PIXEL ART with StableDiffusion + Tileset workflows??
Vložit
- čas přidán 31. 05. 2024
- Creating pixel art with stable diffusion can be very simple.
That's why we explored ways to use it for games, or other general purposes. Like backgrounds, characters, assets and tilesets.
Finding workflows that serve as a base for you to make your own, more specialized ones.
Join our discrod server: / discord to learn about this and more!
Darkstoorm's twitter: / darkstoorm_
Darkstoorm's og guide: gist.github.com/DarkStoorM/4e...
Character Sheet Controlnet references: ko-fi.com/s/65ed598d96
+video on how I make them
------------ Links used in the VIDEO ---------
Pose the characters as you like: app.posemy.art
Checkpoint SD 1.5 (if it doesnt exist anymore, ask on discord): civitai.com/models/195730/azi...
LORA for SDXL:
civitai.com/models/120096/pix...
Extensions used in the video:
Pixel Art: github.com/mrreplicart/sd-web...
(you could also use: github.com/AUTOMATIC1111/stab...)
Regional Prompter: github.com/yankooliveira/sd-w...
Controlnet: github.com/Mikubill/sd-webui-...
Asymmetric Tiling: github.com/tjm35/asymmetric-t...
Fixing the duplicate window issue (by darkstoorm):
"Since this is an old extension, it might appear twice in your webui. To get rid of this, you have to remove the rendering function:
file: sd-webui-pixelart/scripts/pixelart.py
def show(self, is_img2img):
return scripts.AlwaysVisible
After enabling it in Postprocessing and restarting webui, the duplicated tab should go away"
------------ Useful stuff for this ---------
Settings and workflows may very depeding on what you are looking to do. As well as which model you are using.
If you just want to use SDXL, controlnet tile might act differently.
lospec.com/palette-list
Vae by thomaseding that can help with artifacts:
civitai.com/models/366568/vae...
------------ Social Media ---------
-Instagram: / not4talent_ai
-Twitter: / not4talent
Make sure to subscribe if you want to learn about AI and grow with the community as we surf the AI wave :3
#aiairt #digitalart #automatic1111#stablediffusion #ai #free #tutorial #betterart #goodimages #sd #digitalart #artificialintelligence #latentcouple #couple #composableLora #posing #controlnet #SD15 #inpainting #openpose #depthlibrary #AI #midjourney #interaction #relation #comic #storytelling #pixelart #pixel #game #character #animation
0:00 How to create PIXEL ART
0:28 Intro
0:43 How to use the checkpoint
1:16 Repeatable backgrounds
1:54 Hires fix, yes or no?
2:30 Limiting and controlling colors
3:14 Character and animation
4:30 Bad and good news for characters
5:04 Inpainting
5:22 Thanks to Darkstoorm!
5:48 Img2img tips for Pixel Art
7:35 Finding new methods + ours
7:55 Asset sets and Tilesets
11:32 Other styles?
11:50 Thanks for watching!
Already watched before release and enjoyed the humongous channel spam 😎
I hope this gives people a bit more clues on how to get some help from SD with pixel art for game assets, not necessarily getting a production-ready quality, but something they can work with faster. It might be a bit questionable, but at least in my case, this will help me with learning pixel art a tiny bit by working on the base image, because I'm not very creative when it comes to drawing from scratch, even though the sprites are 16x16 🥲still, feels nice to correct SD outputs
I've only managed to watch about 4 minutes of video so far, but I've already heard more useful information than in many 20 minute videos. 12 thousand subscribers for this channel is clearly not enough, you need more, deservedly more.
thank you so so much!!! Really appreciate the kind words, hope it is helpful!
thanks for using my checkpoint.
Thank you for making it!!
You have a great sense of humor my dude.
Thank you!! 😂😂
I wish every tutorial was as straight-forward and easy to follow as yours are!
Thank you so much!!!
Great video! For me the most difficult thing to achieve with AI image generation is consistency. I would love to see a project which successfully achieved to create all the assets but maintaining massive consistency,
Agree, it's something to work on atm. Bc with no lora or assets that you can implement afterwards, its kind of hard to get that. As well as super important
And tyty, btw!
I was literally thinking about this and your video appeared, uploaded 34 seconds ago. hahaha
wtf hahhahahahaha, hope it helps!!
again a masterpiece tutorial. plus it's one I was actually struggling with so yeah wish I could thumb up twice
Hahahaha glad it was helpful. Thank you!!
Hope you come back with more interesting stuff and great tips, your videos are always top quality.
tyty!!! I 100% will. I have a video pretty much ready, but I need confirmation from a 3d party before publishing it.
Idk how well the vid will do, but it's a topic I loved playing with, and has some cool stuff in it too imo
@@Not4Talent_AI Your work here is amazing, truly showing what AI is capable to do, we need more people to know that this is a real tool and how complex it can get.
@@ososimiospub6406 welp, happy that you say that bc it's pretty much where the next videos are going. Or at least where I'll try to head them haahaahha
tyty for your support!!
Aye a great collab of DARKSTOORM. guy to ask in R&D pixels'n background.
Thanks for sharing 😊.
I need to watch it later.
Outstanding work!! Give me a lot of help, I sincerely thank you!!!👍👍👍
Glad to hear that. Tyty!!
awesome video, thanks!
Tyty!
Probably one of the best AI tutorials I've seen in a wnile. Thank you so much for your quality work 🙏
Thank you so much for the kind comment! Hope it helped 👍
Hi! I know this has nothing to do with video's topic, but I've been watching your tutorials over the past weeks learning to use SD from 0, and you are truly amazing. I've learned so much that I'm already working on a project I planned for so long! But there is something I've been wondering for a while, especially while watching the ultimate hands tutorial, and it has been: "HOW DOES HE DO THAT WITH PHOTOSHOP!?"🤯the only thing I know is how to cut out an image, I've always wanted to learn but I have no idea what to look for on YT! I know your channel is not focused on teaching PS, but I really want to learn how you touch the parts you want to fix in order to help the AI, for photobashing, erase artifacts, correct hands (as you did with Katara's example long ago or on the ultimate hands tutorial), change colors, blend images, you know, that masking thing when you erase backgrounds and make them fit on the image so then you put it back on SD haha. I don't know if I explained myself correctly but, maybe is there a tip you could give me to look it up on YT? A course or a specific video to learn how to do what you do to fix images on PS? Thank you for your top quality content as always, I'll definitely come back and watch this
hi!! Thank you so much!
I still need to learn a lot too, but if I had to name some stuff to lean it would be this:
(first, this is one of my fav. photoshop learning channels www.youtube.com/@PiXimperfect)
-How to use photoshop layers
-Photoshop basic tools
-Selections
-Masking
-Free transform
-Adjustment layers
-Blending modes
-Filters and image adjustments
-Brushes and how to paint
Tbh most of the important stuff would be easy to learn over the course of a week by watching the pinned series on the Pixiperfect channel (Photoshop for Complete Beginners | Lesson 1)
After that is just practice. Learning how to draw over all will also be a massive help with retouching AI images. (assuming you dont already ofc).
Mainly how to paint, and blend.
The most important parts for retouching are:
adjustment layers and filters--> for the over all look of the image and blending new stuff together
Layers and blending modes --> to work non-destructively and be able to add stuff
Learning about clone-stamp tool + content aware fill + generative fill --> you can do this in SD too, but sometimes it's better directly in photoshop. It's to erase stuff
Selection, brushes and clipping masks--> to paint new stuff or correct things without affecting the rest of the image by accident.
General knowledge of the tools and what is possible--> to know how to approach a problem. Even if you dont really know how to, knowing all the tools might give you a starting idea. Like "oh, I remember a tool for this, lets look up on youtube how I can use it for this scenario"
If you ever have any doubts and I have a bit of free time, feel free to dm me on discrod and I'll try to help. ( at the best of my ability ofc, I'm no pro hahaha)
Great video! Thanks!!! ❤
Tyty!!
this is so epic, thank you good sir
thank you for watching! hope it helps👌
Awesome!
Tyy, hope it helps!!
thank you for this! This is very helpful for creating animations
glad it was useful!
@@Not4Talent_AI i hope for your next video, you please make a tutorial on how to train concept loras like example, a concept lora for "hugging" so the characters if using this "concept" lora the subjects will always be hugging.
Thank you so much. Your tutorials always my go to for help!!
Tyty!! Srry to say that, if I make such tutorial it will be in a pretty long time. I need to make some other ones first. And then Id need to form myself on that topic.
I think it isnt that different from training characters though. Pretty much having a big dataset and tagging anything that serves a purpose
@@Not4Talent_AI ok thank you!
np!! if I find a good tutorial for what you need Ill let you know!@@billyn7757
Cool!
tyty!!
yeah
always a GODLIKE content.
thank you so much!
Do you know if it is possible to use these generated images for a commercial game? Thanks for the video!
The law isnt really coght up to this yet. So idk.
Licening on the model I used says you can't, but it's also, I think, not really enforcable. (Not a lawyer tho😂)
In any case, if you can demonstrate that you manually retouched and planned the assets, I think you could.
Just as you can use images that are not yours if you retouch them enough.
My guess would be that, as long as you create the base input, and then retouch the output to fit industry standards. You should be fine
I do know that Steam allows it to some extent since january.
Label have to be disclosed with like a ai tag or some sort
Can do all these with normal model meaning non pixelated stuff too
ofc, nothing special about pixel art workflows. just that you use a pixel art model. and that some of these dont make much sense without pixelart hahahha
I coincidentally posted a VAE the other day to fix pixel art aliasing when the model generates 8x8 "pixels". (On both Civit and Hugface)
oh sht really? do you have a link? May put it in the description of you let me
@@Not4Talent_AI CZcams will block the comment links I post. It's the same username I have here on the other platforms. Feel free to post the link yourself if you find it.
Since this is pixel art, what happens if you just scale everything proportionally instead of up-scaling? Generally the problem with scaling an image normally is that it just duplicates local pixels, but that would be ideal in this case right?
Depends on of you want to upscale the image and give it extra detail. For that you need either img2img or highres.
If you just wsnt to resize, using "nearest" and resizing should be enough
This method is okay but it takes way too long compared to pixellab
not sure what pixellab is but sounds interesting in that case. ty for sharing!
Ai community is the best community
Why are all generations in thumbnails with their mouth open, it always looks wrong.
easy to do I guess xD
Finally, the game development with one button!
Hahahhaha idk about that one, but at least its one step closer 😂
if we only could fix the consistency problem that comes with the generation of character stuff