Wait, what? HuaPaint works great for outpainting. I have even outpainted MJ images with SD with it. Been using it for a month now with perfect results.
You mean PaintHua - yes, i need to make a video about this too. But with this method you have actual control over what will be in the parts you create :)
Great technique! Outpainting is usually a disaster for me as well. The only thing I caught that you should be wary of is the resolution, the original image was 1024x1024, and since you inpainted at 704x704 you lost out on some detail! Thanks again, I'll be using this method for sure.
Oh i’m super excited to try this out! haven’t had the best luck with out painting in the past but with trying to make a comic it’s definitely essential to figure out!
Great tutorial. If you GPU doesn't have enough RAM, you can get better quality if you paint top of left left side separate from bottom of the left side. The AI network will always render a square patch internally and if your to-be-filled area has roughly 1:2 aspect ratio, half the AI work will be lost / resolution will be half it could be.
Good video, is there a chance when you use automatic1111 , you use darkmode? its real hard to see with the bright light it sends out. Just a suggestion.
Interesting. I'm sure I'll be back to watch again when I need to do this. I very naively tried outpainting earlier today with laughable results. I'm anxious to try again (but Dreambooth training is keeping my GPU busy this evening).
Basic stuff... Is surprising that people need a video to try out this workflow and they didn't figured out alone. More tips: Export the inpaint/outpaint mask from your drawings software (no need to get mad with web UI mask tool) and add noise layer on the part that you need to repaint with a mask to get much more detailed controls on the outpaint. We call in our studio this technique : "paint with noise". Don't let SD mange the noise when you can doit ;-)
2nd tutorial im watching today where the controls have obviously changed from that version of auto1111. the inpaint at full resolution does not exist anymore and inpaint seems to have changed. maybe an updated video as this one does not have relevant info anymore?
The software is called "Affinity Photo" but you can also do the same in the free "Krita" Software. And the Stable Diffusion is a local install called "Automatic 1111"
@@OlivioSarikas I have Photoshop, I was asking for the later, I'll search for it, it links to your account on Stable Diffucion for charging credits? Thank you!
Hi! Thanks for your wonderfull tutorila. (btw I think it would be better if intead of suggesting quantities, you actually explained what each slider is for,).
I dont know why but area where Im trying to add something has a little bit chabged color. Therefore I can see initial rectangle in center and outpainted area with a little bit different color ((
Great results! Now this is something Automatic1111 could improve on so that the additional step of resizing and image editing in photo editor is no longer needed. If they ever implement something akin to Nvidia Canvas, that would be huge gamebreaker for Stable Difffusion.
@@OlivioSarikas I had to roll back to one of your previous tutorials as I didn't install the sd-v1-5-inpainting.ckpt yet. Also, I just notice... nice...good job on the astronaut's hands! 😅
I tried it and it was very bad. You have to change the prompt a great deal from your original, maybe even preview the seed and the new prompt to use for inpaint. You also do not show the render, which means you tried alot. The masked area is way too big. There is no way you got this level of detail with 704x704 in one pass. What you actually have to do is loop the output back into the input several times for many many more masked areas to get a detailed result. This method here might work for comic and limited color palette. It certainly did not work on the more realistic images I tested this on. If you don't adjust the art style, even when using the same seed, the weights will change and the artstyle might look very different from what your original picture looks like. IT WILL NOT replace the prepainted color blobs, instead it will mix the new render with it which looks utterly terrible in anything but comic style pictures. So do a test render of your seed, adjust the style and loop at least 3 times to get a result, but then still it will need alot of work because it will be pretty pixelated, might even have to adjust the mask blur. Either it is a bug or I am doing something wrong, because even with denoise set to 1 it will not erase/replace the prepainted blobs although it uses the colors, but the new render looks like it only was overpainted with way less denoise
@@OlivioSarikas i tried with this one and nothing happend, so I downloaded the model and put in the models folder, let´s try it! thank you man! nice channel!
My AI Facebook Group: facebook.com/groups/theairevolution
My Instagram: instagram.com/oliviotutorials/
It's actually better that just Outpainting because you have control of the composing. Great idea !
Nice one! I've been throughly disappointed by SD's outpainting so far, but gave it another shot with your method, and it actually works really well! 👍
Wait, what? HuaPaint works great for outpainting. I have even outpainted MJ images with SD with it. Been using it for a month now with perfect results.
You mean PaintHua - yes, i need to make a video about this too. But with this method you have actual control over what will be in the parts you create :)
@@OlivioSarikas Yes, it has the same painting features... much more powerful.
@@Theexplorographer I see what you mean. Good point
@@OlivioSarikas I look forward to your video!
I used your trick on a MJ image and it worked like a charm! Thanks!!
Great Job as always Olivio
Thank you :)
Never used inpainting this way, very useful information, thanks!
MANY THANKS
UNBELIEVABLE TECH...
You're welcome!
Thanks for the advice, using inpainting instead of trying to make outpainting work helped a lot!
amazing and non-obvious way to fill in the missing areas, thank you very much!
Great technique! Outpainting is usually a disaster for me as well. The only thing I caught that you should be wary of is the resolution, the original image was 1024x1024, and since you inpainted at 704x704 you lost out on some detail! Thanks again, I'll be using this method for sure.
That's a good point. You can set it to a higher resolution, but the rendering will also take longer
You saved my day! THANKS!
Oh i’m super excited to try this out! haven’t had the best luck with out painting in the past but with trying to make a comic it’s definitely essential to figure out!
Give it a shot. It works with all kinds of images. not just comic style
Great tutorial. If you GPU doesn't have enough RAM, you can get better quality if you paint top of left left side separate from bottom of the left side. The AI network will always render a square patch internally and if your to-be-filled area has roughly 1:2 aspect ratio, half the AI work will be lost / resolution will be half it could be.
ah, awesome trick that makes so many things possible! thanks
So helpful! Using this technique with the Painthua app with great results! Thanks!
c'est magnifique !! bravo
Good video, is there a chance when you use automatic1111 , you use darkmode? its real hard to see with the bright light it sends out. Just a suggestion.
hey, nice work.... can you give me the link to the inpaint model ?
Olivio!!!!! I love you dude... Thank you for this.
Thank you :)
hey man! Great video, noticed you have switched from photoshop to affinity,
Any reason why that is?
Very useful trick! Thanks for sharing!
Why not pick center spot in resize canvas instead of centering layer afterward ?
Really very very useful tip, thanks!!👀👍
Cool !
can I ask you for your GPU , what GPU did you use for run SD smoothly like this,
Best AI Content!!! Thanks.
Thank you, so nice to hear that :)
Very smart! Awesome results thanks 🙏
My pleasure!
Interesting. I'm sure I'll be back to watch again when I need to do this. I very naively tried outpainting earlier today with laughable results. I'm anxious to try again (but Dreambooth training is keeping my GPU busy this evening).
Great video. For the people that have an old machine and where SD can't be installed, is there a web solution? Thanks in advance
don't give your money to crooks websites who make you pay to use something that's free.
Save money. Buy a decent Computer.
Basic stuff... Is surprising that people need a video to try out this workflow and they didn't figured out alone. More tips: Export the inpaint/outpaint mask from your drawings software (no need to get mad with web UI mask tool) and add noise layer on the part that you need to repaint with a mask to get much more detailed controls on the outpaint. We call in our studio this technique : "paint with noise".
Don't let SD mange the noise when you can doit ;-)
could you send me a email? i would love to talk more about "paint with noise" technique and also love to hear about your studio
wait wait huh!?
you meant you select "upload mask" and use the mask we made from Art Editor Software(PS,AP,GIMP)?
2nd tutorial im watching today where the controls have obviously changed from that version of auto1111. the inpaint at full resolution does not exist anymore and inpaint seems to have changed. maybe an updated video as this one does not have relevant info anymore?
Awesome video, thank you
That is amazing, thanks a lot for this trick !
Where exactly you did that? Thank you!! great job!
The software is called "Affinity Photo" but you can also do the same in the free "Krita" Software. And the Stable Diffusion is a local install called "Automatic 1111"
@@OlivioSarikas I have Photoshop, I was asking for the later, I'll search for it, it links to your account on Stable Diffucion for charging credits? Thank you!
PaintHua let's you have much more control over the generation!
that's also great. I will make a video about that soon too :)
Cool stuff!
Hi! Thanks for your wonderfull tutorila. (btw I think it would be better if intead of suggesting quantities, you actually explained what each slider is for,).
"You could not live with your own failure... and where did that bring you? Back to me."
- Drawing
i also thought of this workflow, we think alike.
awesome. and i am really happy how well this works
I dont know why but area where Im trying to add something has a little bit chabged color. Therefore I can see initial rectangle in center and outpainted area with a little bit different color ((
Where can I find this SD1.5 inpainting file for download?
Great trick!
Thank you!
It looks like the outpainted sections are lighter and less detailed. Is that just the contrast?
Very cool trick....
Great results! Now this is something Automatic1111 could improve on so that the additional step of resizing and image editing in photo editor is no longer needed. If they ever implement something akin to Nvidia Canvas, that would be huge gamebreaker for Stable Difffusion.
Since the UI has changed a lot, maybe a new video for this on the new UI?
Beautiful video as usual Olivio! What happened to my blue friend at the end of your videos? Please bring him back!
been doing this with after effects😄what ever works for you, as you said :)!
Nice video
Is the picture at the end 2048x1024? If yes, it is posible to do even more resolution like in dalle2?
yes, but you can always upscale :)
Smart move !
At 3:00, you say "next, we go to the browser where we have installed 'automatic 11 11' ". what is automatic 11 11?
Seriously ? What are you even doing in here dood （´∀｀ ）
@@HanSolocambo learning. You?
Ingenious! Thanks Olivio!
Thank you :)
@@OlivioSarikas I had to roll back to one of your previous tutorials as I didn't install the sd-v1-5-inpainting.ckpt yet. Also, I just notice... nice...good job on the astronaut's hands! 😅
How much VRAM did this take?
brilliant
👋 It's about time we're getting an Outpainting that has some promise
totally. the classic method is just too random
What is outpainting?
How did you manage to get a perfect Hand out of Midjourney?
Random Florentin Will appeared
😂
Can't you just do it all with inpaint sketch, which allows you do draw with different colors
I'm giving you a like just for using affinity and not adobe. lol
My Eyes 😵
FYI apocalyptic is pronounced "a PAW kah lip tick", though I may start calling it Apo-Calyptic lol
its got the perspective wrong.
I tried it and it was very bad. You have to change the prompt a great deal from your original, maybe even preview the seed and the new prompt to use for inpaint. You also do not show the render, which means you tried alot. The masked area is way too big. There is no way you got this level of detail with 704x704 in one pass.
What you actually have to do is loop the output back into the input several times for many many more masked areas to get a detailed result. This method here might work for comic and limited color palette. It certainly did not work on the more realistic images I tested this on.
If you don't adjust the art style, even when using the same seed, the weights will change and the artstyle might look very different from what your original picture looks like.
IT WILL NOT replace the prepainted color blobs, instead it will mix the new render with it which looks utterly terrible in anything but comic style pictures.
So do a test render of your seed, adjust the style and loop at least 3 times to get a result, but then still it will need alot of work because it will be pretty pixelated, might even have to adjust the mask blur.
Either it is a bug or I am doing something wrong, because even with denoise set to 1 it will not erase/replace the prepainted blobs although it uses the colors, but the new render looks like it only was overpainted with way less denoise
facebook no.... please, reddit!
I´ve this model v1-5pruned.ckpt [a9263745] this inpant is already inside it? plz
you can do inpaint with it too. but there is a separate model for inpaint that will give better results
@@OlivioSarikas i tried with this one and nothing happend, so I downloaded the model and put in the models folder, let´s try it! thank you man! nice channel!