reference to regenerate your characters every time, every pose and expression you use can be based on the ground truth of what your

character actually looks like. To make

it as easy for you as I can, I'm going to share all of the prompts you'll need to make this happen for your characters in your style, as well as break down all of the tips, tricks, and best practices

you'll need while you use this workflow yourself. I don't want to waste any more

yourself. I don't want to waste any more time. Let's get into it. So, I've

time. Let's get into it. So, I've

developed four prompts that make this all possible, and I want to go through each one and explain how it fits into this workflow of turning videos into assets into scenes. Of course, all the prompts I share will be available for

free in the description. The first one is designed to help you get a consistent face and full body reference, both within a single image. This prompt, and these kinds of images are a really nice

thing to have for this method because you're going to split them up and turn them into videos separately. But

generating them together first solves one key problem. And it has to do with what I call prominence. If you had started by generating a closeup face for your character, it's going to be really

detailed. But your options for turning

detailed. But your options for turning that into a full body reference don't really work. We need a full body

really work. We need a full body reference to pose our characters. But if

you try to do that with omni reference, it's possible to ask for a full body image within your prompt and it will generate something. But it takes massive

generate something. But it takes massive liberties on what your character should look like. And I rarely appreciate the

look like. And I rarely appreciate the choices it makes. It's also possible to zoom out on that close-up face if you start from here and do that in the editor. But so often the anatomy ends up

editor. But so often the anatomy ends up really awful. So that doesn't work. But

really awful. So that doesn't work. But

if you try the opposite, if you start with a full body reference, it might be a great design for your character, but those facial features being less prominent means they'll end up

distorted. And if you look closer at the

distorted. And if you look closer at the face, those finer details will be missing a lot of the charm you could have gotten when you generated the face close up. And what's worse, if you use

close up. And what's worse, if you use that slightly deformed face as an omni reference to create a close-up version of your character later, that wonk will just propagate into every future image,

and you'll never like the result. So, by

generating these cool dual images as your starting point, you get both highquality results at once. And then

you have both references you need. You

can use this part when you need close-ups of your character and this part when you need zoomed out versions.

Really the best of both worlds. By the

way, for each of these prompts that I'm showing you, you're going to be sending them to my custom GPT. It's free to use and I've optimized it for generating great prompts for MidJourney version 7

based on a ton of testing. I posted this video about that here. But to get this kind of image for your character, you'll need to copy the prompt off my website.

I'm not asking for an email or anything.

It's just a link in the description. So,

you'll take the prompt and you'll paste it into the gibbitry art designer.

That's my GPT. Replace this part here at the bottom with your character idea and a description of the style that you want your character generated in. This prompt

is flexible enough that when you take the resulting prompt to midJourney, you'll see that no matter what your idea is, you'll reliably get an image in this format with two references, one close up

and one far apart. And of course, as you can see, compared to omni reference or expanding in the editor, these character designs are going to be much more cohesive and consistent to one another,

and they just come out really, really well each time. A much better way to get that first copy of your character. And

now that you have these, you're in an excellent place to use the next prompt we'll need. And that, of course, is

we'll need. And that, of course, is generating our actual videos that we'll be pulling stills from. By sending this to my GPT, you'll be teaching the language model about midjourney video

and how to write a perfect prompt for creating assets. The point of generating

creating assets. The point of generating these videos is to get real control over your character in 3D space and simultaneously be able to describe any pose or expression and get your

character to move from wherever they were in their starting frame to exactly what you asked for. I think of it like this. with a reliable way to prompt the

this. with a reliable way to prompt the video model. Suddenly you turn into a

video model. Suddenly you turn into a director. Once you have a starting

director. Once you have a starting point, a character reference, you are essentially treating this exact character like an actor that will do what you ask. And it can do basically

anything. It just only understands

anything. It just only understands well-written video prompts. Doing it

this way is not only more controllable than Omni Reference. Obviously, omni

reference loves generating these straightforward expressions, but I also find that it's more character accurate than omni reference, right? It's working

from the best reference you have of your character and then just moving things as it shifts into place. So, if you have this cool full body image and use video to generate a new still, you could count

the leaves going up this guy's legs and it would be the same as the reference every time. And what I really love is

every time. And what I really love is once you've done this and pulled assets from these videos, you can use them again and again throughout your project.

To me, this is amazing and it's where your time saving comes from. Instead of

hoping to get lucky and spending hours re-rolling until MidJourney finally gets your character's pose right, illustrating your story becomes a game of clicking and dragging your characters

that are now assets where you want them to be to build out the scenes you're trying to make. So you take this prompt to my GPT and when you send the message include the starting frame for what you

want to be animating your actor. You

could just download it from Midjourney and drag that into chat GPT. And then

here at the bottom you have the option to describe exactly the assets you want the GPT to write prompts to achieve the orders you're giving as a director. So

you'll find doing it this way makes it so easy to get your characters to follow the motion you describe. And I spent a lot of time just optimizing this prompt

so it really makes you feel like that director who has all the control you could want. What I spent the most time

could want. What I spent the most time doing was making sure that when you generate the videos, one, it does what you say, but also your characters stay in frame and move slowly enough that

nothing gets too blurry, which makes it much easier to later cut these out and use them as assets in whatever scene you're trying to put them in. Of course,

I'll show you the easiest way to do that later. But first, the next prompt is all

later. But first, the next prompt is all about editing your images once you have them. You can use this for basically

them. You can use this for basically anything. You can change the color of

anything. You can change the color of clothing on one of your characters. You

can put something in a character's hand or just add interesting background elements into your scene. When you send it, it's just like the other prompts.

You can describe what you're trying to change at the bottom. But the prompt this time also assumes that you'll send it a screenshot of what the editor canvas looks like so that chatpt can see

what you erased and understand what midjourney will have to generate. So you

can just use print screen to save the screenshot into your clipboard and then paste the whole screenshot into chatgpt.

If you're generating characters and building scenes with this method, you will be using the editor a lot. And

without optimizing your prompt in this way, the editor will be pretty good at filling in the gaps between two things and sometimes fixing messy parts. But

essentially, with simple prompts, the editor always makes your image simpler.

So, I highly recommend you use this prompt whenever you're trying to use the editor to add or change details that are complex enough that you might worry MidJourney will just leave it blank in

the edit. But, okay, the last prompt I

the edit. But, okay, the last prompt I want to give you is still about editing, but less about precision and more about saving time and keeping consistency.

It's for batch processing your edits across several images at once. That

might sound a little weird in midjourney. So, let me explain. Let's

midjourney. So, let me explain. Let's

say at this point you've generated a bunch of videos of your character.

Videos are imperfect, so the stills you're taking from them are sometimes going to be a little bit of a mess.

Generating an image is like eight times cheaper than generating a video. So, if

you really like a pose, but there are some things you want to fix, it makes sense to edit the result rather than generating a new video and hoping to get luckier next time. And at the same time,

videos on midjourney are low resolution.

So you might want to upscale the result before you really turn the stills into assets. You could just do those two

assets. You could just do those two things one at a time. Grab a still, edit it, and then use the upscaler for each still you want to turn into an asset one

by one. But there's two problems. First

by one. But there's two problems. First off, it's slow and tedious. The second

problem is everything you erase and try to fix has a chance to come out in a way that is not character accurate. I mean,

you erased it, right? So, Midjourney

doesn't know what it's supposed to look like. So, the solution here that fixes

like. So, the solution here that fixes both of those problems is to take all of the frames you want to fix and mash them together side by side before editing and

then upscaling them all at once in this stitched image. This is probably the

stitched image. This is probably the most powerful thing I do to speed this up, right? It's as easy as bringing them

up, right? It's as easy as bringing them into the editor and then you get a midjourney canvas that you can highlight the problems that don't work and fix them all at once. And the cool thing is

now while you do that, Midjourney has a lot more context built into this image about what your character should look like and all these costume pieces should look like. So your character is going to

look like. So your character is going to stay much more consistent while it fixes those things you erased. So the point of this fourth prompt is to teach my GPT to

write the prompts you need for this situation for whenever you have a sidebyside grid of images like this and you want midjourney to make them all more refined and all more consistent. So

to use this one just upload the stitched image and submit the prompt. If you have specific edits you're planning to make, it also works to explain those here at the bottom, just like the last prompt, but this time you're making those edits

across all of the characters in a line.

And it doesn't have to be fixing something. Let's say your character gets

something. Let's say your character gets injured in your story, and you want a few of your assets to have a bruise, but you want that bruise to look the same in a bunch of the images. You could write a

prompt to edit a grid like this and add that bruise in the same way across the entire stitched image. You could do the same thing to change something about the character's costume or whatever your edit might be. So, prompt one,

generating references. Prompt two,

generating references. Prompt two, generating videos. Prompt three, editing

generating videos. Prompt three, editing single images, and prompt four is making multiple edits all at once to fix consistency. And I think at this point,

consistency. And I think at this point, you have a pretty good idea about how the method comes together. But I don't want to get ahead of ourselves. There

are a few essential things you're going to need to know to follow this method and build the exact scenes you want. To

start quickly, I want to go over how to split up these template images into starting frames. So, after you take the

starting frames. So, after you take the result from that first prompt into midjourney, by the way, I tend to generate them in the landscape aspect ratio 3x two so that both references fit

well. And from what you get, open the

well. And from what you get, open the reference you want to use for creating assets and click edit. At this point, we are splitting up this image into squares that are wellframed for AI video. And

what I'm showing you is the fastest way to do that. Drag one of the handles on the left to crop the image so it's skinnier than a square. That makes it so when you click the 1x1 aspect ratio, the

image doesn't zoom out at all. Now, hold

the Alt key on your keyboard to turn your mouse from the eraser tool into the move tool. And then drag your subject to

move tool. And then drag your subject to the center for either the close-up or full body asset. Now, let go of alt. You

have the eraser tool again. Scroll in a bit to make it bigger and erase anything that's visible from the other side of the image. Click submit and then we're

the image. Click submit and then we're going to hold alt again and move the image over back to the other side.

Switch from alt to shift to enable the restore tool to paint your other character back in. And then let go of shift to erase the other side out if you need to, if it's in your frame at all.

And now you can click generate again.

And back on the create tab, these are finishing up. And when they're done, you

finishing up. And when they're done, you have two starting frames that are ready to animate. I really like putting them

to animate. I really like putting them on exactly a square aspect ratio just because I think midjourney does a better job with video at one:1 width and height. And in general, there tends to

height. And in general, there tends to be a good amount of space for the characters to move around. And when we grab the assets, it helps that nothing bleeds over the edge. So, we have nothing that we have to regenerate. So,

now we have frames that are ready to animate. I also want to give you a

animate. I also want to give you a time-saving tip on generating the videos themselves. If you've used my GPT to get

themselves. If you've used my GPT to get your video prompts for one of these cropped images as a starting frame, normally you might generate a new video from here and then clicking under manual

low motion and then the prompt comes up and you can replace it and then paste in your new one and then once you generate it, the video as a starting frame is gone and now you have to go find your starting frame again and then do all

those same steps one more time. And it's

so easy to hit the wrong button. What I

want you to do instead to speed this up is just drag the image onto one of these other boxes first, right into the prompt box. You can do any of them. I'll use

box. You can do any of them. I'll use

style reference right now. I'm going to have you click the lock button. And now

you can drag it into this starting frame space. We just made the starting frame

space. We just made the starting frame sticky. So now you can go back and forth

sticky. So now you can go back and forth to chat GPT several times and generate multiple prompts and the starting frame stays the same every time. This just

makes this way more streamlined. If you

want to use a new starting frame for a new prompt, just drag in that new image and the lock button stays locked. Of

course, if you ever want to stop using that image or stop using any starting frame at all, just hover over it and you'll see this X button and that makes it go away. But this saves me a ton of time when I'm generating multiple videos. Next, once you've generated

videos. Next, once you've generated videos, you need to know how to pull still frames from them so you can turn them into assets. If you've never done this before, you might just pause the video from within MidJourney and take a

screenshot wherever it looks good. But

this will end up turning it compressed and the screenshots are going to be hard to work with. So instead, what I'm going to have you do is click download raw video. Assuming you're using a browser

video. Assuming you're using a browser like Chrome, you can just drag the video right back in as a new tab right from your downloads. It opens a built-in

your downloads. It opens a built-in video player to your browser where you can pause it and then drag the timeline around to pick any frames that you like.

When you found one, just rightclick and hit save video frame as. Just so you know, while you're dragging, you can also use your arrow keys to move exactly one frame at a time and really narrow in

on exactly the frame you're planning on downloading. And you can download as

downloading. And you can download as many of these frames as you like. If

you're not using Chrome, most video players will have a similar snapshot tool and you'll also be able to move frame by frame to narrow in on a particular image. So, if you use

particular image. So, if you use something else, I would just look it up for whatever tool it is that you look at videos with. So, at this point, you have

videos with. So, at this point, you have everything you need to turn references into characters you can direct and move around and then grab the stills from those videos and use them as assets. One

thing I'm going to have you do to make it way easier to work with them now in Midjourney is to remove the background first. You could do this in Midourney

first. You could do this in Midourney itself. You do that by using smart

itself. You do that by using smart select, but it's slow and rough and it's not a super accurate background remover.

So, I have a recommendation that's going to make those lines crisper and actually select your character more accurately.

In my opinion, the best background remover by far is the cloud select subject in Photoshop. But if you don't have Photoshop, you still able to use this for free. Adobe hosts a little

utility on this web page where you're able to just upload an image and remove the background from it. Adobe is a little bit painful, so if you're removing backgrounds more than once in a row, it'll probably force you to make an account to keep going. It's still a free

account, so it's not that big a deal.

But when you're on this web page, just drag an image into this box. And then

when it's done, you'll see the background is removed. Click download,

and it will download the image. It also

tries to pull up this Adobe Express thing. You don't need that. You can just

thing. You don't need that. You can just leave. If it opens a new tab, you can

leave. If it opens a new tab, you can close it. Otherwise, just hit the back

close it. Otherwise, just hit the back button in your browser. Again, Adobe is a little bit painful. It's a bit of a hassle, but I promise the quality of this background removal is worth the effort. Though, if you have and know how

effort. Though, if you have and know how to use Photoshop, you probably want to just do this in Photoshop instead. It's

going to be faster and less painful. If

you do this after an upscale, you have really highquality assets that you're ready to build scenes out of. The

feature in MidJourney that lets you build these compositions is the layer feature in the editor. To use this, you want to make sure you're in the edit tab in MidJourney. If you go there directly,

in MidJourney. If you go there directly, you can upload any image or if you want to work on something you've generated.

You can find it in midjourney, click edit, and then the button open and edit tab on the right hand side. Once you're

here, you'll see the layers on the bottom left. Starts with just the image

bottom left. Starts with just the image you're working with, but you can drag any of your saved images from your computer and they appear here as a new layer. Because we cut out those

layer. Because we cut out those characters and they're surrounded by transparency, we're able to drag them right into the scene and put them wherever you want. Every layer works

like a mini version of the entire editor. If you remember, you can hold

editor. If you remember, you can hold alt or click move and resize to grab your layer and scale it. And if while alt is held, you hold your mouse near the corner, you can rotate the entire

thing. This works so great for putting

thing. This works so great for putting your assets exactly where you want them.

It's important to understand that smart select and the brush, all of these tools work on one layer at a time. And so I'm going to recommend that if you're trying to erase anything, you'll save yourself

a huge headache if you move the layer you're trying to erase from to the top of all the layers. So whenever you're actively doing something like that, it wants to be on the top and not just the selected one. Because clicking on a

selected one. Because clicking on a layer selects it, and the one on top is really easy to accidentally click on.

But now building scenes really feels like clicking and dragging. You can add as many characters as you want. You can

put some in front and some behind. The

more assets you've already made, the more fun this process of building scenes is. If you have an image you want to

is. If you have an image you want to split up and use different pieces of it in different parts of the image, the best thing to do is to add that image in twice and erase the parts you're not using in each one. And then you can put

both halves wherever you want them.

Super handy. I also recommend even for full body scenes using the close-up asset as the expression and use it in place of the head. That way you get the highest quality expressions, the ones

you've controlled from that early video generation phase. If you've already

generation phase. If you've already generated the background you're using, you'll only need to erase tiny little bits. Things like where shadows or

bits. Things like where shadows or lighting effects go to call a scene like this that you're arranging good enough.

And when you're doing it this way, having your assets on transparency is a huge timesaver and it takes very little additional generation to make the scenes

really great. But if you do need to

really great. But if you do need to generate a background for your scene and you're trying to do it in the editor, midjourney isn't going to play well with the partial transparency. But sometimes

it's what you got to do. So, my tip for when you are just arranging characters and you want to generate a whole background, I'd recommend erasing all this gray stuff and all of the character to make this as clean as possible. And

this is how you'll get the editor to do a better job. And only when your background looks really great, drag that saved composition back into the scene.

And now you don't have any of this mess that comes from the transparent background. I still think it's way

background. I still think it's way better to take bits of scenes that you want to use. You could take some floor from one and some foreground from another and a cool background element from a third scene and piece those

together all behind your character so that midjourney is filling in as little as possible. When the gaps are smaller

as possible. When the gaps are smaller and you're only trying to make little adjustments rather than generating a whole thing in the editor, especially if you use my prompt three to refine the editing prompt, it can turn out really

beautiful. But don't rely on midjourney

beautiful. But don't rely on midjourney filling out too too much while the scene is so complex that it doesn't truly understand it. But yeah, that's the path

understand it. But yeah, that's the path to getting characters and scenes, poses, expressions foregrounds and backgrounds all under control. I feel

like that's all of the essentials you need to know. And now we're on to kind of the fun stuff, right? What do you do if the scene you're building is almost perfect, but it feels disjointed? Let's

say you've put your character into a scene, but even though the poses are much more dynamic than Omniref can achieve, the characters don't really feel like they're a part of the world.

Maybe you want them interacting with stuff, using an object, eating something, maybe fighting another character, or just hugging them. You

want everything to feel like it's in the same place. And at this point, what you

same place. And at this point, what you need is what I call second order assets.

When the goal is a little bit too complex or too specific for one video to go straight from a basic pose directly into what you need for a scene, the

strategy is to add additional context into the scene you're building and then generating a second video that can get everything to settle into a more natural place. I'll give you an example. Let's

place. I'll give you an example. Let's

say you've generated a cool pink hat in this style for this character to use as kind of a prop. There is no composition you could build with this hat asset and

character asset to make it look like he's playing with the hat naturally. So

instead, what you do is you build a simple scene with this character and this hat together, not worrying about it feeling right. And if you're noticing

feeling right. And if you're noticing your scene is disjointed, this is probably where you're at. You still want both the hat and the person to be as prominent as possible, but other than that, you don't have to worry too much.

Essentially, you just need room for our star to grab the hat, spin it around a bit, and put it on their head. The idea,

like always, is that all the motion we're going to ask our GPT for is going to fit into the frame when we turn this into a new video. So, it's a matter of giving the result from our GPT to

MidJourney with our new composition as a starting frame. And suddenly, if we

starting frame. And suddenly, if we generate this video, we have a collection of all sorts of ways our hero could be interacting with this pink

fuzzy hat in this scene. Right? You're a

director. If you need him to do something new, give your virtual AI actor a new order. So, whenever your scene isn't perfect, you get to give orders to your actor as well as your

camera operator at the same time to let the video build your scene into something that works a lot better. And

now, instead of starting from that straightforward expression, you're starting from a much more interesting pose and a scene that has a lot more context and a lot more ways the character can interact. And with my

prompt, my GPT will write something that lets you stack motion onto other motion.

And as you do that, it doesn't take long to get exactly the pose and angle you're after in video form that you can pull as

a still and then work on from there. But

even with all this complexity, it's not going to become perfect immediately. We

should use video for what video is good at and image processing for what image models are good at. Even after building out compositions from improved and upscaled assets like I showed you for

that final prompt, once you have scenes, there are naturally going to be adjustments you need to make. And for

that, the editor is going to be your friend. That editor prompt I shared with

friend. That editor prompt I shared with you, prompt three, will take you a long way as far as making those adjustments.

It's a matter of erasing the things you want to change, using layers to add assets in, and videos to change those assets. I hope so far at this point I've

assets. I hope so far at this point I've been able to demonstrate how powerful this method and approach is. But

depending on how far you go with your project, there is one place midjourney won't take you on its own. And that

comes down to that pesky issue of prominence and context. Right? Video is

a little messy. And even with my prompt, the editor is less effective when it comes to those smaller details. You can

fix a lot of the assets early on and all at once, just like I showed you with these sidebyside grids, but there is always going to be parts of your scenes that were never assets, and the assets

you do have will need to blend into the environment you put them in. This

doesn't happen automatically. And this

is another way the custom tool I built is going to help you get over the finish line. If you don't know, Imager Rag is

line. If you don't know, Imager Rag is the online tool I built to help people manage context and make detailed edits within AI image generators like

Midjourney. First, it makes it a whole

Midjourney. First, it makes it a whole lot easier to build those sidebyside grids I was showing you earlier on in this video. Literally, just upload your

this video. Literally, just upload your collection of images, then drag a rectangle around the parts of each one that you want in the grid. You switch

between your images over here on the right. Then hold shift and you can

right. Then hold shift and you can select the ones you want to put together and create a collection out of. That's

this button here. This collection is perfect for making a bunch of edits all at once early on. And yeah, like I was showing you, you can improve like five different versions of your character all

at once and upscale them all. I should

say five different versions of any asset. But the real magic is when you're

asset. But the real magic is when you're getting close to the end of your project and your scenes are almost done. If you

followed this whole process to build out your story and there are a few final adjustments you realized AI isn't very good at yet, this concept of a collection will get you out of a very

tough spot. I have to show you because

tough spot. I have to show you because not only does my tool let you build a collection out of parts of any image and take it to midjourney for changes, but it keeps track of how those parts of the

image fit back into the originals.

Because then if you drag that image back into the collection, it was like you made five individual fixes to five different images all with one midjourney

generation. Except the difference was

generation. Except the difference was this way. All of the different images

this way. All of the different images are consistent with each other. I don't

have to tell you how useful this is.

Obviously, being able to make all sorts of subtle adjustments to different things about your character across different parts of your entire project all at once is amazing. I've been

playing this little montage of a tattoo that wasn't sharp and clear enough after all those videos I generated and the upscales I did. It just lost fidelity.

But because I was able to zoom in on it and using the fourth prompt I shared with you to describe it in detail, MidJourney was able to reapply it to

this character's face in all five places at once. And normally erasing anything

at once. And normally erasing anything from the face ruins consistency, but midjourney had enough context to get the face right. Let me slow down a bit. Like

face right. Let me slow down a bit. Like

take these images here. When I finished building out the scenes, there were a bunch of places where the magnifying glass my little fox detective guy is holding just doesn't have the right background in the glass bit. It must

have just gotten stuck with whatever was behind the original reference video I generated. So, without image rag, what

generated. So, without image rag, what I'd have to try to do is take this whole scene to the editor. But, we have a problem. If you try to erase the thing

problem. If you try to erase the thing that's wrong with it, Midourney just fills the whole magnifying glass in. It

doesn't know that in every other scene it had this sort of half-transparent blue shade. and tint to it. But with

blue shade. and tint to it. But with

image rag, I can upload all of these scenes together and they're otherwise finished. But by selecting the

finished. But by selecting the magnifying glasses, I can set up a grid for midjourney to address this problem across my entire story all at once. So

after you've made your selections, you can hold shift to click on all of the images, create a collection, and then I hit download and lock. This is just how

you tell my tool that you're about to make changes. It's locked because while

make changes. It's locked because while you're editing it, my tool protects you from accidentally changing the shape and size of your selection so that it fits back in perfectly later on. I'll show

you. I send this grid to my GPT first, get an optimized prompt, and then take it to the editor. Now, I can safely erase the centers of the magnifying glasses I need to change. This one that just has his clothes behind it, there

was no problem with, so I'm just going to leave it unerased. And this is how Midjourney knows within the context of the image itself what the ground truth of the magnifying glass should look like. You can see now that it's only

like. You can see now that it's only worrying about this one thing and it has all of the right context. Midjourney has

no trouble fixing every single magnifying glass all at once. Now in

image rag, you're able to check in that fixed image. You just drag it into this

fixed image. You just drag it into this check-in box. You're able to preview the

check-in box. You're able to preview the change for the entire grid. And also

before you click accept, you can preview the final images one at a time and just make sure that each one is perfect. What

an amazing result for these images.

Sure, you could achieve this same kind of stitched effect using a photo editor like Photoshop, but it is a lot of work to keep track of, and Image Rag makes it

a lot easier. If you're interested in using my tool, I've got a great deal going on right now. For the first 500 users who sign up for the annual version of my premium plan, I'm cutting the

price in half. That's 50% off on top of the annual discount you'd normally get.

This is my early bird special. And if

you claim that discounted rate now, I am letting you lock in that price forever.

But once 500 people take the deal, it's gone and it's never coming back. If

you're curious about how effective the tool is at making revisions and fixes to your midjourney images, especially as they get more complex, I highly

recommend you check out this video next.

Not only is it a great overview of how powerful controlling prominence can be, but it also gives you step-by-step guidance on using my tool to level up your images in a huge way. Definitely

worth a watch. All right, thanks so much for sticking to the end. I hope you learned something and I will see you in the next one.

Loading...

Loading video analysis...