LORA with ControlNET - Get the BEST results - Complete Guide // Jenna Ortega

Рет қаралды 69,098

Күн бұрын

Use Lora in ControlNET - Here is the best way to get amazing results when using your own LORA Models or LORA Downloads. Use ControlNET to put yourself or any other person in any Pose you want. There are a lot of Settings to consider: The Stable Diffusion Models, Sampling Methods, CFG Scale, ControlNET Weight. Also the Resolution you render at is very important to get better results.
This Tutorial also shows how to Inpaint using Lora and ControlNET at the same time. That way you can easily Improve you results even more.
#### Links from the Video ####
How to install ControlNET: • ControlNET Stable Diff...
LORA Install Guide: • LORA: Install Guide an...
My AI Facebook Group: / theairevolution
civitai.com/models/4201/reali...
civitai.com/models/4384/dream...
civitai.com/models/4823/delib...
civitai.com/models/7394/jenna...
Support my Channel:
/ @oliviosarikas
Subscribe to my Newsletter for FREE: My Newsletter: oliviotutorials.podia.com/new...
How to get started with Midjourney: • Midjourney AI - FIRST ...
Midjourney Settings explained: • Midjourney Settings Ex...
Best Midjourney Resources: • 😍 Midjourney BEST Reso...
Make better Midjourney Prompts: • Make BETTER Prompts - ...
My Affinity Photo Creative Packs: gumroad.com/sarikasat
My Patreon Page: / sarikas
All my Social Media Accounts: linktr.ee/oliviotutorials

Пікірлер: 99

@jameshughes3014 Жыл бұрын

I've never seen anything advance this quickly. It's honestly amazing how the community is coming together to build this collection of tools. Thank you for helping us navigate all the new stuff, it's easy to fall behind.

@OlivioSarikas Жыл бұрын

Thank you. I'm amazed by that too :) Shows what happens when science opens up and the general public get's to put their idea on it :)

@Antonio.dei.Demanaro Жыл бұрын

They're making artists go out of business like it's the cure for cancer. Only if they applied the same energy to actually finding the cure for cancer.

@jameshughes3014 Жыл бұрын

@@Antonio.dei.Demanaro how many artists do you know that have lost their jobs?

@user-mr1ph2ie3e Жыл бұрын

@@jameshughes3014 Maybe he exagerated that bit because well established artists in the art community will stay for sure, but i've seen plenty of people just give up pursuing art bc of it, which i find a bit dissapointing

@jameshughes3014 Жыл бұрын

@@user-mr1ph2ie3e i feel like if you really want to make art, you will make art. If a person gives up because of something like this, maybe creating art wasn't enough of a reason on it's own, and it really should be

@ninjuhdelic Жыл бұрын

shoutouts to all you geniuses making these tools, yall doin the lords work

@frankschannel2642 Жыл бұрын

Wow. Another superb video. Thanks, Olivio!

@OlivioSarikas Жыл бұрын

Thank you so much :)

@vapedragon983 Жыл бұрын

Thank you for the kind words grampa. needed this

@odawgthat3896 Жыл бұрын

THANK YOU!!! THIS IS THE BEST LORA TUTORIAL, PERIOD!

@OlivioSarikas Жыл бұрын

Thank you!!!! :)

@michaelm7901 Жыл бұрын

Thank you for the training, I could not do it alone.

@TheMrawesomest Жыл бұрын

Controlnet has been a revelation for posing! Been having a lot of fun with it for the last few days.

@OlivioSarikas Жыл бұрын

I absolutely love it

@gloxmusic74 Жыл бұрын

Control net is groundbreaking, I'm having so many good results with, also super impressed with scribble using colouring book pics !!

@gopolarisstudio Жыл бұрын

Scribble?

@AI-edu-minchan Жыл бұрын

You are a non-stop source of ideas! I respect you! I'll take a good look at the video!

@OlivioSarikas Жыл бұрын

thank you so much :)

@friendofai Жыл бұрын

Amazing video as always!

@TheElement2k7 Жыл бұрын

Thanks for the great video,,to show how the different models looks like, 😊👍

@lamstar70 Жыл бұрын

Many Many thanks, from Hong Kong.

@leppie Жыл бұрын

BeefMaster FTW! :D

@OlivioSarikas Жыл бұрын

BeefMaster knows what's good: Everything good 😄

@dreamzdziner8484 Жыл бұрын

I think you were the first person to bring that daz posing trick and now most channels are focused on that method 😁. Great content as always 👌❤

@OlivioSarikas Жыл бұрын

Thank you. I'm planning to do a video on that too :)

@EriIaz Жыл бұрын

There's a major misunderstanding of Controlnet's canvas size. The height and width settings only kick in if you use an actual canvas for scribbles and whatnot manually. As soon as you LOAD a guidance image into the ControlNet extension tab, it automatically aligns with resolution of the loaded image. You can confirm that if you set it to 0x0, the result will be the same. And you can load a ridiculously highres image into that tab: the webUI will outright crash because it's not supposed to handle something like preprocessing a 6000*9000 pixel image in that section. Source: the developer of the extension, Mikubill, as well as my own experience with it.

@OlivioSarikas Жыл бұрын

ohhhhhh! You are right! I set it to 0x0 and same result. Good to know! Thank you!

@IceMetalPunk Жыл бұрын

Small tip: the recycle icon button next to the dice button for the seed will populate the field with the last generation's seed automatically. No need to look for it in the generation parameters text block.

@OlivioSarikas Жыл бұрын

Cool! Thank you!

@OlivioSarikas Жыл бұрын

#### Links from the Video #### How to install ControlNET: kzfaq.info/get/bejne/mc9ngcmF2M_KlGQ.html LORA Install Guide: kzfaq.info/get/bejne/b7OEZNFqnK-4kXk.html My AI Facebook Group: facebook.com/groups/theairevolution civitai.com/models/4201/realistic-vision-v13 civitai.com/models/4384/dreamshaper civitai.com/models/4823/deliberate civitai.com/models/7394/jenna-ortega-lora

@p_p Жыл бұрын

honest. i love this vid

@OlivioSarikas Жыл бұрын

Thank you so much

@ysy69 Жыл бұрын

Thank you for this very helpful tutorial and comparison. Question: Did you train your LORA on each model, or you trained using SD base 1.5 and then applied the Lora file with the prompt using each model ? Thank you

@MikkoHaavisto1 Жыл бұрын

the format of this video is really good, it stays interesting throughout for both beginners and advanced.

@OlivioSarikas Жыл бұрын

Thank you. I was worried it's maybe too technical at some points ;)

@norianorlan6093 Жыл бұрын

Great guide as always! Liked to see examples of the different weight in controlnet. I wanted to ask if in your lora you used also photos of yourself standing or just photos of your face?? Your results are amazing

@OlivioSarikas Жыл бұрын

In this case the lora is just using photos of my face. That's maybe the reason why i get so fat at 0.0 weight, because the lora doesn't know what i look like. Though usually SD is pretty good at estimating the body form the face alone. I need to train a full lora model on various photos of me soon

@EllenVaman Жыл бұрын

you are funny and very skilled ;))))))

@Lykon Жыл бұрын

Hey Olivio, I'd be curious to know what you think about NeverEnding Dream, my latest model. It should be between DreamShaper and photorealistic models in terms of photorealism. And it works with anime loras making them realistic.

@backster4744 Жыл бұрын

Great video! I believe they updated the ControlNET extension recently and there are new UI options, if you could do a video explaining what they do that would be great, thanks!

@OlivioSarikas Жыл бұрын

Thank you, i will check that out

@pastuh Жыл бұрын

Video conclusion: 07:30 - 07:45

@benjamininkorea7016 Жыл бұрын

Olivio. I have loved all your videos so much. You are the AI master! I have a question-- do you have any experience modifying skin tones of existing objects? I have some photos taken indoors, and when I use the faces for inpainting in outdoor scenes, I cannot find any way to make them look good. These are me, my wife, etc. So I really want to keep the facial features but only adjusting lighting or skin tone. I'm a little hopeful that ControlNet might have settings that will force a total face recreation but allow for lighting changes?

@OlivioSarikas Жыл бұрын

hm... i think if you need exact tons and colors you might be better of adjusting them in Lightroom Classic.

@aiforall217 Жыл бұрын

Thanks for these useful tips , I learned a lot from you, only thing is, If you could kindly use the dark theme for the videos, I have eye problem and it's really painful to watch on light theme , just a request, Thanks again

@ysy69 Жыл бұрын

Dear Olivio, did you Lora-trained your images for each one of the models, or you simply inferred these images against each model ? Thank you.

@bugattiowner Жыл бұрын

is there somiething to put face on video isntead of image?

@SamHardie Жыл бұрын

how did you get your face on it? i see some kind of a file reference in your prompt bar, how does that work?

@FrankLeeMadeere Жыл бұрын

Thanks for the good advice! btw if you need to lower the cfg on a lora really low, that's evidence the lora was over-trained. You're probably seeing pictures very close to the training images.

@OlivioSarikas Жыл бұрын

Good point. Although, like i said, the model works perfectly fine with just the prompt and the overbacked look only appears with ControlNet in the mix

@AeonPrime92 Жыл бұрын

I used i2i for the last few months to transfer poses, it works quite well if the pose isn't anything too special (you're not going to recreate yoga poses with that, obviously). With the additional hassle and storage requirements, do you consider ControlNET worth the extra steps?

@OlivioSarikas Жыл бұрын

yes absolutely, because ControlNET can to even complex poses with high precision (als long as all limbs are clearly visible) and even helps with hands and fingers

@SamHardie Жыл бұрын

How did you get it to use your face? i cant find where thats detailed

@stas_lu Жыл бұрын

7:55 the perfect photo ! )

@OlivioSarikas Жыл бұрын

The BeefMaster is always perfect. No matter what he does ;)

@kylewang6704 Жыл бұрын

How did you use stable diffusion to control pose your own photo?

@CultofThings Жыл бұрын

So what is Lora, and how do you install it into stable diffusion?

@LegendD112 Жыл бұрын

is it possible to use controlnet to process a bunch of images like img2img? to make animations, thanks for your videos king

@OlivioSarikas Жыл бұрын

I will look into that next. i'm pretty sure we will see controlnet for deforum very soon

@mugocloud Жыл бұрын

It would be awesome to see a feature for animating the openpose output (that colored stickman image). For example creating images using same seed and cfg but controlnet will always move the hand back and forth (left to right) like a few degrees to create a wave animation.

@OlivioSarikas Жыл бұрын

Yes, i would love to see that too

@relaxingheart682 Жыл бұрын

Would love to see a video on using img2img to keep the original uploaded image, but only change the colors, or only the pose, without changing any other details. If this is possible to do.

@OlivioSarikas Жыл бұрын

I will test that, but i doubt that is possible because in that case the AI would need to know what that clothing would look like in another body pose. I suppose if you trained a lora model and a person wearing just that outfit that could work to a certain degree

@piggypooo Жыл бұрын

7:51 Beefcake!😂😂

@Jake-bh1hm Жыл бұрын

I keep getting RuntimeError: mat1 and mat2 shapes cannot be multiplied (77x768 and 1024x8)

@Smudgie 9 ай бұрын

Can you find the seed information from any images you have rendered in previous sessions?

@OlivioSarikas 9 ай бұрын

yes, put them into PNG info in A1111

@rainerzufall1868 Жыл бұрын

For ControlNet, any reason you use Canny instead of HED? I prefer the latter after some initial experiments.

@rainerzufall1868 Жыл бұрын

Or maybe, since you wanted to copy the pose, maybe OpenPose would have also worked?

@OlivioSarikas Жыл бұрын

not really. they are both good. but you are right, hed would probably be better for photorealisitc portraits. but just by a little

@djgaming3409 Жыл бұрын

how to download?

@starflyai 10 ай бұрын

tip please use dark mode

@RasmusSchultz Жыл бұрын

9:50 "bearing a white suite" should be "wearing a white suit" - might get slightly better results? 🙂

@jolotschka Жыл бұрын

In invoke up to now that's not running? And also not with sd v2?

@OlivioSarikas Жыл бұрын

Invoke doesn't have extension, so sadly no. For SD 2 i haven't actually tested, since most models and loras are for 1.5 anyways

@marioilieski7088 Жыл бұрын

beef master 💀💀💀💀

@Maeve472 Жыл бұрын

can i add an any text like "ilovepeople" whats the prompt for that?

@Maeve472 Жыл бұрын

i wanna make logo

@danielnico3950 Жыл бұрын

What gpu do you use?

@OlivioSarikas Жыл бұрын

a 3080ti

@danielnico3950 Жыл бұрын

Ohh okay bro.. so your vram is 24gb right?

@mykhailyevzhenko8287 Жыл бұрын

Best for tinder 😄

@synthoelectro Жыл бұрын

it makes one wonder if stable diffusion will ever be simple

@OlivioSarikas Жыл бұрын

sure it will. but these are the very early years. Not sure how old you are, but in the early days of computers a PC did not have any memory inside, so every time you wanted to use it, you had to write the code for the program by hand. later you could load it from a music cassette and the a floppy disk. only later PCs got easier, but it was still mostly CMD with no UI. Windows was like magic back then

@synthoelectro Жыл бұрын

@@OlivioSarikas I had my first computer VIC20 in 1984 ;) Atari 2600 at the age of 6 1983.

@synthoelectro Жыл бұрын

@@OlivioSarikas oh yes, those days were pretty interesting. I didnt' get my first actual PC, until 1995, Packard Bell, loved DOS though.

@skol600ml Жыл бұрын

"That's isn't good because looks too much like me in real life"

@edd9581 Жыл бұрын

Have you tried to use the end animation with stable diffusion? 🙀

@OlivioSarikas Жыл бұрын

end animation? what do you mean by that?

@edd9581 Жыл бұрын

@@OlivioSarikas the animation of the blue cat of the end of the videos

@OlivioSarikas Жыл бұрын

@@edd9581 not yet.. Good idea

@claudianreyn4529 Жыл бұрын

Stable diffusion makes bad faces. I will give up on it. Until they fix it maybe.

@OlivioSarikas Жыл бұрын

The face are so good you can't tell if they are real photos or not. Look into better prompting and using the right settings, models and resolution for your results.

@GomezBro Жыл бұрын

Why are you obsessed with the girl bro

@OlivioSarikas Жыл бұрын

Jenna? This is literally the first time in the history of my channel that i post a image of her (as far as i remember). Also she isn't a "girl", she is a women and 20 years old. Stop being a Incel. Women are awesome in every single way. And they made us. Stop being a hater