Transcription provided by Huntsville AI Transcribe
So welcome to Huntsville AI. Tonight we have David walking us through mid-journey.
David, I will let you talk a little bit more about your background a little bit, but David knows a lot about mid-journey.
So a great person to have around.
for me it feels like I’m back in the my normal way to explain it is arguing with a teenager that has different opinions and I’m trying to talk these models into doing what I’m trying to get and wind up with the whole then you find the secret you get on okay if I say it this way I get what I need so with that we will turn it back over to David thanks for thanks for doing the thing All right, so what’s playing is midjourney.tv. This is non-stop 24 hours a day now. Videos that different people have generated. You can see the names down the bottom left.
They just stream this non-stop now, so they have a video. They do a little filtering to make sure nothing too bad. It’s out here, but it’s kind of amazing just seeing a little bit of the difference. styles you can get.
And I promise it’s a lot smoother when it’s not going through three different connections to the room soon.
When I started making the presentation, there was way more info than I thought there was going to be.
So I’m going to try to keep it not painful and just give like a little touch to every section. And I’ll put out this PDF on our Discord, just if anybody wants to reference what’s there. First thing, just going to go over signing up. It is a paid service. There’s currently not a way to really use it for free. They do different promotions. Cheapest you can get is about $10 a month, about what you start exploring. If you’re using this for work, I highly recommend going for the pro planner involved. $16 a month, but what that starts giving you is what’s called stealth mode.
prevents your images from just being shared online to where everybody can see it. So you can tell, you know, it’s still going to be accessed by their model, but they have so many millions of images. It’s not like they’re going to be looking at everything.
But if you have anything less than promo, your images will automatically be shared to the gallery.
Anybody can find them online.
So if you are trying to develop concept art for presentation, it’s gorgeous.
Word or make you look like Jay was talking about earlier.
You’ll probably want to go for one of the higher tiers The journey is completely private.
So they do some cool promotions from time to time like over the holidays and around New Year’s they did I think they called it relaxathon Anybody with the subscription can do free generations really fast as much as you wanted non-stop Just because they had extra servers that they weren’t using and they’re like, well, we’re not going to waste them. So let’s just give them to the customers to use.
Sign up. Just go to midjourney.com. There’s about a million copycads out there that try to get you to click on their site.
So midjourney.com is the best place.
They also have a mid-journey discord.
You can sign up on either one.
You can also use the model on Google or Discord.
Sorry, through Google, sign up. I’m just going to go over the web UI. Discord UI is starting to fall out of favor, and it’s a lot more manual commands. But if you get into mid-journey and start using it, like, there’s a trade-off. It’s more powerful, but it’s more technical.
So what Jay was talking about, arguing with the teenager. So a lot of people ask me, how do I make this exact thing that I have in my mind Exactly like I picture it technically correct down to the pixel.
And that’s not what mentoring is for.
You just want to kind of experiment with it, you know, throw out an idea, see what it gives you back and just start working with it like somebody that you’re collaborating with. So speaking of, while we go through this, I’m just going to do some examples. So what is something we should prompt?
You know, idea.
Raccoon on a rock. Okay. And don’t worry, I’ll go for all of those little bells and whistles here in a second. We’re going to start off with the base model.
That would be more interesting to look at my slides.
When we’re coming up with the prompt structure, I like starting how Josh just did.
I like starting really simple and building up from there.
Like you have just a subject and something about it.
If you want to get technical, there’s a mid-journey guide that breaks it down really well.
You can lay out the subject.
What are you talking about? A raccoon?
The medium, do we want it to be a painting of a raccoon or a photo of a raccoon? The environment, I assume in space, or it might be like Toyota Field.
Yes, it’s wrecking on a rocket.
What kind of lighting, what color, what mood you want to have. The framing.
So mid-journey will work with all of these aspects all at once.
You can give it some really long in-depth prompts, but I really recommend starting out simple and then just add to it bit by bit and see how it changes.
So… We have our raccoon on a rocket. It automatically generates four images based on whatever you prompted.
And so this is kind of a cartoony style.
So one cool thing in the web UI is you can click on anything in it basically and add it to your prompt.
So your prompt window is up here.
What will you imagine?
You’ll see I can just click on the text Raccoon on a Rocket to automatically add what I had previously up there. So we’re going to tell it now. So get a little fancy editorial style magazine photo of a raccoon on a rocket. And this is just the, you’ll see a little V7 here. If you guys can see it, this is just the base current model of mid-journey.
This is like company headshots.
If you had a copy full of records, and this was your board of directors. You’re probably seen in all public AI models.
You can help improve it by giving feedback anytime you generate any image.
You can just tell it, hey, I like this one.
I don’t like this one. I’m neutral.
But you don’t really have to give feedback if you don’t want to. Any time you do like an image though, it will get added to an easy to find place where you can look at all your liked images and quickly find them. So let’s see, start small.
Let’s go next.
So I think the most powerful thing in mid-journey is something called personalization and mood boards. So personalization, you can go in and start ranking images side by side, picking out which one you like, and it’ll start making a profile that customizes the base model based on what you choose that you like. And show an example.
It’ll just give you two images to pick between, and you’ll notice how different the images are.
Supposedly, it compares every element of it.
I’m going to go ahead and tell you guys, when you make a personalization, if you start using it, think in your mind what style you want and just skip anything that doesn’t fit that. Skipping an image does not influence your personalization at all. And so you can just keep going until you find something. Let’s say I was going for realistic fantasy or your personalization. You know, I’d go until I saw one like the one on the left and I’d make that.
And hey, I get lucky and I get number two for the one on the right and number one.
But then I’d have to start skipping again until I got back.
But you’ll find if you stick to just what you want, your personalization will come out a lot better. mid-journey really likes to push, hey, rate everything and it’ll eventually learn what you like. But if you start working with the super users, there are people that have 15,000 ratings and their personalizations just junk.
Storyboard that I would do, or is it like a global version?
Yeah, so you’ll see I have a whole list of personalized profiles down here.
And then if we go back to the prompt window, so I’m going to just use the Raccoon photo again. There’s a personalized button, and I can choose any personalization I’ve made, and then apply that. And you can apply multiple personalizations as long as they’re from the same one.
So I know, let’s see, I know this one, ignore my horrible naming conventions thing that it says to me.
So all I’m going to do is apply these two personalizations to the same prompt we had last time.
And it should give a little bit more realistic photo based on the ones I picked.
I’m also going to choose another. Yeah, I was going to ask if you can have multiple personalities, but Jack phrased it much better. As you’ll see, you know, here were the four just based mid-journey model, but when you applied my crystallizations, it’s just, you know, a little bit different style and then I have one I call abstract representation and you see how it changes the nice model based on that.
Mootboards are even more powerful.
So a mootboard, all you have to give it is a few images and a style you like and Mid Journey will mimic that style for whatever you’re making with that mootboard. So for an example, let’s look up. Once an artistic style or a movie that we should base our record on, something distinct to the painter photography style, you will also notice that when we’re working in here now It currently does not care about IP.
It is up to you.
Create things and put it out there and you can get into trouble for it if you decide to mimic somebody and try to make money off it. I think Disney currently has a lawsuit against mid-journey in general.
So the argument falls down to can you sue a box of crayons because you can drop Mickey Mouse with it.
That was what mid-journey doing different than it. I’m not going to get into the argument here. So we’re just going to save this image and this image. And then we want to find one that’s way different.
So you have a report tab.
You just create a new mood board.
Drag and drop whatever image is into the mood board that you want. I’m going to go ahead and rename this. And you’ll see I now have a HSVAI mood board down here just by selecting it with personalization on. Yeah, which is using this prompt. Now this one might be a little weird because the style of Wallace and Gromit you could tell was not a photograph and your prompts and your styles can conflict sometimes. So telling it, that’s an editorial style photo, while also telling it, hey, be this Wallace and Gromit style. It’s sometimes either one. That is a very happy record.
Is that the mood or the style? Claymation?
Yeah, so what we’ll do is also try just taking out the editorial style photo and just saying raccoon on a rocket and see how it does. So when do you use mood board versus the personalization? Are they interchangeable or do they have different uses?
So you’ll start seeing how everything kind of can start working together.
So I might explore in mid-journey and just see what kind of style I come up with. And then if I find something I really like, I’ll generate a few images of that style, add it to a mood board, and now I can replicate that style without the wrong. you can combine mood boards and personalizations.
So everything is another layer of customization.
You can have 20 performance and personalizations all applied.
Just see how many of them are added.
The mood boards too, they don’t just do style, they’ll do the concepts and things like that.
So like you’re seeing on here, it’s doing the claymation, but it’s also kind of doing the expression they were doing above the photos of the, ah, that sort of thing. So it kind of just, general cross-reference between all of it. Now, I highly recommend it, unless you like to be very experimental, keep all your styles in a new board similar.
Because rather than it combining everything into a style for you, it more kind of picks chunks of it to apply to each image.
So if you have very different styles in a new board, you’ll get varied templates when you apply it.
So if we add this little I don’t know what it’ll do to it, but we’re just gonna edit our main board.
And do the same prompt again.
That’s running.
So one of the things about this display will be, like we were saying at the beginning, mid-journey is very experimental.
Like, even if you think you understand how every single thing in your product will work together, sometimes you’ll get things off the wall.
So it’s good just to have an open mind and keep kind of playing with it. So, similar to the base model and the personalization of the mood board.
There’s a whole lot of parameters that you can change to affect how the images work.
So we’re going to go back to our basic raccoon on a rocket here.
And I’ll put this in Discord, but I think the best way will just be go through it quickly and then cash show some examples. So you can’t set the pixel. count of an image.
You just set the aspect ratio.
So you can set it to portrait mode, square, and ceramic mode, and it’s just a ratio.
You can set it 20 times as wide as it is tall if you want.
You can set it, you know, just come up with numbers 171 by 233, and it’ll figure it out and make it that ratio.
But you can set the pixel count.
It’s always about the same size of an image.
The journey does have an upscaler, it’s so-so.
But aspect ratio is key.
One of the main notes from any image generator, Josh, I don’t know if you’ve seen this differently before once, but if you want to make a tall subject, put it in a tall aspect ratio.
Like if you want to make a skyscraper, make it a tall aspect ratio and it’ll have an easier time generating the skyscraper. If you set it a panoramic view and say skyscraper, it might give you the first three stories and not give you like a long out view.
If you want a wide landscape to a panoramic aspect ratio. So the first main setting is stylized.
So you’ll see it have all of these dash dash. This is kind of a holdover from Discord.
This is how you can manually add into your prompt parameters and then journey. So under this tab, you’ll see stylization here.
So we can just have all these sliders.
We can adjust it here or typing dash dash s 200 is the same as going in here and setting it to 200.
It’s just alternate ways to do it, however you’re comfortable with. Stylization, the best way I can describe it is it’s how much of the base model or your personalization is applied to what you’re working on. So 100 is just kind of default, 1000 gives it a lot more of whatever it is.
That’s when I’m curious if you have a different definition or not.
Uh, yeah, it’d be equivalent to like guidance, so classifier free guidance.
So it helps it, if you’re trying to train it to do something else with the post training sort of stuff, it pushes it more towards that.
Yeah. It might ignore your problem. But if you have a personalization you love and you want to make sure that it’s a hundred percent your personalization, whatever that may be, rather than mid-journeys based model, crank the S up to a thousand. as much your personization as it can be. So I’m going to say another most powerful thing in mid-journey are the style references. So you can add any image rather than making a mood board. You can just drag an image into mid-journey and tell it copy the style of this image.
So going back here.
I’m just going to leave it raccoon on a rocket and drag in all of some grommet and make it a style reference. And in the same way you can do S for the style, you can do SW for the style weight. And so if we tell it SW1000, it’s going to say, hey, I see this image you gave.
I’m going to put everything I can of this image into what you’re generating.
I can’t see if I finally added two raccoons. You can smile when away. Right, because it’s not using the mood board. That’s what I was saying, the concept. He was holding on to that. The style is just style. Yeah. So be, you know, whatever medium it’s in, the colors, all of that.
Now, one of the most fun to explore in mid-journey is SRF. You can also put in member codes and you can put in random. for the Zatting.
And mid-journey on the spot will generate a random style based off of some of its trained data.
It’s not a pre-existing thing.
The numbers go up to millions, I think. So we’re just going to do Raccoon on the Rock and SRF for style reference random.
And we’re going to tell it to do three more of this.
So you’ll see each one has a different SRF number.
I just learned something.
Yeah, you can tell it to do up to however many your plan will let you generate at a time. So I can do up to 30 revisions simultaneously.
So you see each one of these has a very distinct style.
And now from now on, if we want to copy, let’s say this top one here, which is stink cubed, we can just click on this SRF and use that for anything.
So we’ll just put something else we should generate.
What you got?
A futuristic turtle floating on Mars.
A futuristic turtle.
I like that.
All right. So I can’t remember the command right now, but while we’re waiting for this, if you see the percentage that it’s giving you, like it’s 43% complete, you can put in a command to force it to stop any percentage.
so you can tell it to stop at 80% generated.
And sometimes doing things like 95% generated can give a really cool effect.
You know, make photographs look more realistic because there’s a little fault in there, a little bit of blur to it.
Yeah, we got a futuristic turtle.
I think I like the turtles you generated a little more. But yeah, copying the SREF lets you copy over style to anything.
And something else I’m gonna, I think the link’s in the email.
But there are plenty of sites that do SRF mining. People will put in random and they’ll just experiment and then they’ll share it.
Most sites online, people actually try to charge for that, do a whole subscription service. This is run by a couple of the mods on mid-journey that are really nice. And what you can do is it’s always updated.
You can just scroll through until something catches your eye.
And once you like a style, just click on info.
and it’ll give you the SRF and you can just start and see whatever prompt they did to make it or you can just copy the SRF and use it on your own.
I like just experimenting and putting in random a bunch of times and seeing what it generates.
As I said, a lot of the fun and maturity is just going with the flow of it.
So a standard practice is you have to put in something for the prompt. But anything will influence it.
If you put in a smiley face emoji, it will influence it to make a smiley face. So if you want to get the base idea of what AstraF is, usually a period is the best thing.
And doing that, you can also see what sort of… like your timeout from reports with SRF you can cast see what it’s weighted towards.
Like, some things might, if you want to generate portraits, some styles will tend more towards guys, some towards girls.
This one apparently, uh, superhero. Hi, sorry.
What’s for a teen type?
So what I like to do after I do a few, let’s see which one grabs our attention the most.
You just lose hours at night just doing this?
Oh yeah. I was gonna ask you know that when we looked at client and had like the dollar sign at the top that kind of tells you how much I would expect to see like how many hours I have been at this along with the notice to get back to work somewhere.
And the cool thing is like tonight I’m using you’ll see I have fast mode selected so I’m actually using the GPU hours that I’m paying for but if you’re at each pro plan gives you a relaxed mode. So I can set it on Relax Generations, and it doesn’t use that medium of my compute time. It’ll take a few seconds longer to generate, but I can generate up to 12 prompts at a time in relax mode.
So if I’m doing work at night or watching a movie, I’ll just put through 12 prompts, let it run for a couple minutes, come back, look at it, revise some.
And I think it let me generate something like 1200 videos in the first few days without using any compute time.
So after doing some random ones, if something catches my eye that I really like, we’ll just stick with the high school team type.
So let’s say I want to use this again in the future.
I just keep the SREF and put in school.
And I’ll put in whatever it made me think of, just so I have a generation of it with the SRF.
And then in the future, I can always go back through my organized folder and see, oh, that’s what I was thinking when I saw that. Just an easy way to keep ideas in mind.
And I forgot I lifted on relax. So let’s do that again. You probably saw there’s also a turbo mode.
I don’t really see a me and you see how fast, fast generation is, but Turbo really does crank things out and just uses a good deal of processing time.
All right, so now sometime in the future, I’ll look back and see, okay, for some reason I have a high school team Titan’s.
I’m trying to prove you wrong on the girl guy thing. Yeah, well some are good for both, but I found some S-traps that tend to one or the other. Yes, so a lot of times I’ll use purse as a base prompt.
We’ll see one of these will probably tend towards, you know, three out of four or four out of four will either be women or men.
And you’ll see it tends towards age groups, like some S-traps might prefer to create old people, some will refer kids. Is there any like meeting behind the SRM number in relation to this human model?
Not that anybody has ever found out.
There does seem to be some similarities between really close numbers, but even ones that are one number off can be very different.
So, unless you’ve seen anything like that.
Uh, no, not the actual numbers.
The things are, it’s basically, uh, cause Firefly, I don’t know if you know Firefly, but they, they have one of these F2R sites and they, they mapped it out and they’re definitely like in the similar spaces sort of thing.
So it’s definitely a data distribution thing where they have something behind that.
It’s all, you know, nobody actually knows.
Yeah, I was just curious, can that something to do with like that initial noise generation for starting the prompts?
If that was tied to it in some way, if it was either like an Ashlet that or something.
Like a thing, you would see some similarities between things, but the opposite goes into the user model.
It’s going to be all of the things.
So on the third one, sorry, the one that was on the top, that SRF, could you ask it for a doctor with that SRF? Yeah. Let’s see how skewed it actually is.
So let’s just do a doctor. It’s also dry hospital. And right now I’m just doing it where the S and the SW are both 100.
It has kind of equal weight.
So now I’m going to tell it to go full out the style weight. OK, so it looks like it decides that doctors are in this. There’s other I mean even even way back to the early bias discussions.
We had all kinds of different ways that they pick up on You know Things that you didn’t intend So majority does a lot of interesting polls Participate in the meeting of those so they do demographic surveys. They do style surveys. They’re really trying to decide who likes what So one problem they had is if they just put what was actually most popular on their explore page, it would be beautiful women. It’s a lot of people that come to that, you know, they’re the ones that are rating it. But they don’t want their explore page to show just one thing they want to show what it can do.
So they’re starting to look into, okay, what is a 30 year old woman from Japan?
What aesthetic do they generally prefer? compared to a 50-year-old man in India, and they’re looking for very widespread, like, how do you think… I don’t know what they’re doing with it yet. They keep promising to put out a research paper, but… Like you said, the bias is a question. I have a quick question. Yes, sir.
Can we change one particular picture?
Let’s see.
Can we change the color of one particular image?
keep everything else the same. Okay, so if you don’t mind, let me get to that in just one second. So there’s a feature called Edit, and there’s one called Retexture, and that’ll do just what you’re talking about.
So we can, you know, I’ll go ahead, there’s multiple ways to do everything in their journey.
So we got through SRF’s style weights.
Chaos is kind of self-explanatory.
Chaos is a setting and it basically just adds random variations to the images that you get. That’s the best way I can describe it.
Yeah, it actually, it looks at each image and it forces them to be a distance away from each other. So instead of injecting randomness, it forces distance. Yeah.
It is interesting. Yeah, so it basically guarantees the four images you generate are going to be more different than if there wasn’t a chaos in it.
Weird.
There’s another setting, which is pretty close to how it sounds.
Like I found the more you turn up weirdness, the more unusual elements get added.
Temperature, that’s what I think. No, that’s a good comparison.
Yeah, temperature setting. If you wanted to follow more of the prompt, you can tell it raw. This is definitely true on video.
Not so much on regular, but if you have a very descriptive prompt you write out you want to try to make it to follow it or adding raw sometimes helps The new exp setting is sort of the same as raw So you can do dash dash exp and set in a value from zero to a hundred and if you said to a hundred it basically tells it really really followed this prompt when I’m trying to get out of it I find adding an EXP of about 5 or 10 just gets better prompt coherence than having to work there. Like I said before, there’s a whole lot of settings that can kind of get you to the same place. It’s just all about playing and finding out what works best for you. Now, for a way to do changing a coat to blue, one of the tools we can use is OmniReference.
So we’re going to take, let’s say, this doctor.
we’re going to add it to this omni reference window.
Now we can set in a manual command or do it here.
You’ll see there’s an omni string.
If you can read it up there, it goes from one to one thousand.
The higher you set this, the more it’s going to try to keep the person in the image the same as what you have.
What OmniReference does is if you have like a prominent subject, a person, a car, whatever, it tries to take that subject into another image.
So if you turn it up to a thousand, it’s going to try to keep this doctor exactly as she is in here.
If you want to change it and give her a blue code, you know, we might want to go down to OmniReference 50.
So the lower you set it, the more variation to the prompt.
So we’re going to say, we’re going to talk to her very blue scrubs. And because it’s all about experimentation, we’re going to try it with a couple of settings. So if you’re trying to create characters, like consistent characters, would you have to use a thousand on that hobby way?
Really making consistent characters is still kind of hard, but Omni does help with that.
And I found you can keep them pretty similar, you know, even around a hundred or even less than sometimes.
So generally sort of the same look on these.
I definitely gave them a blue code instead, but it’s not exactly what you’re asking for.
I’m making it the same image with the blue code.
So, next up is editing and retexturing.
So, what we’re going to do in here is, let’s take the same image. And we’re going to choose edit.
So now we can highlight the area. By the way, it looks like there’s still having issues with the number of fingers, right?
Sometimes it’s getting better now, but about 50% of what will be off. the time. I use different features, it’ll be worse too.
So omni reference is really bad.
It’s a finger thing, but if you don’t use any of that, I haven’t seen it or noticed it in the normal model at all.
And I’ve even found I made one personalization where all I picked out were people that had five good fingers on both hands, clearly defined, and it was better at generating accurate hands.
So whenever you choose a personalization to focus on, it does influence it. My big thing that I’m looking for is propellers on planes of boats.
Trying to get four full propellers on a airplane that are 90 degrees above from each other is almost impossible.
So we’re just going to say we’re going to include scrubs.
Let’s put that in it. We’re going to try to force it to really do that. So you’ll see it changed the outfit they were wearing because it changed everything that I selected that it kept all the rest of the image the same. So you can change the clothes or whatever you want. Sometimes when I have a character who hands are wrong. I’ll just do an edit just on the hand and one of the four images will usually be correct. So something else that it lets you do is you can resize images in whatever way that you want to. And then we’ll try to fill in the surrounding, both based on the image that’s near it and whatever your prompt is. But if it’s too far different, it might not do it.
Like if I try to prompt it, as you know, more of this hanging over, like it might have trouble making that leap or it might mail it.
So we’ll see it automatically filled in more of the image just by expanding it.
That’s also something you can do here inside the web UI. We can choose just to pan the image a certain direction. And it’ll automatically change the aspect ratio, keep the original part and just start adding them and it’s on to the side. So a lot of this you can do in Photoshop and with Firefly, like I just mentioned. I found mid-journey just sticks with its own style really well.
whatever style you give it.
That’s really good.
And the key for me is you’ll see just doing this demo, how easy it is just to run through an idea from thing to thing and just kind of follow your imagination. So next up. We’re going to do something called re-texture. We’ll just use Wallace Ferrami. Use your own image or evalysis image. I’m going to set this to a realistic profile. Let’s say an editorial spiral.
photo of goofy little man in his dog.
And re-texture will basically take an entire image and can change it to another style. Now, how much it can change it? Oh, no.
That was on purpose. trying to take this one. Yeah, let’s say it gives retext your whole new meeting. Having a long in-depth prompt can be better for making an image sometimes.
But I found for re-texture, then keeping it simpler is better.
If you give it too many instructions, it can get confused on everything it’s trying to fit together.
But if you focus just on the subject and location, it allows it to work with what it sees.
Not horrible, but not great.
I think retexture still uses the old model.
Yeah, so it’s still used two years behind or something like that.
So there’s mixed results with retexture, but it should be updated.
Eventually they get around to updating each feature.
I think he said that’s what might wait till eight.
Yeah, just because they have so many engineers, so some of the features on some models don’t get pulled forward if they’re like, well, why don’t we just go that way?
So that is… I think it’s a lot of texture.
All good, Jay?
Yeah, just got focus input. So something you can find in several places is Describe.
We can drag and drop any image in here to describe the image.
And mid-journey will try to decide what it thinks a prompt should be to recreate that image.
So you can take an image off the internet, a photograph you’ve taken, whatever. You’ll see it gives you a prompt to try to recreate it. Just like on that prompt. I hope I still have my personalized. Command line switches.
you’ll notice sometimes it gets really weird wearing blue scrubs and a t-shirt with the text mama They are getting better about text. It’s not as good as GPT’s text.
The GPT text is pretty astounding now, how quickly it gets it right. But that’s something else you can use edit for. If you start getting text and it just messes up a couple of letters, just edit that little area and it’ll work out well. That’s pretty good. So using the same model, this is what it generated based off of the describe of this image.
Alright, if you guys could, if I keep plugging along, any questions?
Yeah, I really want to see some of the video stuff.
Cool. Alright, so, video is as easy as… Let’s go back to our raccoon. Let’s see. That’s the new feature, right, video?
Video’s been on a couple months now. So first thing we can do for Lacey is just click the animate button. And it is just going to try to determine based off of our prompt and what’s in the image, what it should make the record do for 5.2 seconds. While it’s working on that, let’s go back. Down here. We’ll see there’s option to animate manually with low motion or high motion.
So yeah, high motion is good. So low motion is, let’s say you have one character against background and you just want the character kind of looking around like you want one thing in your scene. That’s what low motion is good for.
If you want a dynamic scene with multiple things going, high motion is better for that. So we’re going to animate manually.
Let’s see, which of the raccoon do you?
Hold on, it’s running out of gas.
And we’re going to tell it raw because we want it just to try to follow what we say. So if we don’t put in raw, what I found is sort of like some other AI mid-journey will kind of add in some of its own text to the background to try to interpret what you want. But if we just want one simple thing, let’s say that simple thing and say raw and it’ll follow it better.
So another option is… So let’s go back to create.
Let’s take one of our other raccoons.
See, we like this one right now.
So we’re gonna make this the end frame.
So you can give a start an end frame.
where it’s guaranteed to have a single frame at the beginning is your start, a single frame at the end at the end, and how it gets between the two.
You can prompt it and just kind of have fun seeing.
I found morph is a pretty, pretty strong word.
This will say… Okay, I’m on the rocket.
We’re sent to you on quarantine. Alright, so our first video should be done. I promise it’s a lot smoother on my screen, but let’s uh… So again, this is just mid-journey deciding what it should make based on what we’ve prompted. And if you want a quick view of, when I say it’s a little smoother, I don’t know how well you can see, but it’s very smooth frame rate. You could do 30 FPS, something like that.
Yeah.
What’s up? Can you make the video longer? You can. That’s a great question.
So let’s say we want this to keep going as it is. There’s an extend video option and we can choose either to give a new instruction or just to let it keep going like it was and mid-term you decide what happens.
So I’m just going to click extend. High motion. and it’s going to add a few seconds to the video and again give four options for how it extends from there. You can extend up to about 21 seconds to a length for a video.
As it goes on in time, coherence drops some, it’ll start changing style, but that’s where the start and end frame can be so powerful is you can put in an end frame to kind of bring it back to the style that you want.
So here’s the one where we say the rocket runs out of gas, the rocket starts rolling out of gas.
There’s the one where we tell it turns into the cartoon version of itself.
What happens if you give it just like a completely different start?
I love playing with that.
It can be a lot of fun.
We’ll show one. You can also see as it’s working through each of the four.
84% complete, about every 12%.
It’ll give you like a frame and part of it so you can get my dear friends going. If you have one of the lower tiers and you want to make videos, you’ll burn through your hours fast. But there is a command you can use, which is batch size. So you can either put in the manual number You can do one two or four So if we do this one, it’s only got generate one option for the video and it takes one quarter the process of time down here you’ll see video that size one two or there is an HD resolution I believe the standard video is only 480 and then the HD is 20 HD takes a lot more time to generate. And honestly, I thought 480 would be painful, but I’ve got used to it. It looks decent.
So some of the command lines after doing what did show up in that filter.
So is that your dash dash like?
So that comes in where if you drag and drop an image, you’ll find how many reference there or style reference.
You can also do image prompt, which is… It tries to make it sort of like the image you drop down, like overall composition.
It tries to keep it the same.
So if you really want to copy something, you can just force it… using the same image prompt as the style reference as the omni reference.
And just tell it, I want to copy this composition style and character ball in this. All right, so.
Since it’s already seven, I’m going to go ahead and hit the end and then we can kind of play around some more. I want to show, I already showed you the SRF library and I’ll put that link out on discord.
So version seven of mid-journey is really good at understanding problems from current other AI.
So for example, I have a custom GBT mid-journey proctor that I made.
to where I can put in any character from any IP or any idea or subject matter.
It’ll generate four different prompts for me to use.
And also, to note, I didn’t want it to exactly copy a character.
I don’t want to make John Wick exactly. I want the essence of John Wick kind of thing. So what’s an idea for, you know, we could put in Pyramid or Superman or whatever, or put something that we want to try out. You try and protect IP by not making it exact.
No, I just I’m not looking to copy. I just I’m like, you know, what would the essence of this character be rather than me exactly copying it? How about the raccoon, but it’s Indiana Jones?
Cartoon, real life.
Let’s go cartoon.
Okay. Cartoon, raccoon.
Fix us a hat and a whip. jacket. So the way I have this created is it gives me four of them.
The first prompt it gives me just tries to match what I gave it. The second one tries to put out a real life. Third one tries to go crazy artistic. The fourth one goes for more of a cartoon. So I get the three main variations and then whatever I ask for.
So you could set up Mid Journey is very much against scripting and automation.
So I would not suggest automatically feeding anything into Mid Journey. But it’s pretty easy just to copy and paste. Yeah, I looked up if you could use an MCP server with this.
They said, don’t turn your account off if you try. Yeah. On the plan that you have, do you run out of credits or GPU or?
I could just as much as I use it, but since I use relax mode so much.
I always have fast hours left at the end.
Especially after like December, relax mode just became so fast that.
Now there are certain things I like to do that do eat up the hours, but I always try to make sure to have a few hours left so like. My plan resets in two days and only had five fast hours left. So I’m like, okay, I can’t use any more. So I have them available for this demo. But normally, like you said, I might have 10, 15 hours processing left at the end of the month.
Okay, that’s pretty good.
So here’s what it gave.
Here’s just trying to match what we said.
Here’s where it tries for realistic.
Here’s where it tried to go crazy artistic. And here’s where it tried for cartoon.
And I also have it because obviously the current models, oh, it also spits out a simple video prompt that I know will work if I want to change things into a video.
I could just copy and paste that for the video conversion. But let’s go back to our Wallace prompt.
Yeah.
So you guys will see one of my estimates here.
So if we just drag in Wallace and Gromit.
It will do the same for prompts and video prompt based off of whatever image is dropped in here.
So I will Let’s see here.
As soon as the Wi-Fi catches up.
Just drop the instructions into it. Nope, never mind.
We will.
instructions into Discord.
And this is something I like to follow.
I know this is more on the custom GPT side than majority, but whenever I’m trying to get an AI to write prompts for another AI, I generally follow a format where I put in what’s most important in my mind and then I point it towards any online documentation that exists for it.
So it’s, it’s really simple instruction you guys can look over and you see it works really well for just if I want to explore ideas I don’t have to think about what to prompt out I’m just like let me explore a raccoon to Deanna Jones and I immediately have four good prompts to start working for. So uh one more room quick Just because I mentioned it before, what’s a tech question somebody has?
This was my first favorite custom GPTI I ever made. What’s the difference between UDV and TC?
I’m not sure what it’ll say for that.
There’s a big difference between UDV and TC.
Simple tech advice.
Generally, let’s see, can I play?
I love custom GBTs.
They’re simple and it’s just a lot of fun.
Let’s play a game. So just make sure. I know it’s a whole lot of info and kind of back and forth. That’s just kind of what I can think of to cover the most mid-journey, you know, about an hour that I could. Would you guys like to keep exploring? Any questions or just want to, can I call it here?
We’ll probably, well, any questions first that you want on the recording?
Yeah, we’ll stop the recording and then kind of shutting down my laptop because my battery was about to die.