Unity AI proves highly effective for automating development workflows, yet its generative capabilities still produce inconsistent, low-quality assets. It is currently a valuable tool for logic that remains a "slop factory" for creative content.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
Unity AI - Slop Factory or Awesome New Tool?Added:
Hello ladies and gentlemen, it's Mike here at Game from Scratch. And I said earlier uh there are three things in life you just don't talk about online.
Politics, religion, and AI. So again, we're going to talk about AI because that's coming to the world of game development. Specifically, Unity. Unity launched their AI in open beta. Uh it is available. There's a free trial available here as well. And if you've got a pro account, you get 2,000 credits a month. So we're going to go ahead jump in and take a look at what their AI is all about. Here is their launch. uh trailer. I got very good views. Uh definitely. So for in the last day or so, it's got almost 300,000 views. So what is the reception to this? Well, uh hey AI, send emails to 25% of staff layoff notices. Once read, lock computer and building access. Oh yeah, use algorithm to choose which to fire. I'm currently away from the studio. Uni, create me an open source version of yourself in case of any more licensing shenanigans. U then asset flip now, uh AI flip. Uh, and yes, make game button, make car shooter, make multiplayer, make no mistakes, no yapping, make it fun.
Uh, what are you creating? Prompts. The initification of gamedev only makes those handcrafted experiences even better. This is a massive backfire. AI should be tackling the mind-numbing grunt work, spreadsheets, docs, and clerical tasks. Games, however, need to be modeled, sketched, and envisioned by hand, and so on. So, you get an idea.
The reception of this is um it's not great and I find AI tasks in general very very very polarized. So what we're going to do is jump in and take a look at this right now. Uh again if you want to check it out yourself they are doing a uh free trial that's available that will give you a th000 credits. You got to use them within uh 14 days in that case. Uh if you have a pro plan however you are getting 2,000 monthly credits.
Uh and we're going to deal with exactly what the heck a credit actually buys you in just a sec. So, let's move on to the hands-on portion of this. Now, this is a live demo, so you know exactly how well this is going to go. My job is not to make AI look good or AI look bad. I'm going to try and leave my opinion out of this completely because I know uh 30% plus of you vehemently hate AI with a passion and then some of you love AI.
The question is ultimately going to be, has Unity just turned itself into a slot factory or have they added a very cool new developer tool? And you be the judge after this video is over. All right. So, if you want to go ahead and check it out, what you do is you click on this tab. The first time you click on that tab, it will ask you to agree to the terms and then install the required packages. If you do not do that, none of this AI stuff is installed. Something key to realize. Uh, also by the way you go into manage your account and turn developer data sharing off. If you do not want to train their model with your code, I would recommend doing so. So then you can look at AI as one of two tasks. Basically you have the um the agentic and chat side of things and then you've got the generative stuff. Now the generative stuff is basically what we had with Muse before just many times better. They've launched AI back like two and a half years ago in Uni and it was completely and utterly absolutely crap. It was basically a reskin chat GPT. Muse was garbage. Pure garbage. Now they've kind of basically done a complete rewrite. So how have they got it? Well, this is the side of things I think is actually kind of cool. This is the agent side of things. You got agent, you've got planning, and you've got chat. Chat's straightforward. This is a version of like Gemini or Chat GPT or whatever that's been trained on Unity's data uh databases, instructions, manuals, that kind of thing. So, let's say uh how do I apply a um forward uh force to a physics body object? Uh so, like say boom, select that. Now, you may be wondering, okay, what is this costing me? Like 2,000 credits is 2,000 credits.
What does that actually mean? Well, you're going to find out after you ask.
So, it's actually going to give you the credit cost at the end. Um, basically, it's a couple of credits per action.
Some things take more, some things take less. And the thing is, you don't actually know. So, that's kind of one of those recurring things I'm seeing a lot with AI lately is a lot of people are like getting these surprise bills and going, "Wow." But, in this case, you're actually going to find uh if you use up your credits, you've got to actively buy more credit. So, thankfully, you can't bankrupt yourself. You're also going to notice as it's going through this process, it's going to constantly ask you for permission. You can just cart blanch grant permission. I'll do that later on so you can see it. But basically, you can see its thought process as it's going through all of this. It's even kind of showing you stuff you're kind of like, uh, okay, why are you showing me this? Uh, and then the other side of it is uh for asking questions like what I just asked here.
This is many times slower than what I've got if I just tap over to like Gemini or Chat GPT or Perplexi or any other LLM and ask the question just uh by a comparison. But here you go. It walked through. It figured it all out. So here is the final instruction. It gives you the code for what you're working on. Uh the code does seem to be correct. It is well explained. All of that is here. Uh you can also then switch over to aentic mode and have it actually apply that code into your object. So, agent agents can do changes. So, let's go actually check out an agent now. So, we got this really cool level here. Uh, this, by the way, is um from a bundle that is going on right now. Uh, the links are down below. There's a Unity, Unreal, and GDO Springtime bundle uh with a number of different asset packs in this is one of them. Um, again, use code SP60 if you do buy it using my link. It does support me. Thank you if you do that. Again, a little bit more about that later on. But here you can see we got a number of different race cars here. Uh, we got uh 1 2 3 4 5 six race cars here. all different colors. So, what I'm going to do is I'm going to fire this up and I'm going to ask it to change the colors.
So, like here, and I can say, all right, um let's switch over to agent mode like so. And by the way, you can see right here, four credits. So, that took four of my 2,000 monthly credits to answer and come up with that code, just as a frame of reference. All right, so uh change all of the race cars in the scene to be red.
And this is actually going to make changes to your uh scene. Now, one of the good things about it, and I will show you after the fact, there does appear to be a pretty solid roll back feature here. So, if we do an undo after this agendic change, it's supposed to pull back all of the changes that it made. So, again, this is going to run through. It's going to do the logic. And then this time, I'm going to say, you know what, just for this conversation, do what you want. So, it's going to go through and figure things out. Now, it could be a couple of different prompts that it figures out, but it's basically explaining its thought process as it's going through and trying to figure out what cars are, what red is, and how to apply red to the car. At the end of the day, I don't really care about its thought process that much. Although, if you're learning, hey, it could be a good learning experience, at least to see how a robot comes up with this stuff. But, it's going to go through the reasoning process and then hopefully it's saying there it changed five of our six race cars to red and one was already red. So this is the side of things I think could definitely be useful even if you are like a died in the wool AI hater. I think you can see how like an agentic function built into um the uni game engine could be a very valuable thing.
Let's see how it did. Boom. There you go. So now we have just red cars uh and then the thought process behind how they did everything. So that again it's one of those things that would take time to do and this is definitely a timesaver.
And then boom, I hit control-z, we undo the results. That in my opinion is genuinely useful. And I'm going to show you another thing here. So let's go ahead uh back here and we're going to do another thing. So again, this is the part I think is genuinely useful. We can also go ahead and pick some assets in the game world. So here we got three different kinds of balloons and we're going to attach them in. And I'm going to say create up to 30 instances of the hot air uh balloons uh aligning with the track location but slightly above.
And then this is going to theoretically instantiate using those three different assets we sent in uh and put them into the scenes. Now, you can also use Ajentic to do things like coding, like we did earlier on. If I wanted to apply a rigid body to a selected object, I could drop the object in like I just did here and say, "All right, apply blah blah blah code to it." And it will uh do what it does. Now, this is going to go ahead and run. I'm going to pause while it's finishing. All right, probably about 2 minutes later, I would say we are now done. Again, you have a summary of everything it did as part of its thought process. And let's see how it did. So, we should now have some hot air balloons. There we go. Hot air balloons.
Uh, are they following the surface of the racetrack? Yeah, pretty good, actually. So, in this particular case, I think it actually nailed it. Now, I did play around earlier with trying to get it to create a full racetrack. Uh, and the results are bad. Like, very, very bad. I might try that again as a bonus at the very end of this video. But now, I want to move away from the agentic side of things and go to the most polarizing side of AI. And that is the Janet AI stuff. And you've got a bunch of things here. Now, they've moved away from the way they did things before. And what you're going to notice now is it's all using um third parties for the most part. So, example here, if I wanted to create a 3D object, uh I come in here, I create my new object. So, you see it down at the bottom left here. Down here, it's saying what do I want to call this?
I'm going to call this an RX7 for reasons you're going to see in just a second. And then what you're going to notice is you have multiple different options here. Now, this is the part that's getting very scary cuz this is mesh generation either from an image or from a description. Uh, but here I'm going to feed it three different angles.
And I think you artists out there, you're not going to fear for your job when this is done. So, we got a bunch of different ones. We'll do this tripo 3.1.
There's also Hanan 3D 3.1. These are common models out there. They're using them. I'm going to use triple uh just basically at rando. And then I've got already so I went through uh with an image editor and I took uh same source RX7 uh gen two and I basically clipped out front reference, side reference and back reference images for it all from the same car. Uh and now I'm going to go here front reference. So let's grab our front and then our back reference. Let's go ahead and grab our back and then our side reference. I think that would be left. And let's go ahead and do that one. And then uh right oh hell same thing. All right there we go. So there we go. We generate our car and this is generating it from images and uh this takes a while. By the way they went to the Microsoft school of progress bars.
So it jumps to 25% right away and then uh then nothing. Uh so I will be back when this is done. It's going to take probably a minute or two. Actually while it's running I want to point something else out that's kind of interesting. One of the big problems with this Gen AI stuff is what data sets are these being trained on? Especially you get into like ethical issues. Are you ripping off artists? Did you acquire from open sources? That kind of thing. And how is uh Unity dealing with this? They're dealing with this with a disclaimer.
Generations may be subject to copyright.
And then AI generated assets may be subject to copyright. Always verify the appropriateness for use. So, uh yeah, they're just throwing the buck down to you, just so you know. Uh so there there is uh again definitely copyright and ethics issues behind any Gen AI stuff.
If you do not know what the training data set was, uh that has not gone away.
And the way that Unity are dealing with it is basically with this little disclaimer here, just so you're aware.
All right, so that took about four minutes. Here is our end results. And artists, I don't think you're at worry for your job quite yet. So uh that is what this particular model change created. I tried this test on a couple of other ones, uh, like the the different options that were out there, uh, like this guy over here and so on. I tried it just straight out from a prompt. It actually works a little bit better if you just use a single image as opposed to multiple images, but every single time the results were well, they're monstrosities. So, I wouldn't worry too much if you're an artist, at least not yet. This technology is definitely going to improve when it comes to 3D generation. Uh, but I Yeah, so this is the results that I'm getting. Anyways, uh again, not cherrypicked. I let that one run live as we went. Now, there are areas where this does get a little bit more scary. One of the things that they they introduced with Muse, which actually had a little bit of potential, but was janky as heck, uh was this new animation system. So, let's go ahead create a new animation here. We'll call this cartwheel, like so. And then was a prompt do a cart wheel. And then we can say don't um Okay, not jerky. Okay. Uh uh and then we can say how long to make it. Uh so 2 seconds and generate. You see this is costing me five in this particular case. So everything's got a slightly different price going on. Here is our cartwheel.
And honestly for for raw animations, the animations are improving. They're still like sloppy sloppesque, but again they're improving. Actually, let's try another one. Do a roundhouse kick.
And we'll generate that one. And we'll see what a roundhouse kick. Now, the animation stuff is actually pretty fast.
So, again, it's it's getting there. The animation side of things I actually think has improved quite a bit from when I last saw it. And I could definitely see how you could use something like this um you know, at least for prototyping for sure. And then, by the way, you do have trimming tools in here.
And you can also uh set it as a root motion so that it stays stand like so it stays stationary uh in this supposed to anyways uh so that it uh you know doesn't move away from the center point.
It might have been this one edited anyways gives you an idea that's what the animation side of things is. The other one that is kind of getting a little bit scary uh and this is definitely an area where people are using things and uh voice a voice artists etc are should be pretty scared by this is the sound side of things. Now their sound approach they did uh my my debut sound or whatever is kind of similar. They're pairing up with other people to do these models. So we've got uh 11 Labs and LIA uh here. So you've got sound effects and uh voices. So, if you want to create someone saying, "So, uh, ladies and gentlemen, start your engines." You need someone to say that.
>> Hi, nice to meet you.
>> All right, we'll let Adam do it. We'll go ahead and generate this one. Again, five tokens is what it's costing me out of my 2,000 monthly. And this one, I think, was pretty quick to gen. Again, same deal. Maybe subject to copyright and so on. Here is the generator result.
Ladies and gentlemen, start your engines. It's >> It's not bad to be honest. It's not It's obviously AI, but I could see how you're going to see more fully voiced characters for better or worse because this is very accessible and easy to get a hold of at the same time. Let's try it for sound effects. We did an RX7 earlier. Uh rotary engine from an RX7. That's very specific. starting up and revving a few times. All right, let's make that 5 seconds long and then see how it does on that effect there. So, you know, you this is where you'd use it for like, you know, gunshots and again engine sounds and various different sound effects.
This is AI gen on that side of things.
And let's see how it did.
Okay. Uh, that one was definitely a fail. So, let's just do uh a sports car engine. I was being a little specific with an RX7 rotary engine. And they have a very specific sound. It's not that sound, by the way.
But let's try something a little bit more uh mainstream. Let's see if it does a little bit better. Again, I'm just showcasing one AI model. There's a bunch out there. if you're into the Geni Gen AI.
Okay, that was definitely better and I could see how you could draw. So, if you're out there and you're making sound effects packs, this is starting to get very scary. And once again, where is this trained off of? Uh, like where did they get the data set to do this? That is always the moral question you have when dealing with any of these things.
Now, I'm not going to go through the rest of the Gen AI stuff. You get an idea of how it works and what they've done. It's basically they are integrating with third parties quite a bit more to create these instead of you know a halfbaked license of their own.
Uh and then yeah, let's head on back over here. So again, you see if you've got pro, you're getting those 2,000 credits. Uh I started with somewhere around 1918 when we started this and now I'm at 1881 after everything I just demoed. So it gives you an idea of what you went through. Um so that was my account. It's like 50 or so for what all everything we just saw here. You want to buy more credits, it's 20 bucks for 2,000 credits. So, that is the pricing here. Nice thing is you can't overspend.
I find the pricing reasonable, although it's a little bit weird that you don't actually know what the cost of something is. Like sometimes you do know like when you're uh doing just straight Gen AI stuff, it'll show you the price when you hit the generate button. But sometimes, especially in agentic mode, you just say, "Hey, go do your thing." And then you see what the results are. So, let's go ahead with that final bonus thing.
Uh, by the way, more details of it are in the discussions if you want to get in details of all the other options that are available there. There's some profiling and debugging tools in there, some UI tools, some Figma integration, etc. Stuff that I didn't cover in this video. Uh, but you get an idea of what it's like and what it looks like. Um, and then, uh, the bundle by the way that we used, the asset was from here. Uh, so you see it's it's one of the There it goes. That's what we used right here.
this whole pack. Uh, basically use the code SP60, knock 60 bucks off and it's $39.99 for a ton of assets if you're interested. Um, and you find that the Gen AI stuff just isn't for you. Uh, by the way, one thing I want to point out here, this is very interesting. If you want to connect a AI gateway to a third-party agent, so if you've got your own thing, you want to use their MCP, it does look like you need to subscribe to Uni AI to hook up uh cloud cloud or um Gemini or whatever. If you've got your own external one you want to use, it looks like you need to be subscribed at least $10 a month to be able to do that.
And then the number of connections seems to be tied to the plan that you are working with. It's one of those things to be aware of. I know a lot of people that wanted to do bring your own AI are not going to like that fact. Uh but that's it. Okay, now let's head on back over and go out with a bang. Let's go here and we will open up the assistant.
So once again, the assistant is definitely the part of this I find the best. I I would like personally if they got rid of all the gen stuff and they just did this assistant and they just built this in. This would make most people I think very delighted. Uh you can use your own assets here. So, for example, here I'm going to go ahead and grab the whole prefabs. So, let's grab the prefabs folder in general. So, just drop that in there. And then I'm going to say make me a new scene of a race track. Make the track an oval uh surrounded by trees uh with stands for spectators outside the track area. See if that makes any sense. Agent, let's go ahead and run that and see how it does at generating a level from my particular assets. This is going to take quite a while. It's going to prompt me in just a second for uh some permissions. I'm going to let this run and I will get back to you with the results.
All right, generation is done. That cost us 48 credits. It created a brand new scene. And actually, I may have lied. I still have this progress bar going on down here. Uh I think I'm done. Let's go ahead and find out. I was going to move this off screen. Here is our track. And uh well, I think uh we can safely say your job as a game designer, you're probably safe for a while. So yeah, ladies and gentlemen, that is the end of the presentation. Uh it can work better again. Let's see that that now my 1881 is now down to Okay, wait. You know what? I'll give him a minute because it may not be done that. Nah, it's done. This is not a progress bar. This is a context used thing. So, yeah, that was the final result. I've done tests like this a couple times in the past. My results generally were substantially better than this result. So, this is the worst I have ever seen. Uh, but yeah, that that is it, ladies and gentlemen. So, what do you think of Unity AI? Uh again, it's one of the most polarizing subjects out there. And the thing that you always have to keep in mind is this is as bad as it is ever going to be. So this technology is going to constantly get better. Is this good for the industry?
Are we going to have a lot more slop from this kind of stuff? You tell me.
Let me know. Comments down below. I'll talk to you all later. Goodbye.
Related Videos
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
5 Mind Blowing Omni Uses Cases
PaulJLipsky
1K views•2026-06-02
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29











