How to Make Game Asset Art with AI (Free and Easy) - Stable Diffusion Tutorial 2022

[@zyzzgang]

This is awesome not only for icons in the inventories, but to make your own references for 3d modeling without being a concept artist and convert that into a 3D asset

[@albertbozesan]

I‘m planning a tutorial on doing this with Blender :) keep an eye out for it and leave a sub

[@hieugao4695]

@@albertbozesan Bro! I'm so excited lol

[@heigonsoldera]

@@albertbozesan Would be awesome

[@OceanSlim]

@@albertbozesan Doing this for 3d models would be amazing. Subbed for the blender tutorial. please deliver!!

[@USBEN.]

AI to generate 3D models from images will solve that soon.

[@vincentkuipers9577]

This method of passing the ball back and forth between AI and artist looks like the future of 2D art taking shape to me.

[@albertbozesan]

Totally! Can't wait to see more people adopt these tools.

[@anonco1907]

The artist won't need so much skill anymore and the art of art will be lost to AI, fr sign me up lol

[@codesymphony]

photoshop is still a very valuable skill, as well as having an eye for what works and doesn't

[@scottyfityoga]

What is this method called? Would these be considered “steps”?

[@michaeledwards6890]

meh

[@ravenoftheredsky]

I've seen this video before. I will watch it again. It's so satisfying

[@jimmyrochette7536]

A truely game changing video and the fact that it free and the content needed is free is mind boggling.

[@onerimeuse]

I can't believe this is fast enough for you to do this in real time. What a time to be alive.

[@ConnorFishcat]

Yo this is so creative! You're essentially kitbashing ai generated art to make something more unique and high quality in multiple steps. This really is a new skill on its own!

[@programaths]

I was thinking that, contrary to popular beliefs, AI will not make artists obsolete, but change their job. This video is a good argument 🤣

[@DarthG33k]

I could totally use this for making art for my CCG. Might have to...

[@ConnorFishcat]

@@DarthG33k Do it!

[@wolfsongcg6497]

How is it a Skill to let an AI create art for you? Interviewer: What is your skill set? What can you bring to this company and it's projects?" Interviewee: "Well, I'm very good at typing words into an AI Art Generator, and moving sliders". Later On, in HR Manager's Office: CEO: "Yeah... We're going to need to fire our current 2D art team, and start using this new AI Generator an applicant brought to our attention".

[@programaths]

@@wolfsongcg6497 Tel me you don't understand creative process without telling me you don't understand the creative process ^^ In programming, there was something called "conversational programming". Even primitive AI were able to guide developer in their work by asking questions based on flow analysis and heuristics. It didn't work well. Not because of the AI being limited, but because the developer still had to bring in a vision and direction. Here, you can see that the artist and the AI works hand in hand. You wouldn't be able to do that job that easily, because you would be much more quickly faced to your own limits. The AI speed up the process and mostly tackle the mundane part, the artist brings direction and vision, he tells the AI what he is aiming for and refine it. The AI can't work it out alone yet, because there is too much parameters involved. Some subjective, other subjective. One simple example is coherence. You need all of your art piece to have some coherence, AI could do part of it (in the prompt, he mention WOW), but what if you need to break the coherence to bring more attention to a part of your design ? Now, the difficulty becomes that all the technicalities are handled by the AI and you are let with the hardest part. So, all it is going to do is to raise the bar and shifting jobs around. Artists who were more "technicians" will have to upgrade their skill set. Artists who had vision and direction will have an easier time. Still, "technicians" will be needed to correct AI as the AI can't always do a good job in labeling/separation. So, those people will have to have an eye to check work and understand how to correct. Still need their basic skill set! Here are the answers: - What's your skill set ? - I am able to provide the vision and direction to an AI in accordance with the target audience and the medium you want to use. - What can you bring to this company and it's projects ? - I have a high success rate in gearing AI toward set exceptions in compliance with expectation set by the company and needs of our audience. Later in the HR Manager office: Holly molly, that's the first guy who really understood the job, we should quickly hire him before competition snatch him!

[@Arcwise]

Here's my workflow which is much faster and less involved, if you wanna give it a try. Set denoising strength to around 0.5 - 0.65 to introduce some variety, generate a bigger batch (like 12 images) and pick your favorite one, (roughly) fix up details in PS and feed it in again using a low denoising strength (0.12 - 0.25) to smoothen things out. That's it, just 3 steps, excluding upscaling. Some other findings: I try to never exceed a guidance level of 8 unless absolutely necessary. Even 5 is fine for img2img. Otherwise it forces the AI put stuff in places where it doesn't belong. Also some keywords seem much more impactful than others (LAION-5B bias?). The more famous/ubiquitous, the higher the influence. Artist names only seem to matter when they are world-renowned. Including art platform names (e.g. artstation) also yields high quality outputs for illustrations.

[@originalShea550]

the obsession with 'workflow' needs to gtfo

[@Optable]

@@originalShea550 It's for progression efficiency sake. Time is the most important asset for us creatives when it comes to deliverables for those depending on us. For messing around and personal study alone, sure.

[@originalShea550]

@@Optable the obsession with effeciency and 'workflow' leads to shoddy results and bad quality. I personally work on jets. We NEVER rush, and will actively get in trouble when we do, because it leads to bad quality, and bad quality leads to people dying. Perhaps game design should learn a lesson here.

[@Optable]

@@originalShea550 The obsession with workflow also leads to masterful results like motion pictures by Christopher Nolan (you can view the editor's workflow in Avid on KZbin), full brand identities for companies like Nike, and the very 3D schematics that engineers used to efficiently build a new jet. You're confusing lazy loopholes with workflow efficiency and effectiveness. It's pretty redundant to suggest shortcuts are the answer to solutions in absolutely any vertical or industry. It's a mastery of tools, not the absence of mastery. Can be both complex and easy to understand at the same time. My family worked on jets and jet engines for the military at rocket labs in Ohio, Iowa, and Florida for over three decades. The latest stealth recon plane being one of those projects. Their opinion on workflow is quite different. Nobody is simply arguing to throw out the extensive time and dedication it takes to bringing those types of projects to life. The essence of latest available software is the culmination of advanced workflow, and they'd argue to the teeth how much easier there lives would have been with the software we have available at our fingertips now to work on these projects.

[@albertbozesan]

I can’t agree with this. Clicking around copying images slowly won’t make me any more creative or improve my art. Any tool that removes purely mechanical steps is very welcome. You mention working on jets. Do you do your calculations by hand? Or did calculators/computers improve your workflow at some point?

[@Fustercluck06]

This is so amazing. It’s like having your own concept artist

[@albertbozesan]

And this vid is old - it’s gotten so much better in the meantime!

[@Conquered]

@@albertbozesan And even better since this reply haha

[@TheChromePoet]

@@albertbozesan This is insane!!

[@TheChromePoet]

It's been 4 months I'm just checking up on you to make sure you're making progress in your game.

[@squitist]

As an artist, I really liked how you took advantage of the AI tech to add intricate levels of detail to your art as a tool. AI art will open up new possibilities or even art genres that weren't possible before by easily experimenting and playing around with details and textures a lot more like you did. I'm really considering using AI tools when I make art.

[@albertbozesan]

Absolutely check AI out! It may or may not work well for you and your process, but trying is always a good thing when it comes to new tools!

[@Scripture-Man]

Just insane. Stable Diffusion has to be a dream. It must be April 1st. My brain cannot handle how a computer can create beauty!

[@Optable]

All of that data farming. All those likes you ever pressed. Those pinterest boards you ever made. Those deviant art pages you ever followed. The work that engages with us is the work filtered through this AI. When great work can be broken down into theory like composition, color, aesthetics, setting, etc. it can be plugged into a computer

[@mohamedsaid320]

Like 2 minutes in and you've solved the issues I was having. I didn't realize that box for showing what was going on was there, I didn't

[@Cesargermanico1980]

I watched your tutorial about 4 times now. I managed to install Stable Diffusion and been messing for some hours now. It amazes me ...... and scares me Thanks a lot

[@Shredzy715]

pls tell me how to do it, I have no experience and can not understand the tutorials linked below

[@g-funny2171]

I appreciate how there’s no ads, thanks.

[@Percavius]

The way you explain this is very easy to understand. Thank you for sharing your process.

[@Espada_Abstracts]

wow, this just opened up an interesting way how I can go about creating a lot of my art designs and concepts.

[@TheRogueBro]

This was just a random recommendation, but this has to be the coolest thing I have ever seen! I'm going to have to mess with this and make some cool DnD assets!

[@xrealitythe2ndx]

You're the reason I started my Stable Diffusion journey around a year ago! My life has changed for the better since then. genuinely, if I had money, I'd send some your way to show real appreciation.

[@albertbozesan]

Thank you so much for sharing that with me!! It’s worth more than money, and cash is better invested in GPUs anyways haha.

[@lifeofnatureforever2160]

Man tNice tutorials is on of the best tutorial in general on youtube. Clear, simple, constant, good to listen. Thank you for sharing.

[@jap3243]

Would love to see character concepts using "full body" on stable diffusion and ICON: Implicit Clothed Humans Obtained from Normals for a 3D character. Great work!!

[@albertbozesan]

Thank you! I will put that on my list of future videos, good ideas.

[@vbot]

@@albertbozesan I want to make a character in 3d like that and use it for vTubing hahah is it possible? Which tools or softwares good be ideal? Do u know? Please

[@zonyzhao469]

I tried as with characters, just didn’t find a way to make characters that look right.maybe someone can provides their workflow and prompt

[@nerdcodestudios2840]

generally "full body" doesn't seem to work well as a prompt. If you include words like "standing" or something like "orthographic front view" referencing blender images of course.

[@albertbozesan]

@@vbot sorry for responding so late. I don’t know how to build VTubers, but you can check out my newest video for 3D game assets to see how I turn an AI result into a 3D model :)

[@opxv]

Great tutorial! Not only the result (the icons you created), but also the pace and left out stuff that is covered by other videos. Great stuff!

[@Ghost-cf8jr]

TNice tutorials tutorial was worth it, I understood everytNice tutorialng now

[@Marbul]

Incredible tutorial. I can see this process being extremely useful for any kind of game development. Thank you for this

[@CrniWuk]

Particularly for putting creatives out of work.

[@Anon-wo2xg]

@@CrniWuk Not yet, someone that can draw will draw a better looking chest in the same amount of time. Atm it's only good for people that lack practice. I don't think it's gonna put creatives out of work but it will force creatives to adopt AI in their workflow.

[@CrniWuk]

@@Anon-wo2xg Emphasis on ... not yet. Give it some time. Who knows where it will be in 6 months.

[@LordKarronz]

@@CrniWuk Id argue that it still requires a creative to actually get the AI to give you what you actually want. have you used one yet? I use one pretty regularly and let me tell you, it takes time to get exactly what you want. Its a tool for creatives not a device that is going to get rid of them. the only ones who are going to be out of work are the boomer's who resist change every time they are faced with it

[@CrniWuk]

@@LordKarronz I've tested Midjourney and I am not disagreeing with you. I am not saying Ai image generators don't involve human creativity. The bot does nothing without human input. But that's not the point. I worked long in Graphic Design to see the "potential" of this tool. Think about where it will be in 6 months. Or in a couple of years. Particularly when the user interface becomes easier, more intuitive, where the Bot understands better what you want where changes can be done on the fly and where learning what to "say" to the bot becomes more wide spread. The fact alone that I can spend a few hours with Midjourney or DALL-E 2 to throw out thousands(!) of images which would usually take me weeks if not months, is nothing but impressive and frightening at the same time. You understimate the pure increase in efficiency here. The whole idea behind Ai (for now) is NOT(!) replace humans. That's not possible. Because the Ai is not capable of thinking and deciding like a human. But what the Ai does, is speeding up the work process exponentially. And that is what worries me. Think about how the situation will change when 2 people with the use of AI can do the same workload like 20 or even 100 that don't use Ai. How will that affect the job market? What will that mean for the value of labour? What does that mean for the creative output (production)? When I can throw out high quality work in hours that would usually take a month, how much will the price of that work drop? How are people then still going to create income? Ai has the potential to be to intellectual tasks what the steam engine and electricity was to physical labour. And 250 years ago, it caused a massive devalue of labour. And if that happens in a very short time frame - we're talking about a few years here - then it can lead to very serious social issues.

[@kims_threads]

This is awesome thank you for the tutorial! When you said you didn't like the name of the guide I was like huh yeah "voldy" isn't a very attractive name and then I clicked it and was like danng kinda brutal

[@albertbozesan]

Yeah...and they doubled down. That's why it's "he who must not be named" now - they're making fun of people who...care about hurting others. 😅

[@jeremysmith9319]

This has got to be the coolest tool I’ve seen in a very long time! I can think of so many uses I’m going to experiment with. Excellent demo of this as well!

[@albertbozesan]

Enjoy! It’s an amazing new tool.

[@veteranxt4481]

As i paid 10$ for something similar i stumbple upon some other video and found your tutorial. I wanted to kiss ya now and give huge hugs. Thanks alot.

[@albertbozesan]

You’re very welcome 😄

[@veteranxt4481]

@@albertbozesan I found out that have issue with AMD RX 6600 XT, it takes 10mis per generation

[@Pfoffie]

Surprisingly convoluted and complex haha I was hoping for something even faster/simpler. Still very amazing and I’m sure a lot people will use this tutorial for great stuff.

[@albertbozesan]

Further proof that it still takes a little patience and skill - it won’t put artists out of business, thankfully 🙏😅

[@Pfoffie]

@@albertbozesan 💯 agree

[@C4reful]

@@albertbozesan not this month at least ;)

[@hammerofjustice]

Amazing, thanks. What gpu are you using?

[@IMMANUELFRANCY]

Love you man. You are my favorite digital art Tutor/KZbinr 😍.

[@albertbozesan]

Thank you so much! 😍

[@RandomNewb]

this type of software is going to change the art industry, incredible. with some writing ability and creativity, anyone can create fantastic art. thank you for this. going to try this out when i get a free moment.

[@GonzaVerde]

Fuck every artist that worked his entire life to be good at it right?

[@pastuh]

I think you should sometimes use: shift+alt and click on mask , helps to understand which part is masked

[@albertbozesan]

Good suggestion, thank you!

[@mailleweaver]

Dang, this is a huge step towards what people think digital artists do.

[@albertbozesan]

Haha that’s a great way to see it. 😂

[@GeryNH]

Brilliant and super useful! Saved me hours! Thank you!

[@albertbozesan]

Great to hear! This vid is pretty old, be sure to check out my newer ones.

[@GeryNH]

@albertbozesan will do! But you have no idea how useful this was for me. I started using SD thanks to it and seriously sped up my freelancing project! So a big Big BIG thank you!!

[@sakuraa7]

I'm so confident, yeah, I'm unstoppable today

[@Thimas]

Astonishingly good presentation skills you've got, and what an amazing idea this is. I'm useless when it comes to modelling, so this will most certainly come in handy. Rare to see such a small channel have this level of polish already. Keep it up!

[@albertbozesan]

Thank you very much!! I hope to use these techniques for 3D models soon, keep an eye out for the video 😄

[@xzaviayifu3581]

Thank you so much for your videos. I discovered you today and have watched a few. I'd never heard of Stable Diffusion before so I was quite intrigued. After watching this video I'm champing at the bit to get started. I have so many ideas racing through my brain. Unfortunately, I have a few time constraints today but tomorrow I'm going for it! Thank you again for this, my inner muse has been unreliable for awhile now and I think this is precisely what I need to get those creative juices really flowing again.

[@albertbozesan]

Glad to hear that!! Enjoy 😄

[@Reforge3d]

Loving your process! as an earlier learner of these ai tools this is super helpful for creating concept art or even game ready sprites

[@sephyshen2510]

Good Job. It's very helpful for people who know how to code but are not familiar with drawing. Thank you. You give me inspiration.

[@kawag2780]

Having so much fun with stable diff, thanks for the tutorial!

[@Grishnakh66]

Thank you from the bottom of my heart....

[@albertbozesan]

Fantastic! What a fun project idea 😄

[@binaryalgorithm]

Just amazing on its own, but the fact that this is also free blows my mind. The only limit now is your imagination (and how good you are with your prompts).

[@albertbozesan]

Yeah it’s wild, isn’t it?!

[@mawot_the_videogame]

@@albertbozesan What about to use it for commercial videogames? Is this possible?

[@albertbozesan]

@@mawot_the_videogame Check out this Reddit thread, but the short answer is: yes. www.reddit.com/r/StableDiffusion/comments/ww8k6h/am_i_allowed_to_use_sd_art_i_produce_commercially/ Keep in mind this isn't legal advice!

[@jacobs.7925]

Outstanding!! How well is it running on AMD cards? Might have to get a Nvidia GPU just for this

[@albertbozesan]

Someone else recommended this guide: /watch?v=Lk2syPsVMQM Their explanation was “The problem is that Stable Diffusion uses a specific type of processing that Nvidia cards are "built" with. AMD cards are "built" differently, hence the need for a work around.”

[@kombosabinho]

You deserve a like man. Thanks for the straightforward tutorial and in depth explanation ❤️

[@albertbozesan]

You‘re very welcome!

[@donaldcanonoy3312]

IT'S ALWAYS THE UNDERRATED VID THAT'S LEGIT! THANK YOU!

[@TheMeldanor]

This video showed me the real potential of AI generated images. I thought until this video that it will generate a perfect picture in the first generation - but you showed me that is was wrong and even showed the process how to get there. Thank you, great video. Don't know yet how to use this knowledge, but I will remember it.

[@albertbozesan]

You’re welcome! Yeah, people who say it’s just entering some text in a box are wrong 😅 it does take a little skill

[@ivancanosa8540]

I like it! A very fast approach to get high quality assets

[@Akiruu_Sama]

I did some assets for my personnal use with Dall-E 2, but seeing that we can make it by our OWN, it's amazing. A really good video !

[@ericlewis3444]

so coooool! On each step you can add more description text as well (faceted fluted glass medieval decanter, for instance)

[@RagingGamingBear]

Enjoyed this, should turn it into a mini user submitted series :D

[@MrErick1160]

These videos are so entertaining please do more

[@albertbozesan]

You’re in luck, my new game asset video just finished uploading and KZbin is processing it :) be sure to subscribe to get notified!

[@MrErick1160]

@@albertbozesan awesome! Just subbed 😋

[@albertbozesan]

@@MrErick1160 thanks! And here it is: kzbin.info/www/bejne/paC2iGihmKZpnas

[@MrErick1160]

@@albertbozesan just watched it! I actually like more of the 20 minutes-ish video format!

[@albertbozesan]

@@MrErick1160 thanks for the feedback! Would you prefer multiple parts if a topic is too big to cover in 20 mins?

[@IMMANUELFRANCY]

I think finally we got our new legendary teacher like Andrew Kramer once again. Wish you all the best 🌟.

[@MilkBanana]

This is super useful for indies that need assets on the fly

[@joshbarros1995]

Amazing tutorial!!! Can we use Gimp instead of Photoshop?

[@albertbozesan]

Totally!

[@alenwesker9552]

I like this tutorial because it illustrates how to iterate images with SD.

[@MadRDev]

This is awesome and a great video too! I will use this in the next game jam I'm participating in!

[@mpedrozax1]

Dude!!! So I was playing with some other AI system and received crappy results. Thank You for sharing your workflow in detail. In the end you showed me possibilities that will allow me to refine the art that Im trying to make.

[@albertbozesan]

Im so glad :) thank you for letting me know!

[@jessnesband]

Best Albert, i Know! Thanks a lot, this is really awesome! ❤🥰

[@Baka_Dan]

Thanks for helping us artists and sharing your knowledge!

[@albertbozesan]

Thank you so much! 🙏

[@onlywithbuts1781]

If you are artist and are okay with this, i pitty you greatly.

[@zilibabwei]

This is amazing! Thank you so much for this demonstration! Just... wow... such exciting stuff - honestly kinda blown away!!

[@licenciadopii803]

Thanks for your easy to follow explanations, one step at a ti. It makes learning less overwhelming!

[@albertbozesan]

Glad it was helpful!

[@drawmaster77]

this is kind of insane. I've already been doing this technique but for making funny pictures of my friends and my cat. I've also been using Inpainting feature instead more then masking in photoshop (for me it was more about speed than quality). But I can see how your approach produces better results. While your 3D video still takes a fair amount of work and know-how to require a 3D artist, this 2D one just straight out seems like replacement of concept artists altogether (or at least most of them leaving just one guy to do work of entire team). I'd love to hear your take on this.

[@albertbozesan]

So far, I see this as mostly a tool for indie devs that wouldn’t have the budget to hire an artist anyways. There it’s a clear benefit that doesn’t hurt anyone. If big studios use it, I’m sure statistically fewer artists will be hired overall. But artists definitely remain a key component in this, as it needs skilled guidance to produce usable results. My images here would need a pro to repaint/paint over and retouch before use in a released game. It’s silly to believe we’re creatively all-powerful now and “don’t need puny artists”. People who think like that are showing their insecurity and total lack of artistic understanding. Morally, I’ve changed my views since this video came out. I would no longer use artist names (except maybe historic ones like Picasso, Da Vinci etc) or brand names. You can get good results other ways with more creativity. Several artists in the database have come out against it, and we should respect that.

[@Zalmoxes87]

really cool to see a detailed process, thank you for the help!

[@AlanDarkworld]

Interesting process... I once gave stability diffusion the picture of a badass fire-breathin cerberus and some according keywords, and it turned the hellbeast into puppies XD

[@albertbozesan]

Hahaha

[@kisatom]

Great project! May I ask what video card are you using? it seems you generate a pic in only 5s, I know it is accelerated but the numbers show that your GPU is extremely fast, at least compared to my PC....

[@albertbozesan]

I’m running it on an RTX 2070 Super. But I fast-forward through most of the image generation to not bore you :) it’s not really that fast. Usually around 15 seconds per image depending on settings.

[@zonyzhao469]

I’m developing an indie game and have been working with sd,trying to produce game asset. The result is always not satisfying. Your video provides good vision! Would like to have a discord server, for game devs using ai to make art.

[@kikijewell2967]

I'm here for that! Keep us posted!

[@fredericc.6115]

A really excellent, comprehensive and creative tutorial. Thank you so much for sharing! ❤️🙏

[@Theportraitdude]

This was fantastic! Thank you so much for the step by step explanation - very helpful!

[@tayyabaminhas5565]

Hey man, very clean, interesting and professional. Good sNice tutorialt.

[@clean7886]

TNice tutorials is the best tutorial. Simply explained to a newbie, great job man.

[@MrDarkmatterX]

Completely and utterly deserved the like and subscribe. Honestly, the commentary in this video was some of the best I've ever seen, I feel like I understand and I'm not the sharpest tool in the shed! Thank you so much

[@albertbozesan]

I’m so glad to hear that! I try to explain what my thought process is without skipping anything. Not always super easy 😅

[@wakematta]

This has a lot of potential. I like it.

[@vladislavkisselyov]

OH. MY. GOD. I'M FASCINATED

[@cstoomey]

Watching this tutorial makes me appreciate the skill it takes to use AI to make great art.

[@nathanfranck5822]

Looking good! Nice thing about this process is it could go right into some 2D animation software assuming a particular style

[@progoldfish7441]

awesome video if KZbin still had a star rating system id give you 5 stars!

[@LumberingTroll]

Do you have a video, or source for learning more on how you manipulate the image in photoshop? your techniques are so quick and effective, Id really like to spend more time learning that side as well as it seems crucial to getting good results.

[@albertbozesan]

It’s just a few small tricks, so I could make a short video about it. Feel free to sub if you haven’t already, so you don’t miss it :)

[@LumberingTroll]

@@albertbozesan Awesome I did, and looking forward to it, another thing I am curious about is if its possible to get alternate images with the same primary subject, say for instance I have a ship or vehicle and I want it in different settings, or situations, or perhaps a character, and I want different POVs, or expressions.

[@Progenix]

Great tutorials so far! Keep it up!

[@albertbozesan]

Thanks! There’s a new one up now about character art 😄

[@BlakeEM]

The 3 artists you used, Shaddy Safad, Ariel Fain, and Calvin Boice, are not in Stable Diffusion. They do help with this particular generation, but so does adding by any random name to the prompt, try it.😉

[@albertbozesan]

Ha! That‘s hilarious. So much for the art theft comments 😅

[@jonmichaelgalindo]

Even if they're in there, I doubt Stable Diffusion would remember those exact artist names. Seriously though, how do you know this? Clip front uses Laion 5b's index of 1.6TB of KNN embeddings, so there's no guarantee a search will match your string. (It's just an AI that'll probably match more frequently repeated strings instead of exact terms.) The Laion metadata is 800GB. Did you download that? I've been considering buying an external HDD just to try... But I'm not sure it's worth it. Running an actual search for a single term on that would take over a day on my PC since it has to keep copying into RAM... but solid state would be way too expensive. (I was wondering something similar, how to check if a particular thing was in Laion / SD.) (But it's true that adding any string of characters to a SD prompt affects its output. You can add emojis. 🙂)

[@BlakeEM]

​@@jonmichaelgalindo There is no way to get the training data because it contains paid for stock images as well as many images that came out of a dataset created by the financier of Stable Diffusion and isn’t publicly available. There is the laion-aesthetic-6pls list that pulls from the same sites, but it isn’t 100% complete. This doesn’t contain any of these artists. Stable Diffusion doesn’t forget anything, the artist names were entered in conjunction with the image and is how the prompt references those images. You can test for yourself to see that it takes 3-4 images to get color and style and only a single image to get color influence after about 4 hours or less of training. If there was a single image in the system I wouldn’t be getting amateur pictures of flowers and flat color and often abstract faces. I get these same types of images when entering any name that isn’t in the system. However, just by saying something is by it improves because the word “by” implies that it’s by an established artist so the quality increases a little. It is likely that some art that they made is actually in the system but it’s referenced under the game name and is mixed with other assets.

[@jonmichaelgalindo]

@@BlakeEM "Stable Diffusion doesn't forget anything" LOL what? SD's total size is 4.3GB. LAION 5B's _publicly available_ dataset that SD was trained on (all 340TB downloadable) has 800GB worth of meta data. How, exactly, could SD "remember" 800GB worth of text data in just its 4.3GB of parameter data? And that's _just_ the text. Even if SD used 100% of its memory for just text, it would still need a compression ratio of almost 200:1 in order to memorize all the text meta data it was trained on. That is mathematically impossible. (It's called the pigeon-hole principle by people much smarter than me.) Plus, I guarantee most of SD's 4.3GB of memory is dedicated to visual concepts, not text.

[@BlakeEM]

​@@jonmichaelgalindo Yes, Stable Diffusion doesn't forget it's training on all those images. 4.3GB is the compressed version, the full version is 7.7 GB but the results are nearly identical, with the larger file having slightly better coherence in my testing. I think you misunderstand how it works. There is two parts that are trained separately. The autoencoder (or classifier) that understands putting words to an image, and the diffuser that leans how to de-noise an image. It's often called guided diffusion because the diffusion model is guided by the autoencoder. Learning colors and aesthetics of an image in relation to other learned information takes up far less space than learning the exact details of every image. The more it learns the less space new information takes because it can defer to other learned aesthetics and simply adjust them instead of relearning every idea from scratch. This is how it can blend ideas together so well. Please correct me if I’m wrong. Although I’ve been a software developer for 25 years I’ve only barely scratched the surface of the Stable Diffusion codebase.

[@MorneBooysen]

This is bloody amazing

[@jaycristobel5044]

Hey man, Great video and AI system to create art in a few steps is crazy!!!!!!!!!!!! Bless your skills and wanting to share this skill with the rest of us artists. Thank you very much. Subscribing to you now. LOL Have a great day!

[@albertbozesan]

Thank you! You too 😄

[@neopabo]

This is so insane, I need to mess with this

[@kikijewell2967]

This is exactly how I see AI: not as a threat, but as a _partner_ in building our future together. When I told an artist friend of mine that people were afraid AI would ruin art, she replied, "oh no! Art is dead AGAIN!" lol

[@albertbozesan]

Ha! That's a good line from your friend there :)

[@NukaColaLight]

Art isn't dead, artists making a living from art are dead thought. Game over for them, so wonderful.

[@kikijewell2967]

@@NukaColaLight just like it did before, AGAIN. I promise you: it will create MORE art jobs, not less.

[@NukaColaLight]

@@kikijewell2967 I hope you are right but as a professional artist working on a big video game title, my opinion for now is that it will leave us with less and less working artists, as in people that can make a living out of it, as more and more of the production is automated and teams get reduced to a few core members making the artistic decisions and outsourcing a bulk of the work to machines. Again I hope I'm wrong. My opinion is that it is not the end of art, but it will be most likely a drastic reduction in working opportunities for artists, and so the "death" of many art career.

[@randoguy7488]

@@kikijewell2967 I don't see any new jobs for artists though. Character concept artist - AI, enviroment artist - AI, prop artist - AI Also, I don't think such thing has ever happened. You can compare it to "Photography will kill art" but the adoption and evolution of cameras was slow. This is lightspeed fast.

[@ThatKid22101]

The only good use for A.I, concepting ideas to be improved by hand.

[@warturkey_yt]

Awesome video! Any reason you use a black background instead of white?

[@albertbozesan]

Nope, works just the same. Sometimes a saturated background can help too, makes it easier to cut out later.

[@warturkey_yt]

@@albertbozesan Thanks bud

[@halladall1]

Thank you so much for this. I am still learning but this really helped me out a lot.

[@thibaudherbert3144]

Like it ! Very curious about how much complexe scene you can create with this process, with much more time of course

[@albertbozesan]

I think your scene can be as complex as you want! Check out my other video where I make some Sci Fi concept art: kzbin.info/www/bejne/jqWwp55mmdWhhZY I'm planning to try photorealistic matte paintings soon. That will be exciting.

[@KivySchool]

This was pretty impressive, thanks for posting it.

[@baelgul9099]

Love this content, keep up the awesome work!

[@arpit.soni2001]

TNice tutorials was very helpful thankyou.

[@alekskroskin6101]

Wow, actually mindblowing

[@vanderkarl3927]

It's hilarious how many people think that AI art is a fad. Even as rudimentary (compared to what is to come) as it is, it has become yet another tool in the kits of many artists and, really, anyone who wants quick, free assets with no IP strings attached.

[@benjaminschultz6501]

Very novel and informative! Thank you for sharing your creative process with us! 🙏👍

[@tomashais9123]

ITS REALLY WORKED LOL THANK YOU DUDE

[@_SkyDancer]

Jeez this is unbelievable!!!

[@albertbozesan]

Thanks! It’s also super old now, the developments in the meantime are mind-blowing.

[@MitchWilcoxen]

Wow, great video! Keep up the good work

[@leestorm]

Nice, thats for the tut and download links. very informative and SUPER helpful

[@Foxcoder]

great, thx for the explain, it will definitely boost my game development

[@minhtrongngo8130]

TNice tutorials was really helpful for soone coming from Ableton Live and having used Fruity Loops waaaay back in the past

[@Sharlenwar]

This is a very cool way to use AI.

[@christiannilsen2835]

Wow! Mind blown 😮

[@aaronspencermusic]

Hey, Thanks for the video, I'm in the process of installing the rather shady looking files from the terribly named thing. Is this the version you're using with the Web gui?

[@albertbozesan]

Yes, I used that guide to install what I show here! Most of the install files are straight from the official GitHubs, so it should be safe.