How I used ChatGPT and Stable Diffusion to make a whole video game! - AI for Godot Game Dev
Vložit
- čas přidán 14. 12. 2022
- Play the game here! bozezone.itch.io/submerged-se...
Get 10.000 bonus words when you sign up for Sudowrite here: www.sudowrite.com/?via=albert (Affiliate Link - this supports me!)
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...
----------------------------------------------
Did you like this vid? Like & Subscribe to this Channel!
Follow me on Twitter: / albertbozesan - Hry
Everyone is freaking out that AI will make all of our jobs obsolete, but in reality I feel this is just an chat partner that people who already know what they do can use to structure their thoughts better. Like if you talk to a colleague about something you are working on... But Gamedev is a lonely journey and we don't have friends :D
in reality AI is progressing ultra quick. its nothing to freak out about eitherway. Nothing suggests freaking out as a sensical reaction.
I mean, it is automation to some extent (especially assuming AI will get better, which is a pretty reasonable assumption) and so brings the job replacement any industry automation brings, for better and for worse. But agreed that freaking out isn't a sensible reaction. Automation always brings the doomsday prophecies, but society has yet to collapse from it.
Don't kid yourself. 10 or 20 years from now not just millions of jobs will be obsolete, but by extension and implication millions of people are going to be obsolete. You don't have to have a crystal ball to see there's a very dystopian future ahead with this type of technology.
Nothing which will stop progressing. So get over it ppl ;)
literally this
This is so fucking cool. At this rate we could have so much more solo devs (or just solo creators in general) taking on more and more ambitious projects while maintaining full control with their creative vision. This also gives much more emphasis to knowing the theory of *how* exactly things end up good (composition, storytelling) than to the skill of being able to execute them, since the AI helps with a lot of the heavy lifting.
AI is like a really unreliable coworker with occasional bouts of genius right now. That's already worth a couple thousand a month, but we still need to know EXACTLY what we want. It can't help with project structure at all, or figuring out which of its suggestions are complete nonsense in context 😅
Tons and tons of knowledge in one video! Great content mate! Please keep more videos coming...
Thanks! I have one in the edit right now 😄
Really nice experiment, keep them coming xD
That's amazing. I'll have to give it a try with Unity.
Update: It works surprisingly well in Unity. I asked it, for example, to write code that would bank on object based up the rate of turn and it came back with usable code. It even commented on the necessary components of that game object (e.g. RigidBody). Impressive!!!
Very cool! I can imagine it knows way more about Unity.
Tried out with Unreal Engine and it can handle shader writing (as well for unity), blueprint code and c++ code generation even for most difficult thing and can give you an idea how to handle certain topic.
@@LukiGames0 Very cool!!
What I love about ChatGPT the most is that, with programming as an example, you can treat it as "a shortcut for quickly finding your solution" without having to dig through Stackoverflow or random google pages. I solved many TypeScript issues with complex types and some random code logic, SASS, Unity scripts and even Laravel!
Obviously, it won't give you a ready-to-use solution right away, some code errors might appear, but it's still great, because it will point you in the right direction. It's like having a good buddy sitting next to you, saying: "here, try this code, __maybe__ it will work, I haven't tested it yet, play around with it". At some point, something will just "click" and you will figure things out even faster.
It only takes a properly asked question with some good explanation and extra sample code. This is a huge workflow booster - a totally new experience...
I just fear how lazy this will eventually make me when it hits workplaces. IDE's already made me to ba barely able to type, I program stuff all the time but I cna barely type on the keyboard compared to how I started xD. These suggestions and such are so handy I hit tab or enter more than most of the keys on the keyboard throughout the day.. lol.
When ChatGPT does mistalkes, tell it (I'm referring to us all) where are its errors. Using the payed version GPT-3 but I think ChatGPT is the same, I saw an enhancement to the 68000 assembly code (old Commodore Amiga processor) produced, even at different my logins to OpenAI, after I told him that he mixed wrongly Amiga and C64 assembly code! Probably he can evaluate our suggestions when he has enough internal data and learn/enhance his answers.
I was looking into this, too. I'm going to have to really play around with this!
That’s freaking COOL! Holy moly!
I think it's gonna be a companion for people that know how to use prompts. I had an idea for a 2D survival game that was similar to Tarkov. GPT came up with a title and plot. PlaygroudAI made the sprites for a separate project I'm working on.
Love it man, thanks for showing this gems tools!
You finally finished a project! Each new video is better than the last!
Thank you! Yeah, this one took a while. I planned a couple hours, and it turned into around two days...
Awesome video man! The only thing I think is missing is some simple AI generated background music, but that’s just a suggestion :)
Yeah for sure :)
Nice! Good Video! Love that!
Glad you enjoyed it! Thanks.
I'm actually currently trying to make an entire game with only AI generated art. I'm a programmer by trade and I have plenty of experience with Godot, so coding is not my problem, but art is always somewhat of an issue for me. I can do *some* art, mostly pixel art and UI elements, but I was curious if I can get good results using AI...
The short answer is yes, kinda...
the long answer is... well... The results are ok, in some cases amazing but it's finicky. I sometimes find myself generating, editing and re-generating images for literal hours. I mean good 2-3 hours for something, that an artist and a ui designer would give me in much higher quality and in .psd or .kra files with layers in about 30 minutes... On the other hand, especially with backgrounds, you can get amazing results for roughly the same amount of time as it would take an artist, or less... the issue really becomes style drift then. The more pics you generate, the less artistically similar they will eventually be. You'll start with one idea, and while looking at hundreds of generated examples, your old ideas become outdated and you start picking backgrounds by new and more interesting aesthetic qualities... which means you end up with a mess of backgrounds, or you can re-do it all over now that you have a "better?" sense of direction... All things that you wouldn't deal with at all if you had an artist... not to mention an artist works in parallel with you, but you yourself can only work sequentially...
These AI tools are fantastic, especially for rapid prototyping and sparking imagination. But to really get a grip on the end result, you'll end up spending actual time and effort on using it. For a guy like me that likes working solo, especially when it only comes to hobby projects, it's an investment I can easily make. But for any serious projects, you're going to waste ton of time using AI yourself. Team up with an artist, someone who knows what they are actually doing, and you're going to produce a much more coherent and better product.
Anyway, my game idea is not really art-centric. The main game design is mostly experimental stuff that I'd like to see made. It's for an android game too, so I'll see how well that goes xD
AI doesn't steal jobs, it's the people who uses it cleverly that are gonna do it.
So the best thing you can do is just embrace it and take advantage of it
we especially don't have to worry about it stealing programming jobs because the overlap between people who would work programming and people who are excited to play with this new toy is a lot lol
also it can't really steal the job of a good programmer bc it makes so many mistakes that it takes someone who knows what it's trying to do to read the code and make sure it's good--some rando who's never coded before couldn't be like "program me an app!" and get something that works off the bat
Damn, this is so inspiring!
I've taken this idea and started creating a top down RPG with the intent of creating a pokemon-like game.
So far, I've spent 2 weeks without ChatGPT failing to get anything done, and 4 hours with Chat GPT creating a player character node which switches between keyboard or controller inputs and checks to see where it can navigate based on whether the adjacent tile is occupied, too low or too high.
It straight up told me when I was trying to use the wrong node for this too.
It started with a KinematicBody2D node to start with and when I said I wanted it to go up and down ramps, it changed to an InstantMesh3D with a 2D sprite using Billboarding to meet my purposes, and even explained its reasoning and commented its code for me to check it.
Awesome! I’ve come back to this and tried to make an update. GPT-4 is much better at GDscript, so it’s been way easier.
@@albertbozesan It's definitely still difficult if you don't know what you're doing (me), but as long as I keep checking that it's doing what it's meant to by asking "does this conform to the constraints I mentioned earlier?" and "Will this code function according to the set demands stated above?" It'll generally continue fixing the code.
In addition, I started asking "Are there ways to optimise this code to run faster in Godot 4.x?" and asking it to implement some of the solutions.
@@cptncutleg yeah, same! I'm far from a good coder. Fair warning, ChatGPT, even with 4, is not adept at Godot 4. It will make a lot of mistakes that you have to translate into the new syntax.
@@cptncutlegif u want 2h-30 minutes create games and apps I'll introducing AI CHAT DEEPAI the AI CHAT DEEPAI is so powerful and much better than OPENAI since start 2015-2018 and u can show ur picture and video and anything u want
This was so so so cool, thank you for doing this experiment!! This is incredibly inspiring !!!
I'm glad you thought so :D it's super promising for sure!!
Very inspiring!
That said it was a pretty cool little game you made. If it wasn't explicitly stated that it was produced with AI's help I probably wouldn't have noticed it.
Thanks for pointing out openjourney, it's amazing!
It's my favorite right now!
Nice! Was just about to get to this too :)
Let me know what you learn!
Wow I was surprised that ChatGPT knew the Godot language, but as you noticed, it needs to improve. But anyway something can help, it is likely that later it will be more effective. For creating shaders, it would be nice
You are genius!
Haha these tools are the real genius!
Well done video with great graphics as always.
Why didn't u use ChatGPT to create the story?
Because Sudowrite is specifically for fiction writing and has a lot of extra features that help. It also uses GPT-3. If you’re curious about it, check it out: www.sudowrite.com/?via=albert (affiliate link)
@@albertbozesan Thank you for the link, I bookmarked it for future use.
now, I can become a game creator
thanks for tutorial!!
I recommend trying it out with Unity, you will probably have more success with coding!
@@albertbozesan Or rather, the Unreal engine. ChatGPT must have a lots of data on C++ code from Stack Overflow and elsewhere.
@@somedude5951 also an excellent option. I've even seen some people get Blueprints out of ChatGPT.
What is the name of tool to auto generate image in your video ?
I used the Auto1111 UI for Stable Diffusion. Check the description and my other vids for more info :)
I tried using chatGPT to help with GDscript, but it is not good indeed. It writes methods that are from another classes, doesn't really know what parameters a method takes. I believe the model had low amount of samples to work with
I wish we could fine-tune it like dreambooth. Can’t wait for open source alternatives!
Very good, just there are some "holes" in the explanation when you get faster... Don't listen who wants short videos, or divide the contents in more videos, part1, part2, etc. Another good video idea would be to have a Stable diffusion training of an original character, and use it in a game, since while using Stable diffusion this way you did here is good for locations, drawing the same original character on screen in different poses is difficult without training, or not?
Thank you! This was very much not a tutorial, but an experiment - my game dev skills are too bad, I went back, fixed and redid things a lot. I cut out around 4-5 hours of frustration here 😂 I will check out the character pose idea. You’re right, it would probably require training.
@@albertbozesan Really thank you for checking some ai character solutions! 😀
Nice game! Well done! Nice channel!
Thank you!
I always had the idea that coding takes decades, and didn't want to do it for a living. Now coding is valuable and viable
I’m not gonna lie and say ChatGPT makes coding easy. But it’s excellent at making small, helpful applets for day to day work!
dude you are genius
I prefer the term “AI enhanced” lol
@@albertbozesan Thank you for such a hard-worked video, I enjoyed watching it very much.
@@orkun441 thank you! I’m glad you enjoyed it 😄
I started using this ai thing early and was thinking of making a video on it -_-
I’m sure whatever you come up with will be different from this! Feel free to let me know when it’s online 😄
it's a stunningly!
I stop reading on Godot...🤷😅
It’s a very promising engine. Check out Godot 4 previews.
infringement the game
Find me *one* image the AI used to generate any of this content. Or, alternatively, learn how it actually works instead of blindly following some angry people on the internet 😅
@@albertbozesan Find me one image the AI DIDN'T use to generate this content. Or, alternatively, learn how it actually works instead of blindly following dummies without a clue.
@@gordonzar992 there is not a single image in the model. You have no clue what you're attacking here. Go be creative and enjoy art instead.
@@albertbozesan I know very well what I am attacking. I wrote a 101 pdf tut. on making a perceptron and connecting multiple of them to simulate an XOR operation. I have 15+ years experience of programming and tech and more knowledge on tech subjects than you will ever have. How the f*** do you think the model came to existence? Do you think the devs manually adjusted billions of weights in a mutli-dim array? No, they fed it hundreds of thousands of images, from around the internet and trained it on a huge number of parameters. Why do you think the AI is having issues with fingers? Because there are mutliple fitting images "burned" into the weights that overlap eachother producing an opacity result like in photoshops "multiply" or "burn" layer filter. There are literally images encoded into the weights, so many of them they end up overlapping eachother in the result.
As I have said. Infringement, the game.
📸
@@gordonzar992 I understand what you are describing - but it's the process by which we humans learn anything. There's not one original thought, we've always been learning from each other and reinterpreting it into original things from our faulty memory of it. That's literally the opposite of infringement. There's a discussion to be had about where the data is sourced from, but this is not the way to have it.
Can you send me that pic you took of me?
Go-Dot
I’m aware many people including some devs say it that way. But seeing as it’s named after this play en.wikipedia.org/wiki/Waiting_for_Godot, it’s perfectly ok to say GOD-oh.
It’s a joke about how it’s the game engine “we’ve been waiting for” :)
This is a really cool idea :D I recently tried out the dreamlike-diffusion-1.0 model, had some great results so far. Do you know why my AI generation sometimes freezes my whole computer (It's stuttering, Audio still playing and then freeze frame, nothing works anymore)? A couple of weeks ago I could generate thousands of images, now my computer freezes after ~1 hour of using Automatic1111 :( Using a 11GB 1080 TI.
No, unfortunately I don’t know why that happens :/ I usually get good responses on Reddit, maybe you can ask there in the StableDiffusion sub?
@@albertbozesan thanks :) I’ll try Reddit. Keep up the good content!
Actually, I think I know why. It seems to be a bug with never versions of Gradio. Assuming your problem is the same as mine, I've seen a lot of people on GitHub complaining about it and someone figured out that big part is the RAM usage. Essentially it seems that all new versions of gradio have memory leaks and that eventually makes your computer run out of usable memory, including swapping space.
Next time while using it, check your memory usage and generate a couple of images. Notice if the RAM keeps raising after each generation or if it always returns to a stable value. If this is your problem then downgrading gradio to version 3.4.1 is a fix for that. If this is not your problem, then sorry, can't help :D
@@Hazarth Thanks, I tried monitoring my RAM and VRAM, but so far nothing happened. Will keep an eye out. But good tip regarding Gradio! Thanks! A memory leak was my guess too, it's too random.
i strongly advice to undervolt the card, If you havent done it already. Far more stable. runs not as hot. needs less power and speed is the same.