BETTER than PROMPTS - The Future of AI Composition
ฝัง
- เผยแพร่เมื่อ 6 ก.ย. 2024
- This is better than Prompts. In this Video I show you how to get exact results. Have full control over your AI images. Create exactly what you want. This Automatic 1111 Stable Diffusion Tutorial uses Photo Bashing, Img2Img Render and Controlnet to get indepth control over your AI Art.
#### Join and Support me ####
Buy me a Coffee: www.buymeacoff...
Join my Facebook Group: / theairevolution
Joint my Discord Group: / discord
Follow me on Twitter: / oliviosarikas
Support my Channel:
/ @oliviosarikas
Subscribe to my Newsletter for FREE: My Newsletter: oliviotutorial...
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 Facebook PHOTOGRAPHY group: / oliviotutorials.superfan
My Affinity Photo Creative Packs: gumroad.com/sa...
My Patreon Page: / sarikas
All my Social Media Accounts: linktr.ee/oliv...
For VAE Select and Clip Skip got to A1111 Settings -> User Interface-> Quicksettings list -> and add: sd_vae, CLIP_stop_at_last_layers
SOFTWARE USED: Affinity Photo 2
#### Join and Support me ####
Buy me a Coffee: www.buymeacoffee.com/oliviotutorials
Join my Facebook Group: facebook.com/groups/theairevolution
Joint my Discord Group: discord.gg/XKAk7GUzAW
👋
I´m new to the channel can you please explain what software are you using. Does not seam photoshop.... Thanks*
@@OneTap__ look at the sticky comment under the video 🙂👍
Ohhh the software is affinity photo. Is it free ? What’s VAE? And do you have midjourney tutorials on how to get the face like one wants and then how to change it to 3/4, looking up , etc ?? :) and how to move the body how one kinda wants and the clothing ?
OMG!!! to make it real simple. All we need is mediocore photoshop skills making crappy draft and get SD to finish it up. And for small edit. Generative fill or to stick by the rule and NOT using any paid apps, inpaint. And you mentioned about the free app that is photoshop like which streamline with SD!! Thank you for the training. I spent 2 weeks learning about everything but prompts. Mainly because of that. Control. I want to make money as a designer. Not an artist working on the best prompt.
Coming from Photoshop, Illustrator and currently using the free macsoftware Gimp, this is just absolute magic. That last image with the woman and the halo is just mesmerizing!
Great video! I have been working as a high end retoucher for 10 years and whenever I other retouchers say to me they are worried about AI I honestly think it's because they do not know much about it. I've been getting more work than ever creating ai fashion imagery etc and photobashing is one of the key components needed. Then further editing with multiple ai images to use.
Could you perhaps create a new from A-Z tutorial on how to utilize all of the most recent tools?
Things are changing so quickly, it's hard to start from scratch without using outdated tools.
As a AI art and photo editing newbie, really appreciate the tune and speed of the tutorial provided in this channel.
Ease to follow and comprehensive explanation,
Really learn step by step and grow a little every day.
Agreed! That's where I am at too so I appreciated the thoroughness and pacing!
@Olivio You are one of the greatest Stable Diffusion Masters out there! Your videos are interesting and entertaining to watch. Thank you!
Ive been using photoshop and SD for a few months now, hands down the best video for anyone to start getting into this!!! Keep up the amazing work! We all really appreciate you taking the time to make these videos
Thank you so much :)
Sorry I don't understand how this is useful for beginners. He sets a hundred different settings without explaining any of it. How is that better than prompts, which beginners can actually use?
@@NicolasConnault who said this is for beginners? It's easier in the sense that you can get exactly the composition and even colors you like. If I would explain every setting in every video my video would all be 3 hours long and most people wouldn't need 95% of that information
@@OlivioSarikas I was replying to odawgthat who said this was the best video "for anyone to start getting into this". I assumed he was talking about beginners. Personally I get very frustrated when I'm shown what to do but not why, but I do get your point, and a 3-hour video would be a lot of work to make :)
@@NicolasConnault I think he means people who want to use img2img more, not people who are completely new to A1111 and the settings of it. However, i have a lot of videos to get you started on my channel
"The future of AI composition" - goes on to show how graphic designers have been working with Photoshop for almost 2 decades.
He's probably showing it to people who haven't been working with photoshop. I sure haven't.
Underrated comment
Photobashing is amazing, I've used it on a big project and it would have been impossible without it. It's one of SD strongs, it would be very annoying to do it in Midjourney.
It is. I've been exploring this with SD for quite a while now, and fun though pure t2i is, running images from past photoshoots through SD has been really interesting, with even failed shots in some cases being useful for i2i. Some types of retouching that can be a time consuming challenge can also be done very satisfactorily and quickly with SD.
Or instead: Use ControlNet's lineart preprocessor on your 3 images, then combine the lineart in photoshop, then plug the resulting lineart into ControlNet for generating. (Lineart is much easier to work with, although color control takes an extra step w/ either img2img or a second ControlNet.)
Wouldn't that be even more steps and less detail as a result?
@@OlivioSarikas Lineart is easier to work with. You can remove extra fingers, resize eyes, nose, mouth, add / subtract from the hair style all very easily. Redrawing sections of lineart for absolute composition control is much faster than hand-painting. There are no seams / mismatching colors in the end result.
(I don't know about "less detail". I've never had ControlNet cause a problem like that. I get high detail by upscaling with realisticVision and ControlNet's tile sampler.)
(I'm a traditional artist, and I am using SD to make my art photorealistic and to make it faster! :-) )
@@jonmichaelgalindo interesting. thank you for letting me know, i will look into that
When plugging the lineart back into controlnet, which preprocessor and model do you choose?
@@7ens3nButt0n I would assume you would load the lineart you made into controlnet, so no preprocessor, and use the lineart model. This would be for txt2img.
I haven't tried this method yet but that's what I would assume the workflow would be.
Here we go! This is where AI gets good, when you combine artistic skills with it :3
Yep, this is where Art meets AI :)
I tried this workflow for a few days. It's really not an easy exercise when you have a very clear idea of what you want. It's a bit frustrating when you have to spend hours explaining to the AI how to change a very specific detail in the desired way. But I'll keep at it, thanks for the video, it gave me another boost.
You are welcome. You can always try to use the already rendered image, change that a bit and then either only render the part that needs improvement or render with a lower denoise value to get a close result
Yes if words fail. They often do in life. =] Try using another way to communicate that idea. ControlNet is your copilot. 1 image = 1000.
@@OlivioSarikas 8:45 how u fixed the eyes???
@@aibi5532 He mentioned that he used in paint again on the eyes
@@RobShocks i m saying which prompt he used for the eyes
Alright, I tried your first concept here but a bit different. I made a quick sketch and IMG2IMG it and it does the composition really well and is probably quicker than finding all the assets you need and adjusting everything. However, it still seems more reliable to get the background you want and only sketch the foreground yourself.
dunno why I never thought of doing this. I think this is more interesting for compositions that are almost impossible to get through prompts.
Great Idea to Photoshop you a prototype frist. 😁👍
This works for me! This is the way I've been doing it. But you took it to another level with image adjustment. Nice!!!
I really loved the African queen result ❤
I appreciate how he's using Affinity Photo ✨✨
Back to the roots! Love to see it
nothing better than some good old basics
This's my daily workflow. First, I start with my own creation, so no copyright infringment, and second I've fast multiple variation to choose. Just, I keep the result closer to my style to make it more personal.
Hey Olivio - all of your videos are awesome. This one was especially awesome. Really enjoyed seeing this end to end process for how you use these tools together to bend pixels to your will! Funny intro was a nice bonus too. :)
Dude this is amazing, you are filling in the gaps for me. I use photoshop instead but it's very similar process.
thank you :) happy i can inspire you
I've been doing this technique sense style transfer.
It was a way to cover up completely janke Photoshop cut and paste.
awesome!
Absolutely LOVED it! Thank You....
man i was waiting for this tutorial , many thanks and congrats
Someone needs to remix that intro.
Thank would be amazing :)
… with an AI!
Soo good!!!😂😂😂
Amazing, eye-opener! Thanks for sharing!
Terrific tip and end results!
best pedagogue on SD by far ! 🙏
Thank you :)
MINDBLOWING
You could have compared some of those in-between images side-by-side and maybe zoomed in a little. I couldn't tell the different sometimes. Thanks for the video though!
I'd love to see a video showing how this stuff can be done in GIMP.
great work, i do how we can get to a point soon where we dont need to take it to photoshop for it to better understand the context of what we want
It's annoying trying to do this in midjourney and leanardo a.i. I tried it, it's a mess. Going to have to upgrade my system to use stable diffussion because I'll have far more control with it. Great video and really helpful. I do a lot of photobashing in photoshop so this is ideal for my work.
Exactly the same here
I started out in leonaradi ai, then went to easy diffusion then automatic1111.
I'm more excited about a future where we can do similar with 3D-printed full body replacements with an AI surgeon as part of the deal where what you make can be your new body.
What would a world where anyone can be any gender, any color, any shape, any apparent age, any and every option choice one could think of?
Would such a thing end sexism, racism, ageism, and other biases and prejudices?
Amazing! Your natural imagination, talent for prompt creation, and composition is a treasure to us
I keep coming back to this video because it lo0oks like the only way to get close to a Midjourney 5.x quality in stable diffusion. The composition is stable diffusions weak point as far as I can see so composing an image this way seems promising.
Great video! Thank you!
short but powerful video
I have been doing this as an architect , and it is just crazy
Thank you :) I love to play with this
I started just photo shoot myself, its the best way to get correct pose
amazing tutorial THANKS
This is pretty much the idea of what artbreeder collage had started with isnt it?
Great video, I'd love to see how the bear and girl is done, when will your live stream be, thanks again, great content.
Thank you. I will do it this sunday
@@OlivioSarikas Awesome, see you then.
Great rap brother love it! 0:45
LOOLLL that rap!! Hahah!
That was one of the most random yet effective intros I've seen in a while 😂
Amazing art! Wow, I really like it!
Thank you very much
Thanks 😌
you coould usw 'select subject' option fron select submenu, would save a lot of time
Great video as usual!! how do you use multiple Controlnets? I only see one tab on my A1111
I would probably listen to the full version of your intro song
Fantastic Tutorial....!!!!
Thank you :)
Congrats, you discovered concept art. :D This is how most concept art is done, just with 20h of noodling instead of Ai.
Спасибо, мужик! Ты крут.
Thanks Olivio! Didn't know how cool tile was
Tile is super awesome! (when it works)
Your videos are outstanding. Thanks!
Loved it! Thank you for the tutorial
Sir, You dropped this 👑
funny I was thinking this technique might work well with SD but didn't know it had an official name (photo bashing). Someone should make an online photo bashing tool to help people create quick mockups of what they want.. or maybe someone already has made one :)
you are the best!
Hey there, would really love to know why you choose to not use pixel perfect on the depth, other than that stellar video as always !
for the photobashing your using with the winged girl what program is that photoshop or photopea? You skipped the part on how to take the image to Stable Diffusion
Great video!
Great video
Wow. That was so impressive. Thank you!
Thank you very much
"Right click on our layer". It may be worth mentioning to the viewer what tool you use, before jumping into your use of that tool.
Edit: deleted segment anything.
i did, but that got lost with the first part that i accidentally deleted. It's affinity photo 2
that's what I'm talking about, the fundamentals didn't change, it's still the Artist that will take advantage of the a.i. tools. not the common folks. A.I. cannot simply steal an artist JOB 😅😊
Olivio, tsk, tsk, a goose? Really? I went and grabbed the wings off a turkey vulture! Now that is a real bird! Thanks for the vid, I learned a lot!
Hi Olivio - great video as always - could you comment on what photo-editing software you feature in this video? Is this your preferred editorial software?
Never mind! I just saw Affinity in your comments! Thanks!
Hi Olivio! Thanks for your tutorials. Any idea how can I generate a nice professional studio-like LinkedIn profile pic from my selfie? I fail to do it in Midjorney - the resulting pictures don't look like me :)
so basically does the inpainting need to have the same prompt if you wanna change just 1 thing? (For example, a lady is holding her hair in the final image)but the hand has a couple of extra fingers, i paint just the extra part, but the fingers still come out crooked, does inpainting have a feature that you can remove extra fingers? or do you need to be really specific as to painting the area?
If you keep the full prompt it will try to put say, another entire woman in the inpainting space, so along as there's high enough denoising. Most people I've seen, and myself use a simple description of what you are trying to change, maybe with the style included maybe not.
For hands I'll usually try to inpaint the entire hand or all the fingers. Or use Lama cleaner or photoshop to roughly get rid of anything extra then inpaint it.
There's also a "bad-hands" negative embedding that can somewhat help.
those were also prompts, in *clicks* form
It used to be we would say "There's an app for that."
Now we're saying "There's an AI for that."
Meanwhile, humanity is getting dumber by the day.
love your vids
Do you know why I don't see tile_resample? I only have tile_gausian I have controlnet 1.1 and everything is update in the check for updates tab 😕
"Happy little accidents" - Bob Ross
Thanks Olivio 👍👍
Happy little accident - That's what my Mom calls me ;)
Great Tutorial M8, Just Subbed!! Gratz for your channel!
Thank you
Maybe review what tools you are using? How to set them up? What image manipulation software are you using?
May I ask which AI Generator are you using in this tutorial? It does not look like Midjourney.
Hy, thanks for the tutorial, could you made a tutorial for vector Ai ? Thanks
Why not use Krita with it's automatic1111 extension?
Shuri would be
slightly impressed
Which scale factor for face restoration?
I wish Olivio to earn with the help of neural networks for the apartment he has in the background! :)
This will skip manual cleanup and polishing that is needed after photo bashing the concepts. Lovely.
I'm not gonna deprive myself of the lived in experience of painting it myself.
Who said you should?
@@OlivioSarikas fair enough.
Im still trying to learn all of this. Why did you set the pre-processor resolution to 1600 rather than using "Pixel Perfect"?
AGI: Secretly tricking us into photoshopping for it. 😂
lol, true :)
Boss sauce!
Thank you very much
Amazing as always! Thank you again!
I saw that you worked on Affinity Photo there and you did tutorials on it before working with generative AIs. For a starter: is it an equivalent to photoshop?
Not equivalent, but it can probably do 95% of what most people need.
Nice video, can you show how to fix the eyes for the African queen part?
4:28
He did the same for the African queen.
Many Thanks for a very informative and entertaining youtube channel.
In outputs folder you have all images you think lost ;)
Hey Olivio... Really awesome demo... what editor are you using?
It's called Affinity Photo 2
sorry I don´t get it :-( you do net mention in the video when you switch programs and which programs you use for what
You never said what editor you are using
@5:41 the wings don't look attached to her.
It looks like there is a bird behind her.
lacks perspective.
inpaint doesn´t work anymore for me the image dropbox disapear when i put a picture there
Olivio, did you ever had success inpainting midjourney generated images to fix details? I get strange results
Hi, this might be a silly question but could you tell me how you imported your jpeg from photoshop into "img2img"?