Hi guys. Hi. Hi everyone. Uh, welcome

back. I'm Edo. New setting. I changed my

camera. Hope uh this work better. Um,

thank you to be here today with me. And

today we are going to see something

cool. I think I hope we are going to see

the new XLTX2.3

video

or engine. Um,

we already know the the last uh update,

the LTX 2.0.

Um, it's quite good. But this this new

video

uh model, opensource new video model, it

is amazing in my opinion.

um for many reason. Um

some of these reason are here. Uh you

can prompt audio as you can as you can

do with the 2.0 version. Uh you can

create video clip up to 20 second. uh

there is the the possibility to uh to

encode to 50 um frame per second and you

can create something with 4K

uh resolution.

Uh today we're going to see everything I

hope um the image to video model and the

first frame last frame uh workflow. Um,

I want to show you some of my uh render

some some of my um videos and try to

talk with you about um pros and cons uh

of this model. Um

I'm working in local.

Okay. So, I'm working with my Ryzen 9 uh

5060Ti with 16 gig of VRAM and 74 uh 64

gig of RAM uh for the uh the

for Windows. Okay. Uh the system RAM,

this is my

um spec uh and I'm working local. So, no

um

I I I I want just to create something

that works for me, for my spec, for my

confi, for my uh computer. Um I will

guide you in the in the

installation process. And now we can go

uh right here.

This is the confui

uh blog you know um

part with everything uh with the

highlights of the model with the some

with some example output very cool and

with this controversial

workflow um first of all I don't work

with prompt to video um workflow so I'm

just working with image to video I think

it's the only way to work professional

in a professional way. Uh prompting it's

uh not so basic when you want to have

control of your uh scene. Um and if you

want to see something more specific,

this is the GitHub paper with

everything.

Okay, you have to install a bunch of of

things in order to work with this LTX.

But uh but but but but uh we see

everything um to to to work with. Okay.

So um

these are all the checkpoints the 2.3

uh 20 billions dev or distilled if you

want to have some uh um save for your

memory. This is the upscaler. We talk

about the upscaler um in a second

moment. This is the Laura uh to uh to

in order to work better with um spatial

interpolation. This is the text encoder

Gemma. It's very heavy in terms of

memory and space but whatsoever.

And these are all the if you want to

have the the precise control

um uni union control or motion track

control. Okay, I will not use these two

in the workflow but you can whatever you

want. So

this is the um workflow

you can find here. I think I think this

is the workflow. This one

right there we can download it and you

know you can just drag and drop in your

uh voila.

Yeah is the same. Okay.

Uh do um this is the simple workflow.

Everything you need to know it's is this

um your your model checkpoint your

loras. Okay.

And the upscaler. This is a simplified

version of the the LTX um workflow. We

have this right here is is an a scale

node

in order to work uh fast. LTX first

downsize your frame and then upscale

that. So it it can work very quickly

with the downside version of your your

images and then with upscale with

upscaler you upscale everything. Okay.

in order to do something very quick and

very fast.

This is the the you know the subg graph.

Okay, seems to be very simple

checkpoint. This is the Laura. I'm using

another version of Laura, but it's the

same. Okay, this still Laura this one.

Okay, this Laura here we found it.

Sorry, right here. Okay, this Laura,

this still Laura right there. Bam. The

gemma and the upscaler.

The value of your frame rate 24. And the

value of the image. Okay. Heights and

weight.

I just click this one, this button here

to go to the subgraph. This is a cropped

version of a specific workflow

that is this.

Uh it seems to be very you know

uh um

demanding in terms of um understanding

and it's quite quite heavy.

Uh and you know you we have to um

one thing very important this version

this version here is a modified version

of the um confui

version. This is the confi you can

download from and in the confi version

you can find this prompt enhmentment

something

for me it's a mess every time I'm using

this prompt enhancement here I will face

errors errors and errors and errors so I

delete it it's completely useless use

Gemini or whatsoever to create a perfect

prompt this is a mess

And the second thing

um for me for me I had some many

problems with the first key sampler. I

don't know why. I don't know why. So for

the gener low resolution generation

process I swap with the LTX normalizing

sampler. So it's a same simple um you

know um

um node for LTX and remember you have to

update your confi of course everything

is in confi but you have to update

confui. So the the idea of the the

process if is to uh reduce the the scale

of your image. There is a process right

here. Um the less you use this

compression

uh the the most the the the model will

find details and everything very crispy.

So you can you can handle with 18 20 24

of compression to have something very

balanced in terms of quality. Okay. Uh

this part will reduce crop the the the

the image ratio. Uh and then this part

generate a low resolution image and then

there is an upscale

an upscale that creates something with

the right resolution. So he um he reduce

of enough the resolution and then

multiply by two. So it's it's a trick to

uh create something very very specific,

very cool. Um and this is the part with

the um high resolution generation with

the the code everything

it's

right here. Okay. So in my case with

this

um with this node deleted removed and

with this change here you can leave this

one

as the the provided one from the simple

um checkpoint sorry workflow from from

confi but I had to um replace this

Sometime confi is magic. I surfing on

Reddit, surfing on github in order to

resolve my problem. I fix it with this.

I don't know if this will fix your LTX,

your confi.

Uh guys, I hope so. Um I'm trying to do

my best in order to um show my process

to find the resolution. You can also uh

call Gemini Gemini or code whatsoever in

order to solve your problems. But it's

very uh stressful because everything

with Python could be different in every

single um computer. So my my problem my

pro my solution could be not your

solution. But for me this workflow works

fine for the generation of one you know

one

um

one video

um the value of um 168 it's the value of

the time okay the the length of your

generation it's about seven six seconds

7 seconds Okay. And you can work

whatsoever with your um with your uh

model and your specs. Of course,

uh some tests. As you can see, I use

this

uh girl here

for a simple test. And you can see

okay, there is a simple camera movement.

But we have a nice quality and um a nice

sound with the rain. I create something

with the rain. Uh beautiful girl rain

socks or four clothes blah blah blah. Uh

a pandon. Um I done something like this

also.

Okay. I used another

another u single image.

Okay, as you can see, we have a nice

nice video and nice quality with the

music with the the sound of the rain,

precise eyes movement.

I think it's quite impressive, quite

good. Uh let me see if this is this is

not uh sorry another uh um you know

another uh test and this is another one.

Okay I think it's it's impressive. Okay,

it's a open source video generation

model

and in with my spec with my confi with

my 5060Ti.

Okay, I generate this video in about

150 seconds, two minutes

for me is is impressive. And the

resolution, sorry, the resolution is

uh right there

and HD resolution.

So you can simply upscale that with an

upscaler RTX C2. I made a video I'll

link right here to choose the better

tops if you want.

But with this resolution, the sound

it's very very impressive, very good.

And there is also a voice

in Italian

or I made.

Okay, this is Sara that says something

like, "Hi, I'm Sara. Nice to meet you."

Okay, in Italian, it's it's powerful.

There's a good lip sync. Okay, you can

add music. You can add um uh speaking uh

and it's very sensitive in terms of um

you know um

prompting. So, be careful.

The next thing I want to show you is

this this workflow here.

Uh this workflow it's another workflow.

Um I leave the the link in description.

a great work of uh you know um I think

this workflow is from

uh CQ

uh channel but I leave in description

thank you for your your um work it's

powerful okay um I think I will find um

just a second I want

to show you

if I

able

well sorry guys I just want to show you

the page but otherwise I'm going to link

in description because it's very useful

right there. Thank you. Thank you. What

dreams cost?

What dream cost? Thank you for your

incredible nodes for the first and the

last frame generation.

Uh it's very powerful.

uh and I will create another video for

this uh you know um this part it's very

very cool very powerful and I think it's

worth the the the length of a single

video uh just to create a right uh you

know process uh to uh to this this um

node.

So uh for today it's all I think it's

impressive quite powerful and very um

handy in terms of demand of spec. So you

can try with your with your um specific

with your GPU and I think uh worth the

the the

you know the time um it's I'm I'm using

also for professional work um just to

have um a very precise

um way to look at um um camera movement

or uh you know a mode. I think it's very

very impressive, very cool and I think

uh it's uh funny to use. So I hope this

video could help you to work inside, you

know, um inside LTX, inside this um this

uh workflow. I hope some tricks could

help you uh to

to to have something to to work with.

Okay, I'm just trying to to do my best

to provide something cool for this kind

of of content. And I hope to see you uh

next video. And I think we'll talk about

this um uh first first and last video uh

you know um workflow and I'll post um

as as soon as I can. Okay,

for today it's all. Thank you guy to

being here to to be here with me. Um

and to also um give me some feedback um

some tips. Um I I appreciate that. I'm

trying to do my best. If I can I I

answer you with my with my capabilities

and I hope to hope to see you again next

video.

Dr. Sh