CS:GO - How AI-images could Affect/Benefit the game's community
Вставка
- Опубліковано 4 чер 2024
- In this video I show lots and lots and lots of AI-generated images, and discuss how they could benefit the players, mappers, skin makers... and me
The Stable Diffusion GUI tool I was using: stability.ai/blog/stable-diff...
0:00 - AI and Stable Diffusion's magic
1:20 - Generating pictures of CS:GO
3:36 - As a player
7:39 - Mappers
10:30 - Skin making and conclusion - Ігри
Boris Johnson in Mad Max is the epitome of realistic image generation. His generated face is indistinguishable from reality. 10/10. I almost suspect this wasn't AI generated but simply photoshopped by hand.
I am currently watching a big booba renaissance on twitter, and it is insane how good some of the images are from stable diffusion.
Some of the content looks like it could have been in some castle by Lord Pervy.
Boris Johnson in anything has been my favorite since I found these tools. His face just suits everything.
@@loonloon9365 That's disgusting! How'd you find em tho...?
The one of him in the suit is gold
Adding a sketch of an image as input helps Stable Diffusion generate really nice results. I wonder if it could fill a UV Map to create unique awesome skins
That would be really cool. I would love if I could do a sketch on a drawing tablet and then have it refine and add similar style (like the decorative patterns they do on fancy hand engraved revolvers and knives or antique furtiture), do some of the flowing lines and it copies the artistic style for a bounded object shape, then laser engrave the pattern onto objects.
There is a Blender add-on that can do this now, it's called Dream Textures.
The last line of the video being done by the voice AI was a nice touch.
Oh my god. I didn't notice.......... Wow
Is only a matter of time before someone makes him sing "Country Roads","Big Iron","Hino Nacional Brazil","Never Gonna Give You Up" and even "All I Want For Christmas"
I was looking for this. Also when AI voice is fully implemented, you will see a uptick in vtubers...
I'm expecting 4kliksphilip to be 100% AI generated kliksphilip-content
I was actually thinking the other day how useful AI generated art could be for making maps for games and stuff. I noticed in RE7 there's alot of repeated paintings and photos and I thought about how using AI to quickly generate a bunch of classical looking paintings could fix this. They don't have to be super well done either, just enough that someone could look at them and think "yeah, this is a painting". Making posters, photos, newspapers, random product packaging could also benefit from it.
Oh yeah this is a great idea, I keep seeing like 10 paintings in games repeated everywhere because if the paintings don't matter there's never too much detail in them. You could even try to match the style of environment in prompts to have creepily atmosphere matching paintings. That is honestly a brilliant idea and I shall be forever mad if game devs don't get the same idea.
You could even have a human edit the images to make them more believable afterwards and have a quicker workflow when it comes to such details.
@@zarrg5611 In general, asking a human to edit an AI image to look more believable is like handing a translator a piece of text that's been run through google translate 45 times and having them "just edit it so it makes sense." The work required to do that offsets any benefit the automated part of the process might provide. Pardon the hot take but I really don't think that we should be trying to automate art. Like at all. I think that defeats the purpose of art.
There are some forks of stable diffusion that allow you to use "negative prompts" alongside your normal one. for example if you put "deformed body" in the negative prompts it will try to not generate what you wrote and steer in the opposite direction. this allows you to create 99% real looking images with no hints that theyve been generated by AI. you really should try it. i use AUTOMATIC1111s web based UI, it has so many good features.
also if anyone is interested (and willing) you should check out the /sdg/ stable diffusion threads on 4chan's technology board (/g/), since the newest progress made by the open-source community tend to be posted there.
I don't play CS:GO often but I really like your Videos.
Almost every time u have an interesting topic to talk about and your voice/the way u speak is really relaxing.
Nice work again, Mr 3kliks
Philip: please draw this portion of a CSGO MAP
AI: B A N A N A
I've actually been using the Midjourney AI in my game in the way you describe at 9:30 - I have a vague idea for a level after which I use the AI to inspire me, sort of like concept art.
In addition, I put the best images in the game as collectibles for the player (it's a 2D adventure game about exploring dreams).
There's a video about the process on my channel in case anyone is interested (which is definitely inspired by your amazing videos)
That was a cool concept, enjoyed your explanation in your vid and I think you have a great concept using dreams within the game for different levels.
When AI gets to doing 3d modeling, I think that's when alien world items and sci-fi blasters will be really cool to throw into first person games (2d like you did was pretty cool, but I see 3d becoming a thing within the next decade too).
5:40 This truly is a Liklк Liilkpe 3lЮ moment
it's always funny when he says '
The AI generated textures for mapping just blew my mind idk why
Somehow, through your spreadsheet organization and impeccable comedic timing,
you managed to make me watch this video at the exact seconds it would take for my boss to walk in on me looking at the handsome black men section of the film.
Thanks Phil :)
5 days ago I installed stable diffusion on my PC. I need sleep, but I don't want to sleep. It's addicting beyond anything... I watched this video while rendering a batch of sci-fi concept arts.
One nice thing about sleeping is that if you've found a nice prompt you can leave it running overnight and you can wake up to a folder full of the results.
@@jan_h Another nice thing about sleeping, my brain can render prompts live 3 dimensionally based on prompts. I think its called dreaming or something.
Hey there. Sorry if this is stupid, but I can't figure out how to install stable diffusion. I just see a way to use it in a browser. Can you help me out?
@@Blodhevn666 I just followed this simple guide:
ua-cam.com/video/Pyze0seDHzA/v-deo.html
I can't believe you haven't used AUTOMATIC1111 webui for Stable Diffusion when generating textures, especially for an AI upscaling/generating fanatic as yourself! It has support for generating tiling images in it's ui already, and can easily upscale and fix deformed faces using ESRGAN models, GFPGAN and CodeFormer. Additionally, there exists a tool for AI generating normal maps just from a texture called Material Map Generator, which should prove useful too.
Most of these UA-camrs clearly don't understand how to utilize the technology lmao
Too complicated an installation process.
There is a once click installer that literally installs all of it by just double clicking s single bat file.
I can feel slowly AI are entering our world ..
They probably already have we just don't know it
You're behind the curve. AI drive internet traffic and trade stocks to maximize profits for almost a decade already!
@@chyza2012 it is
@@chyza2012 do you even know what AI means
0:19 the propositions for the generated image are something else
Between this and your video on Hammer++, I can feel the mapping itch coming back, and I can't wait to spend weeks working on a project never to be touched again but purely for the enjoyment of it.
You killed it with the last few videos. Missed them. Great job as always.
Hi Philip, your music kit brought me lots of joy!
The texture generation would've been fantastic for GoldSrc mapping back in the day. The example you made up looked like a really funky old TFC or 1.6 style Nipper map, which is an art style I do sorely miss.
The “does it even matter” JUST BARELY activated my my brain, as I’m so familiar with your voice, but for people with less of that built-in voicebank, I could see it go unnoticed!
4:27 That dude actually looks kinda normal in the face but then you look down and see his shoulder is growing another bicep on his right pectoral....
The reason you were unable to generate CSGO radar images is probably due the "aesthetics score" filtering used when training stable-diffusion.
From stable-diffusion v1.2 onwards they filtered out any images with a score less than 5.0, which I noticed removed the majority of 2d images.
(Especially anything flat and monochrome)
I was actually able to get a few, but the prompt has to suggest for it to generate an overview. the one that worked for me was “competitive multiplayer map layout overview” but results may vary
Snowy Graveyard in a sunset is quite possibly my favorite map setting too. In fact, any map based on dusk or dawn settings are always my favorites.
Moreover, Philip has a point though, this A.I. image generator can help mappers to picture their ideas or textures and so forth.
I'm so glad you pointed us to this program! The GUI version is super easy to use and takes about 10 seconds to generate an image (3080 12GB). I will for sure find inspiration from this program
You were ahead of the curve with the Radar Map generator thought, you began to speak of it just as I was feeling a burning urge to comment about it. Although Stable Diffusion might not handle it out of the box, there are some AIs that have been able to work from a small corpus of source images you specify. If you could normalize an unannotated set of radar maps, Wave Function Collapse could even generate level layouts and that's a tool that we have had for years.
I just started mapping and my god this couldn’t have come at a better time. Texturing is the one thing I’m scared of in my future. Definitely going to try using this when I’m ready to start textures
I doubt ppl are interested but my idea is a pirate haven city type thing with inspiration from my favorite areas in lots of games. Like b on breeze in val or the statue site on cobble with some inspiration from old toxic on nuke
@@3kliksphilip thank you! I know it probably won’t come out perfect the first several times 😭but I’m so excited to have found a new way to enjoy this franchise that’s been part of my life for over a decade now, and largely in part to how passionate you are about maps and such🥰🥰
Philip, you are highly imaginative on how to use AI. Awesome work 👏
I think This is actually great for creators to get new ideas or use as textures.
We're in for a treat!
The title feels like the title of a school presentation love it
Interesting idea to get inspiration for skin-making! The AI may get a little too inspired by existing skins currently, but it's definitely something I'll try soon
Perfect timing! I used AI-generated images to spruce up my map just yesterday. It's a Left 4 Dead campaign with a chapter set in a haunted house, and I used Craiyon to generate a few spooky paintings to decorate the walls. I'm too out of practice to make good paintings anymore, and I don't have the money to commission an artist for such a minor detail in a hobby project that will be released for free, so AI art is a perfect solution.
I'm fully sympathetic to the freelance artists who are worried AI art will hurt their job security, but at the same time, any decent employer should know you cannot substitute human artists in the vast majority of projects. Still, I think it's a tremendous tool for concepts and brainstorming (as you suggest in the video) and I imagine in a decade or two, using AI to brainstorm will be as normal as digital painting -- another new art technique that caused just as much controversy back in the day.
Which map name? I haven't played L4D2 in a while, so if it ends up getting published on steam I'd love to play it.
@@jakegarrett8109 It's not done fam. It won't be for another year at least lol
@@QuintessentialWalrus Ok, didn't know how close it was. Have fun making the map!
28k only after so many hours up? That sucks. I really love your videos on developing technologies, Phil; not only do you often show off the tech in both novel and average usecases, you also give us your interesting perspective on the potential future such advancements hold.
Thank you for all your hard work! I don't even play cs or quite frankly, enjoy it outside of the context of esports, but I always enjoy listening to you share your passion for it, and for many other subjects. :)
I think the title is just too specific, his other videos with vaguer titles like “CSGO - Can Water?” still get good views
The only reason I never watched the video until now is because I’m scared of AI generated images. Now that I’m actually watching it, it’s a really great, interesting nightmare.
There was recently an addon made for Blender that also does AI textures. I think this is absolutely the future of mapping. Even Substance/Allegorithmic is on it with Alchemist. Very exciting
Definitely look into textual inversion and Dreambooth for stable diffusion.
That AI generated sand looked just like one of Half-Life's original textures!
Most impressive also I didn't know I needed Boris Johnson in mad max ty!
I have already made a skin with ai textures before, it looks great and i am very excited to see what else ai can do for humanity in the future!
just got the music kit and its awesome
Eyy I got a notification for this one. Thanks for the fix 👍
I love this channel so much.
I'm finally subscribing to your channels. I don't know why I didn't already years ago, I love your videos. It's weird, but I'm finally subscribing. Should've done so in 2015 or something.
There is a stable diffusion addon for Blender that is intended to be used for textures and that already produces tiled images.
“ai works well at the rough, high level, and at the very detailed low level, but it’s the space in between that benefits from human intuition” sums up my experience using AI programming tools like Copilot and TabNine. They save a lot of time that would be spent looking up common algorithms on stackoverflow but they’re definitely not a replacement for a human being.
This video is peak art.
Think you would like Dreambooth-Stable-Diffusion where you can add your own photos for ground truth thus your own model based on input photos (ie, add photos of yourself and thus any output will be based on you)
why am i not getting any notification on this video? nor even this video popped up on my front page. had it not for the short, i would have missed your upload. and i guess this isn't unique to me judging by the view count, idk. hope you'll bounce back viewcount wise.
can't wait to see the 2022 empire analysis video :D
When the time comes, I will be waiting for a fully AI generated kliksphilip video.
adding the ai generated 3kliks voice at the end was a nice touch... but i did notice, so yes... it does matter
The thumbnail has some seriously sinister energy
Great video!
Here's some tangential thoughts on how the game industry will change from AI over the coming years (from the perspective of someone who spends way too much time exploring AI/generative models) - not specific to CS, but your section of the video on game assets reminded me:
Big overall shift towards pre-production period asset-creation, a move away from production-period asset creation; in a way we've already kind of seeing this, asset stores like unreal/unity have been blowing up over the past half-decade with custom created content for any model/asset a dev might need - this'll shift permanently when assets can be generated on the fly real-time or during production, and when assets aren't limited to just whatever the artist can think of in a vaccuum. Instead of artists making assets per-game, it'll be large teams of artists creating assets per-dataset, and datasets aimed at highly specific styles or general versatile sets of assets.
This'll happen not just with 3D and 2D artists, but with sound production too (like foley creation, dialogue, music scores etc). AI progress in sound is admittedly not as far as image synthesis but it's pretty close behind, I'd be surprised if we don't have a DALL-E 2 / Stable Diff quality-equivalent for audio this time next year.
Programming/coding for gameplay mechanics is already possible with AI, and given how UE4/5 has been moving over to blueprints instead of code to enable level designers scripting freedom, it's clear that this is direction the industry has been headed already - enabling designers more creative autonomy, getting them more hands-on with the games systems rather than what is currently the development equivalent of backseat steering a gigantic freight ship.
Overall though I think it's incredibly exciting, this shift will enable ANY game dev/designer to equip the full capabilities of gigantic teams of artists creating assets on their project, but for literally any project by anyone in any direction. It'll be a similar shift to the way audio/music has become so accessible that anyone now can start learning with just a laptop and a DAW, but potentially way more creatively liberating than the DAW was in sheer scope and capacity. Possibly in 5-10 years, Destruction Darius 3 could be made on the scale of any current day AAA game in the same amount of effort it took you to create the original. Since the limitations of scope and complexity will be sort of 'solved', we'll see far more dynamically generated/AI pseudo-simulated content in games (or much richer content in linear style games).
Last point for those that have lost or are worried about losing work from AI: automation comes for all jobs, and while I sympathise with you and can relate to the financial stress, the solution to the issue isn't to grab our pitchforks and stagnate automation to preserve jobs (that'll at most buy you 6 months of time before the models will be retrained on different datasets and be back at square 1). Instead, we should all consider how society/governments can support people of all career paths when automation inevitably lays them off, and restructure the world so that people aren't financially crippled when they are put out of work for whatever reason. If people's livelihoods werent so entangled with their job status, having *the most creatively liberating technological advancement in the world be released open source and accessible to all* would not be such a controversial issue, but instead profoundly invigorating for all artist/creative-types that haven't had the time or physical ability to create before.
I’m sure someone can LiDAR scan and photograph a whole library of stuff to create a texture generation engine. Could be pretty lucrative.
I hope you keep at it, better prompts, image inputs.
Plenty of tools have been available since. Making such a model openly accessible democrasizes creativity. And it's been 4.5 months since this video.
I went to school to be an architect. Graduated in 2013. We sometimes made abstract images in photoshop or by hand to inspire designs and design concepts. This AI shit would absolutely turn what we did back then on its head.
Great video. Using AI to make placeholder textures is a good idea, specifically speaking as a programmer (haven't used hammer in a decade but plenty graphics programming) it'd be helpful :P
I like how some of the AI generated images looked like they had what was supposed to be a parts of a Gamebanana thumbnail.
There are versions of stable diffusion with texture habilities and with Dreambooth you can train the model with whatever you want, even counterstrike and bananas
Sd-webui and automat111c include image to image generation to help the AI to generation versions of a image
And tiling support
It would be cool if there were sliders or tools you can use to make the AI change the current image to your desires.
Turns out that is a thing! If it sounds interesting I would recommend installing stable diffusion for yourself and playing around with the img2img functions. (If your GPU doesn't have enough VRAM, use the optimized stable diffusion branch, which is slower, but at least it works on more PCs).
youtube didn't notify me this video
Seeing you make those textures I figure I would tell you that someone has already made a plugin for blender that takes a prompt and auto generates a tillable texture out of. fairly mixed results but I would assume it will become a valuable tool in the coming future.
Was this reuploaded? Damn now I have to watch it again and spot the difference.
In response to 8:10 there have already been tools made to use inpainting on stable diffusion to create seamless textures, for example "Dream Textures" for Blender. It tends to look much better than GIMP's seamless plugin
The Corridor Crew channel had a video (Title: Is This The Death of VFX?) about training AI with a new Person... the results are disturbingly accurate!
the ai klik at the end was scary
That ending of "If you can't tell, does it even matter" from Westworld where robots revolt against humans and make us extinct was a scary but a perfect ending to such a video 😁
This makes me really tempted to make a VRChat world based off AI images using AI generated textures. One useful thing is that Unity has a way to make normal maps from textures and it's usually pretty ok.
There is an AI-generated mosaic of a shrimp that I used with env_projectedtexture for my shrimp themed church in fof_canyum for Fistful of Frags. Should be going live in the game sometime in the coming weeks.
AI can only draw topics that it has been trained on. Thus, it does very well on prompts that have a large and well-ordered dataset, like digital art from boorus or livraries of stock photos. Unfortunately, there aren't a lot of good places to find labelled screenshots of video games, so its "knowledge" on the topic is limited. More recent, popular games like Fortnite have more images to look at, so it does better there.
One solution would be to gather large amounts of game screenshots and videos and feed it into the training model.
INCREDIBLE HOW OLD THIS VIDEO LOOKS LIKE JUST 3 MONTHS AFTER!
Stable diffusion is the beginning of an era of open for all AI generated art
There's a Stable Diffusion plugin for Blender now, that can automatically generate tiling textures. the examples i've seen barely give you that visible grid that most 'normal' tiling settings do.
Hoooo boyyy
Those chromedsets images activated my neurons so much that I almost had a seizure.
You can actually import a model made by community, which can be used to create images based on other databases, which can result for example in better generation of things from particular field. For example art by feeding AI mostly on Art images. Same could be applied to CS:GO and the results would be MUCH better if trained properly.
I wonder if AI could be more useful in map creation if it was specifically taught about it using something more basic and abstract, like just 3d floor map files in some simplified format, as inputs and outputs.
Neural networks are quite good at finding counter-intuitive patterns, so such AI could provide numerous propositions for spaces that encourage engaging encounters or even completely new, unconventional yet working map layouts that would be hard to come up with otherwise.
img2img could majorly improve all the ideas you suggested; want a radar overview of a map? want a variation of a place in your map? img2img is the answer (or, for even better results, you can also try making your own "style" using dreambooth on some rented server or something)
One thing that occurred to me during the mapping section. I wonder if our ideas will benefit from AI, or if it will simply remove the need to fantasize about things?
MidJourney is another fantastic AI for generating images (which is also discord based!) had much better luck generating images with it compared to any other AI, its crazy. Especially when you research how to make the prompts, can genuinely get photorealistic portraits etc.
4:03 Philip went full TotalBiscuit , kinda sounded like him
6:26 wait why do I feel like I have seen that face before...
EDIT: it reminded me of a guy that was in a few of Abroad in Japan's videos. His name was Alex Shapiro, and he was that real estate dude
What if one day in the future you could plug the maps themselves into training for AI to generate new layouts? I've seen something similar that generates living room layouts based on specific kinds of furniture for example... Imagine starting a match and then just typing a prompt like "beachside mansion" and having a new map to play! The future holds so many possibilities
this video never hit my sub box. very sad cuz its a great video
I've been working on a map in Intruder that's only uses generated textures. It's close to being done.
How have I never seen the seb and you comparison before
Somehow Image storing AI for data compression sounds like the basis for a cosmic horror novel thinking of how the images might degrade over thousands of reposts
here after reupload
1:40 I think also part of the issue is that just trying to generate off of "counter strike" will cause it to blend cs 1.6, css, and csgo's artstyles.
technology is great, I am curious what we will see next year. Will we have Ai generating UA-cam vides from prompts?
So were you using the voice robot of you during this video? Would be really meta if you did lol
Funny enough, I used AI generation for a jungle TF2 map a few months back. The images were actually insanely good, if I ever get around to it I'm going to use a couple of them cause what they generated literally would work amazingly.
please do more ai generated stuff i know theres a lot out there. but theres not a lot that do it well. and you brought it to another level been subbed for a long time and this was a breath of fresh air from a channel i already thought was perfect.
What if you made a map using AI generated textures?
If only he could make a video about using AI generated images for CSGO in general too
It seems I have commented on a video before I finished watching it
0:27 I see what you did there :)
i love your ai image content
While I personally dislike AI art, seeing the uses of this and some of its more unique applications is very interesting to me! Cool video
Great concept. We're still at the horse and buggy stage of ai image generation, it's only going to get weirder.
It's still in the dreamland level of realism, but that's good as a year ago it was completely abstract, even with millions of data images to create more images with
In the end ai will be feeding itself. Reusing the generated pictures from other ai programs until nothing human remains
I want to know how long it took you to generate all these images. Also how much space on your hard drive consists of ai generated photos
you should look into CLIP, textual inversion and of course dreambooth
Ngl Boris Johnson looks really appropriate in May max
so basically, AI image creators are a version of google that instead of getting confused and showing you other, completely unrelated things when you enter a prompt a bit too long; it shows you whatever comes to mind and retries with your request until both of your thinkings come to a proper conclusion.
If anything, I'm most hyped for what this tech can do for search engines as mentioned, cause god damn do I hate treating google like a baby, fool-proofing all of my searches and making damn sure it doesn't accidentally get stuck on a certain popular buzzword inside it just to get an answer to something that's very slightly rare...