Leonardo AI Levels Up With ControlNet & 3D Texture Generation Monzon Media English (auto-generated) English (auto-generated)

Leonardo AI Levels Up With ControlNet & 3D Texture GenerationYouTube↗ IA↗(long shot)
Education 17.9K views 792 likes 16 Apr 2023

Monzon Media 42K subscribers Canada YouTube↗

Click to Play Video

well hello good people you know it's

been a while since I've covered Leonardo

Ai and there's been quite a few updates

since my last video we're going to talk

about control net promptmagic V2 updates

to the upscale feature and the ability

to texture 3D objects first off let's

cover prompt magic V2 if you go over to

the left side of the interface you're

going to see prompt magic V2 if you

hover over the question mark it tells

you what this feature does and basically

it helps to adhere to your prompt a

little bit better so I have a prompt

here of a female Ninja Warrior nothing

too special I'm just going to generate

four Images without prompt magic to show

you the difference currently using the

dreamshaper 3.2 model and as you see

here we have some pretty good results

here except for this one

but if I were to open this one up you

see it's not bad it could use some work

but all in all it's workable right so

now I'm gonna toggle on prompt magic and

you see that it's got a prompt magic

strength if we hover over the question

mark it says how strongly prompt magic

influences the output so the higher the

number the greater the influence so if

you look at this style in this type of

look we're gonna leave it at 0.3 for now

and it should influence the image just a

little bit so here are the generated

images let's open up one of the original

ones you see it's got that sort of

cinematic Vibe but with prompt magic

it's got some additional color a wider

dynamic range even I guess you could say

it's a bit more lively as well and

that's only with a strength of 0.3 now

let's over exaggerate it to 0.8 and see

what we get now here are some of the

generations it looks to me that we're

straying away from the original look and

now it's becoming its own so it's kind

of like applying a filter on top of the

dream shaper model and really it's up to

you on how strong you want the prompt

magic to be but now you just have more

creative options the other thing you can

do is utilize a reference image so what

I'm going to do here is come down to the

bottom make sure you're on image prompt

we're going to click here and import an

image so I'm going to use this image

that I created we're going to import it

to use as a reference image if we look

at the image weight image weight is

basically how strong of an influence the

reference image is going to have so

let's put it to about halfway maybe to

about 53 prompt magic strength will

bring it up to 0.4 let's see how that

turns out so here's one of the first

images as you can see the hands aren't

great but in terms of using the

reference image you see there's sort of

a transfer of Styles not the best arms

and hands in the world will have to work

with this that's for sure but you could

clearly see the influence of the

reference image in the environment as

well so you would have to play with

these sliders to get the results you

want I like to think of it as sort of

like a style transfer or combining the

two that's a great way to do that just a

few minutes ago these images that I

rendered the posing isn't great if we

look at this one it's all over the place

the legs look weird it's not working at

all now that Leonardo AI has implemented

control Nets we have more control over

the posing of our images let's turn off

prompt magic for now and the first thing

you have to do is make sure you're using

a model that's trained on stable

diffusion 1.5 let's change this to dream

shaper 3.2 because I know it's trained

on stable diffusion 1.5 as we scroll

down you're going to notice a toggle

from control net but you need to upload

a reference image first we're going to

start off with this image as my

reference image make sure you have image

to image selected we're going to click

on here and simply click on the file to

upload it you see it's going to be here

on the left so for this example I have a

width of 832 and a height of 640. now we

can toggle on control net and you're

going to see that there's an option here

to control the weight you'll also notice

that you have three options here the

first one I want to show you is Edge to

image I'm going to go ahead and generate

four images with the exact same prompt

let's take a look at the results here so

for one the posing is pretty accurate

the hands could use some work I know but

the way Edge to image works is that it

looks at the details of the subject and

its surroundings you see these patterns

at the background you're going to see

the same patterns in the reference image

in the window in the background with

this image you see that the window

outlines have now become like

decorations in the background so that's

how Edge to image Works yes it'll change

the details of the character but also

the surrounding so basically takes the

outlines and the edges and recreates

them with that in mind let's switch this

to depth to image we'll hit generate

with depth to image you see that not

only do we have the pose but the

environment is totally different because

What's Happening Here is that it's now

looking at the foreground the middle

ground in the background not necessarily

the edges and the outlines of the image

but it's looking at depth this is a

great great way to pose your character

and change the environment around it

unlike the previous example where the

window was influencing the image you see

here that the background is totally

different same thing with this one and

I'm going to switch this to post to

image but you're going to notice here

that it doesn't really follow the pose

as much as I'd like it to it's close and

it does a great job in terms of the

details but we're not getting the left

arm sometimes you just have to get a

better reference image so I'm going to

use this one instead same prompt

generate four more images here's the

generated images this one is very close

I like the way the hands are here a

little janky workable we could take it

into canvas and fix it same issues with

the hand here but as you can see the

actual fighting pose is more accurate

when you're using post image I find

images with less of a background do a

lot better for these types of of images

the good thing is that you have three

different options here to get the

desired result now let's talk about the

upscale features if you hover over your

image you're going to see a creative

upscaler an alternate upscaler and now

an HD upscaler here's the original this

is a carnage venom hybrid character I

quite like the original I want you to

pay attention to the details of the face

the horns here and especially the

texture we'll switch this to the

alternative upscaler you'll notice in

this area some of the details a bit more

prominent now so it's given it a little

bit more of an edge with the upscaled

option those details now have smoothened

a little bit more switching it to the HD

upscale method there's details here that

we didn't even see with the other

options here I find the HD upscaler is a

good balance and I tend to use it more

personally myself but it really depends

on what you're creating moving right

along let's look at texture generation

if you see here on the left there's an

area called texture generation it is an

alpha stage so there is some improvement

to be made first I want to show you a

couple examples where I've had success

and this is supposed to be a dragon egg

that I textured it's sort of like this

chromy Granite type of material I forget

what it's actually called and then

here's a rock where I did the same thing

and here's some shoes that didn't work

so well and I think because there's

different surfaces you've got the

shoelaces you've got the soles here I

don't think this is for complex models

just yet but for simple models like this

that don't have multiple layers where

you have to texture I think it works

best with things like this at least for

now now in case you want to try this

yourself first you need a 3D model the

first place where you can get 3D models

and for free is a place called turbo

squid unfortunately the site is down at

the moment they're experiencing heavy

traffic so if you don't get in just try

again later but as you see here below

there are free 3D models even simple

ones the other site is called CG Trader

one thing you need to be aware of with

these 3D models different platforms have

different file formats currently

Leonardo AI only accepts file formats

that are obj okay so if you filter for

this particular file format you'll see a

range of free models here that you can

try out again start with simple ones

like yeah like this carpet maybe some

grass you do have to register but all

you have to do is click on the image and

then you're going to get a free download

section here it takes a few seconds just

to download it and you're good to go

once you have a model that you can work

with head over to upload new object

click on that give your model a name I'm

just going to put cookie test because we

are going to texture a cookie you can

click here or drag the file on top which

is what I'm going to do there you go

there's my cookie then all you have to

do is click on start texturing you'll

see your model in its raw form I'm just

going to move this down so that we have

it facing us you can use the mouse to

change the perspective and all that

stuff but you'll also see that you can

control roll it by this corner here and

basically you have your y-axis which is

up and down x-axis which is left and

right and your z-axis which is your

depth so front and back or foreground

and background before we enter the

prompt I just want to show you the

settings mesh adjustment is just turning

it and zoom will bring it closer or

farther but these have no influence in

how it textures the model currently the

assets we'll look at after once the

texturing is done so we'll head back

into generation and all I'm going to

type in here is chocolate chip cookie

for now it's best to just use simple

terms now what we're going to do is

actually just generate a preview and

it's going to cost you five tokens so

once you click on that you see here that

it's processing it usually will only

take a few minutes to do that and the

preview is just going to show you

typically on the face of the actual

object it's not going to texture

everything just yet here we have the

generated preview and I know at first it

really doesn't look like much but trust

me it gets better but as I said it's

only going to do the front of the object

and you see the back there nothing on

there at this point what you want to do

is generate full texture just be aware

it costs 30 tokens when you hit generate

full texture you see that it's going to

give you a warning that'll be about five

minutes maybe longer depending on the

traffic on the server you could simply

click out of here and go somewhere else

now in my case I already did this so

let's look at the chocolate chip cookie

that it developed as I said from the

preview it didn't look that great but

once it was fully baked you'll notice

that if we flip on the back end here

it's got a different texture just keep

in mind this is still an alpha they're

gonna improve this a lot more but as you

can see with a simple object even with a

simple prompt you can get some pretty

decent results at the moment well that

was a lot to cover let me know in the

comments below what you like best about

what you saw and how are you finding

Leonardo AI now if this happens to be

your first time coming across Leonardo

AI make sure to check out these videos

to get you up to speed they might be

slightly outdated but it'll at least get

you started until the next video my

friends I'll see you when I see you

Settings Privacy About Contact Donate