How to Make 3D Game Assets with AI (Free and Easy) - Stable Diffusion + Blender Tutorial 2022
Vložit
- čas přidán 19. 06. 2024
- How to Install and Use Stable Diffusion (June 2023) - Basic Tutorial
• How to Install and Use...
Other options for using Stable Diffusion: / dreamers_guide_to_gett...
CHAPTERS
0:00 Intro
8:30 Iterating Results in Stable Diffusion
19:04 Blender
22:17 Adding the Texture
50:53 Cleaning up Mesh and Textures
55:40 Baking the Texture
58:36 Editing the Texture in Stable Diffusion
1:01:29 Texture Painting
1:03:40 Creating a Normal Map
1:05:02 The Final Result
----------------------------------------------
Did you like this vid? Like & Subscribe to this Channel!
Follow me on Twitter: / albertbozesan
This is so incredibly helpful for solo game devs, I'm so glad I've stumbled upon your channel! Keep it up!
handpainted details are pain in the ass if you are not a digital painter or illustrator. but with the power of AI and your tutorial it become easier. thank you
This is so cool. Especially like the idea of bringing the baked UV map back to Stable Diffusion. Thank you very much!
It's amazing the work you're doing with this video series on how to implement Stable Diffusion into your workflow. Thanks
Incredibly detailed and neat ... I love your tutorials. I couldn't deploy Stable Diffusion in my computer yet, but definitely I'll do it.
Yes, glad to see the video made!
This workflow is amazing! Thank you for the great tutorial
This is awesome. Can't wait to see how tools that make this process easier continue to develop!
i think in some time soon maybe 1 year we get Dreamfusion to work, its prompt to 3d, its already in google colab, but very slow and many errors.
Absolutely fantastic workflow! Thanks for the inspiration...
Wow. Thank you so much for this. As an illustrator, I was initially against AI generated imagery. Your game assets videos have really opened my mind to possibilities on how to use AI as a collaborator and not just a tool. Looking forward to seeing more of your videos!
Thank you! I’m so glad I could show you some of the potential. Even though AI is very powerful, it won’t replace anyone with a fundamental understanding of art and creative business as long as they’re open-minded.
amazing stuff!
this is great stuff ! cant wait to see what stable diffusion does next, and what you will do too ! keep up the good work :)
Thank you!! I’m glad you enjoy it 😄
@@albertbozesan I honestly been wanting to get my ideas 💡 out there and sometimes the hill is steep for having more ideas than skills ^_^ theses are amazing tools. I think all I'd truly need beyond this is blender/unity classes :)
Incredibly useful approach! Thank you for sharing.
This is also great for texturing faces, you can just screenshot the face from the front and then let stable diffusion make the face look good and then project it back :)
Thank yoooooooooooooou
That amaziiiiiiiiing
I was waiting for videos like this😁
Respect to your good work, thanks a lot
all work thank bro, you are the best!
It’s so fun to see what you do, you are actually the first person I see that make these kinds of videos 😁
One thing I would really love to see is how to make sprite sheets using stable diffusion, think it’s possible?
Keep up the great work!
Thank you! I don’t think sprite sheets are possible yet, because SD makes wildly different results for different perspectives. It would take so much manual work that one might as well just do it the old way.
@@albertbozesan " a sheet of paper with a sprite cola logo on it , by Greg rutowski"
Solid blender tutorial! Try playing with higher CFG values and lower denoise to get more interpretive results that don't deviate from the original shape/silhouette! (high cfg is like screaming the prompt at the AI)
You did it, you genius! TY!
it works perfectly! Thanks buddy
Sure thing! 😄
Yesssssss. I love you bro thankskkkkssss
I was just thinking of trying to mix Blender with Stable Diffusion. What timing!
Awesome thank you
Your workflow with this is amazing. Thanks for making this, it's really informative.
ps: I think that by the time I upgrade from my GTX 970 to be able to use Stable Diffusion (or maybe some other A.I. software), it will probably already support "text to 3d" or "image to 3d" in blender.
Thank you! I’m glad it’s helpful. I can’t wait for this to be easier, too!
Worked, thx
Create a roughness map to make the normal map pop more.
very useful. need more one-click plugin!
Thanks!
Amazing
I figured out how to this with Maya, by referencing this tutorial. Great help, thank you!
That’s awesome!
This is so fking cool.
I liked the video.Thanks to the author
The problem with your blade, it's trying to create a "sword", which naturally contains a handle, even if you say "sword blade", since those two words are not a single phrase, but two tokens, "sword" and "blade". You would need to add "handle", "grip", etcetera, to your negative prompts, to be more explicit, so it becomes more likely to do a "sword blade", instead of "sword" (which includes a handle) and "blade" (which could be construed as a "flavor" rather than a thing).
Good tip, thanks!
Every video is better than the previuos thu so much
thank you so much dude you're a god
Glad you enjoyed it 😄 I try my best
Awesome content thank you!
Because of my lack of skill and patience doing 3D modelling, I prefer to wait one or two weeks and let SD spits some awesome 3d model :p
Thanks! Yeah this was way more work than I was expecting. I’m excited for the simpler solution that should be arriving soon!
i am inspired!!!
gonna try making characters witht his, gonna take a good time!!
Check out this other technique, too! It goes “the other way” and could help you:
czcams.com/video/zukojwUkHz0/video.html
What you called 'back of hilt' is called the 'pommel'. Really useful tutorial, thanks.
Thanks!
Thank´s ;)
WOW!
worked
With ControlNet SD/A1111 extension, this is going to be so much easier.
Yes!
Let's Goooooo
This video shows a lot of ways to automate the process if you're interested: czcams.com/video/RztJGuhC5so/video.html
Also, I'm wondering if there's a way to make an "asset" of sorts for Stable Diffusion. Like say, you want to create a few seperate pictures with the same prop or character, and have the prop/character looking the same in each pictures.
I know it's possible to train the AI to recognize it specifically and be able to recreate it, but that's very effort intensive and requires already having a bunch of photos of the thing. Surely there's a more efficient way to go about it. Maybe by making it into a 3D object that would allow you to manipulate it into different poses and angles easily? At least that's the direction my mind goes.
Great stuff, thank you!
I did try doing this the other way around - making a 3d model first and then feeding that to SD. But the issue you mention did come up, that different perspectives don’t have a consistent style. I’m mostly waiting for SD to catch up, because that sounds like something we will get in the near future.
@@albertbozesan With the speed stuff had been coming out? Definitely. I feel like the bottleneck (so to speak) is less on the tech side and more in that there doesn't seem to be a lot of artists trying to figure out how to make an integrated workflow this way.
Though this reminds me that recently a feature for mixing two prompts came out so maybe that could help? It would be interesting to try combining an image of a pose with a character image. It probably won't work because perspective seems to be a consistent weak point for AI art so far, but may be worth trying something in that direction.
Thanks for the video, I had another version of SD and prefer this much better. One issue I have is the sword gets cut off on the top and bottom. I have "centered" in prompt and even cropped the image in a photo editor and drew back in the cut off parts and added image back to SD but it cut off the image again. Any ideas?
How close does it get to your input image? If it goes beyond the borders despite your input not doing so, try lowering your denoiser strength.
@@albertbozesan It goes beyond the border, I will lower the denoise.
@@albertbozesan It worked, thanks!
Come back to this video after a long time and @24:27 was 🤯
What about it? 😄
at the end you should change the model the 1.5 inpainting one to do the textures in img 2 img for better results!
Not sure I understand. What do you mean?
@@albertbozesan runwayml released multiple models, one of them is called stable diffusion inpainting, that one yields better results for inpainting/img2img, hope it helps!
@@ene_77 I see, thanks! Good tip.
this is amazing please make a video like this but for a full game character would be great!
i follow how to install stable diffusion
why unable to one klik?
so complicated like bureaucracy
i dont know how to continue
so much work for a puny sword! I dont have the patience to drag all these verts around lol
I nearly went nuts a few times figuring out the right way to do it. This is still one of the easy ways 😅
Is this ok to use in commercial games?
The jury is out for a final verdict, but all providers of these AI tools claim yes! Styles cannot be copyrighted. I wouldn’t see why you couldn’t use what you created :)
Using GIMP instead of Photoshop and I have no idea what you did with the brush step.
I'm trying to make a character model and I have no blender experience, so . . .
In Photoshop I just mask over the images to reveal what I want. Painting white into the mask reveals part of the image, black hides it. Does that answer your question?
Reg Blender: I heavily recommend learning more basics like the donut tutorial before trying character art, or you’ll get super super frustrated.
We have dreamfusion that can make 3D meshes that can be imported to blender with textures. It's still a work in progress.
Yeah I tried it out, the results are where the images were about two years ago. It will get better I’m sure.
For once, the software is actually really useful
Your UI is different than my UI. How'd you get that UI?
It should be the default Automatic1111 UI. Did you follow the instructions in the description?
@@albertbozesan yes I did, but I found a different instruction set that gave me the same UI, so I've got it working now! Haha
I believe the back of the hilt is... A weird pommel?
I suppose it is!
People talking about dreamdiffusion but its limited and you have to pay
well I cant use diffusion because I dont have enough credit and I cant set up this engine too
Check out Mirageml.com, you get free credits every month. But yes, the result quality still needs to be developed. Not quite there yet.
Das ist ja mega aufwendig
Puh, also früher hätte sowas ein paar Tage gedauert. Im Vergleich sind knapp zwei Stunden schon sehr gut.
Plese teach me how to install Stable Diffusion in my computer
The best way changes regularly, so I recommend checking out the guides in the video description :) if any concrete questions pop up then, feel free to reach out!
@@albertbozesan I installed yesterday thank you
IS AI GENARETED TEXTURE ALLOWED IN UNREAL MARKET
WHY ARE YOU YELLING
AI is magic
It really is
why not inpaint instead of photoshop?
Inpainting was not as reliable at the time I made this video. Today I would inpaint.
Bro named the guard the hilt 💀
I named a flail a mace in my last video. Totally helpless when it comes to weapons.
@@albertbozesan I only know because I watch shadiversity so...
😂
Did you create a game yet?
I’ve dabbled in 2D and some VR experiments.
Dreamfusion can already make 3d objects from texts
Yes, but it has the disadvantages I mention at the start. Not really usable yet.
I don’t understand how people can see this as a good thing
i dont have to pay a redditor 20 dollars to make me something
@@scarm_rune make it yourself. if you support AI replacing certain jobs it's not going to stop for you