Pose Editor for precise Control - AI painting in Krita
Vložit
- čas přidán 1. 11. 2023
- For a specific dynamic character pose it's best to have precise control. Create, edit and make adjustments to pose skeletons directly in Krita now!
Krita plugin: github.com/Acly/krita-ai-diff...
Uses Stable Diffusion & ControlNet.
Video runs at 4x~12x speed. - Věda a technologie
Very well done, much appreciated! Like this I can really understand the process, thanks
Finally, I was waiting for a cool Krita/SD integration, this may be it 🎉
Thanks for sharing, got the plugin. Awesome work!
This is downright astounding! My mind is blown outta my head! 😳😳
Get some paper towels and clean up the mess!
LOL
when I use control pose I get the layer but no stick figure, any idea why?
You are a hero
Great job dude!!
Very-very nice.
Genius!
amazing!!
Awesome
This is fantastic! Thank you for sharing this!!
Can we use the same seed for different poses? (so we can generate character consistency in all angles)?
Seeds don't work like that. Subtle changes will be similar, sure but any kind of significant change to any other part of the process and you're getting different output.
Hi Sir, great work. How do you turn on the control pose. I have install the plugin in Krita. Thx
Thank you. Is it possible to switch poses while maintaining the character?
👏
2024 will be ♾ABUNDANCE!
How do you paint the older high detail generations with a brush onto the new image? I couldn't catch that step it just felt like applying and painting the layer.
I use the eraser to delete parts of the upper (new) layer.
@@aclysia Oh felt like something advanced. Thankyou :D
nice work. whats u grapich card model ?
For some reason it doesn't detect my nvidia card nor generate any image even if CPU generating option is selected. Anyway, looks fantastic and hopefully it will become a core feature in Krita!
Hey, is there a way to create a skeleton? Instead of having to put a random image?
just type in prompt a skeleton
@@SomeSimpleGaming Lmao! I realize now my comment is misleading. I meant an OPENPOSE skeleton ^^'. The colored character stick.
I realized the problem, by the way. In order to create an OpenPose, you need to use the preprocessor first. As long as you don't do that, clkicking on the "Add a character pose" does nothing.
After 2:48 seconds, when the mouse traces the image, it looks better. What do you do to make this happen? I know that the brush is used as a spray, but I don't know what else it does.
It's just an eraser and reveals the previous state (layer below). Often new generations improve some part of the image, but change other things you don't like.
How about some character consistency? Instead of creating random characters, I want to add my own original characters to the scene.
You need the 'Reference_Only' ControlNet for that. Then you use an example image of the character as the input, but it only seems to work well with humans, probably because that's what it was trained on. But in order to get the best results, you'll have to train your own LoRA model on multiple images of the same character from different angles and scenes, then use that in your generations. The final way is to use Roop or another face-swapping tool, but for that you'll need Automatic1111 or ComfyUI, or a stand-alone program. These tools can take a single image of a human an put that face on any other portrait image, but both eyes and mouth must be showing for it to work and highly stylized images don't usually come out well, so use a 'photorealiscic' source and generated image.
When I try to move the points it just transforms the shapes instead of moving the skeleton
And when I undo it removes the lines
For me it was the same. The next day it was working properly, don't know exactly why. I realized if I move the pose-vector and then I use the select shape tool it doesn't work. But if I use the tool immediately without moving the layer before it works fine.@@QareenProductions
Why aren't you using the lcm live generation plugin
Cause the quality is ass?
Nice Plugin. but.
My Krita send me this error:
Error during server startup : Traceback (most Recent call last) "C:\...\.server\ComfyUI\main.py", line 69, in import comfy.utils
File "C:\...\.server\ComfyUI\utils.py", line 5, in import safetensors.torch ModuleNotFoundError: No module named 'safetensors' [1]
Could you please help me with this error?
This indicates some python modules are missing that should have been installed automatically. Was server installation interrupted or aborted somehow?
There was a similar issue recently: github.com/Acly/krita-ai-diffusion/issues/33
If this doesn't help, please post your client.log file (find it by clicking Show logs) so I can figure out what went wrong. Best open a GitHub issue, it's difficult in YT comment.
w0w
omg
All that Krita AI plugins have a HUUUUUGE disadvtantage: you can't specify the folder where your AI model resits. So if you have tons of them already installed because you are using ComfyUI or A1111, you have to download ANOTHER 10GB into the Krita plugin folder. That's a waste of resources! I just noticed ONE Krita plugin where you could specify the model folder - but guess what, it did not work 😐
symbolic links
Uh, no, I didn't have to do that at all. Worse thing I had to do was rename and shuffle a few models(IPAdapter) into subdirectories. That was a few months ago.
From what I see , the open pose doesn't really help to create the character to be a pose, you have to paint over it any way, Doesn't that defeat the purpose. I am sorry maybe I missed something here.
Without the rig, it's way more open for interpretation and can easily rotate the character along any axis, meaning you can end up with them doing things you don't want like facing away and doing handstands.
Has art become too easy? Will there be a point where anyone can produce art just by thinking?
Still a loooong way to go haha. It all seems awesome for a moment, but our ambitions and expectations will soon catch up.
Always has been. Art has never been defined as "something hard to do". The only thing AI does is allow anyone to have high quality without wasting hours every week. It will also force the industries to stop producing garbage, as they'll have no excuse anymore.
Great! I would like to generate reposes for Krita, it would be very interesting to use it with the timeline
czcams.com/video/SacK9tMVNUA/video.html
I am very disappointed. You don't learn anything and just make the computer work for you. It might be faster and easy, but it won't become a job.
This is my opinion. Don't attack me
Adapt to changing technology.