AI 3D Mesh Generation...Good Enough To Replace You?

AI is revolutionizing game development, extending its capabilities to include full 3D mesh and material generation. In this video, we dive into a new GenAI 3D project, Rodin Gen-1, to evaluate whether this technology can truly replace 3D artists and create game-ready assets in a matter of seconds.
RODIN GEN-1 ►► s.deemos.com/I...
DEEMOSTECH ►► / deemostech
GET 50% BONUS CREDITS ►► hyperhuman.top...
🔥 SUPPORT THE CHANNEL ►► / stayathomedev_
Thanks for watching everyone!
-----------------------------------------------------------------------------------
STAYATHOMEDEV ►► stayathomedev.com
TWITTER ►► / stayathomedev
GODOT ASSET MARKETPLACE ►► godotxc.com/
#3dmodeling #3dGenAI #3dgeneration #Rodin

[@faridkemyakov2645]

Read their terms very carefully - you don't own generated model, it belongs to the Company (Rodin). Just keep that in mind

[@_KondoIsami_]

even if that's their terms it doesn't supersede copyright laws. Any work generated by machines without enough human input can not be copyrighted, the copyright office has already made it clear multiple times that "prompting" is not enough and it's closer to a client request than creative input. Good luck to them trying to sue anyone who uses their generated AI assets, but also good luck to anyone who spends money to generate enough assets for a game and then get their assets stolen.

[@DeemosTech]

Hi there, thanks for trying Rodin. We do not restrict users from using Rodin-generated assets for any purpose. Commercial copyright is only required for another product on our platform, ChatAvatar. Sorry for the confusion. We will also open source Rodin in the near future.

[@alexale5488]

@@_KondoIsami_ HeLLo I uSe AI to gEnErATe mY aRT aNd i'M aN arTisT HuEhuE HaHa I uSe AI aNd I'm sO sKiLLeD aNd IM beTTeR tHan pEoPle wHo pUrSue aRT coLLeGe hAhA

[@ArksideGames]

AS IT SHOULD, you didn't create ANYTHING

[@dest5218]

Okay so we can you it in our games without needing to pay??​@@DeemosTech

[@silvarafafeitosa]

I want AI to do the retopology, UV mapping and the rigging while I focus on sculpting and painting not for them to do the sculpting and painting while I do the retopology, UV mapping and rigging.

[@Godot-Game-Dev]

Yeah I agree, most of this AI stuff would be much better if it was just assistance instead of trying to completely replace professionals that have spent years learning this stuff.

[@aryabratsahoo7474]

@@Godot-Game-Dev That's what Cascadeur does for animation. It automatically add the physics and secondary motions to existing animations. Sadly you have to pay to use it. You can't export in the free version(which was possible in the free early access version).

[@Godot-Game-Dev]

@@aryabratsahoo7474 Yeah I have heard of it before and it does look like a very cool tool but sadly I don't think it would be worth it to pay for really, especially because when I make games and stuff normally I go about 50/50 with 2D and 3D

[@Longplay_Games]

@@Godot-Game-Dev "ai" rigging has been around for quite some time and it's *brilliant* when it works.

[@malahamavet]

Best AI take I've ever read! If we use AI, it should serve us, not do the fun stuff style we assist doing the boring stuff, seriously

[@monsteratemygame]

These crates would be great for a horror game about a rogue AI crate factory.

[@lemming573]

You're so right. You could generate a lot of weird looking boxes that would be perfect for a rogue crate factory. You could probably drop AI from the narrative and it would still work as the factory descended into madness as the player progressed. Honestly, good idea.

[@stayathomedev]

Some specific thoughts on Rodin tool: - reference images that are generated are too small to truly see any details - the UI cuts some options off and it's not entirely clear what step you are on - I had a lot of difficulty getting true hard edges, even when generating a sword or something that would be sharp - in the same vein, everything seems to get rounded off, so hard surfaces get lost - poly count can be way too high for simple objects, i attempted a box as a simple test and the complexity was overkill with no way to adjust below the minimum polycount - LOD generation was ok, but there are better options already in other software - material generation was either good or really rough, and everything came out with an almost painted look - to that effect, the models in general seemed more suited for 3D printing than 3D games, both in material and mesh generation - UVs seems to need more margins between islands...might be better of just unwrapping in 3D software - the mesh editor was nice, but again...if you're able to do that, 3D software will do it more quickly and better - characters did better, offering actual t-posing - simpler objects did worse...just too much detail that would kill any optimization efforts when rendering

[@skepsys_d]

I love Blender and use it for so much. I would never trade my skill with Blender and the freedom it gives me for wasting time creating prompts. Thanks for another great video. Stay safe out there.

[@SlayerDUDE1993]

at some point you will be using AI just like others do, or you'll be jobless.

[@Sylfa]

When photography was starting to become a legit option for creating art you had people refusing the budge from making portraits with paintbrushes. Some of the other artists learned how to use a camera, combining their eye for aesthetics with the much quicker technology. Who do you think made the most money as a portrait artist? Point is, you don't have to throw away your Blender skills. But the most successful artists will be those who know how to use new tools that speed up the creative process, not the ones that stick with the tools they consider most pure.

[@SlayerDUDE1993]

@@Sylfa xzactly

[@JoseKalatra]

Wow, aren't you brave! Thankfully, no one is making you trade anything.

[@GANONdork123]

​@@Sylfa"Speed up the process" It took him like 20 minutes to generate an "acceptable" crate that didn't look how he wanted it to with bad UVs, bad and unoptimized topology, and less than stellar materials. It would have taken less time to just make it from scratch and the result would be way better.

[@fredericklebeault3467]

I guess this can create placeholder items in a prototype game in the meanwhile the 3d artist are working on those actual real handmade assets.

[@johnnyfatsacks5209]

A general rule for placeholder assets is they should be a very obvious placeholder so no one confuses it with a finalized asset.

[@fredericklebeault3467]

@@johnnyfatsacks5209 you can easily mass print a placeholder logo on the textures

[@JanPospisilArt]

It's like if you photogrammetried a memory.

[@3dguy839]

Lol Exactly Almost like starting off a sculpt with an ok Maquette

[@JakeCattrall]

Meshy is, at most, $1.00 for 50 credits, Rodin is, at most $1.50 for 1 credit. Somebody in the business department is obviously a little disconnected from reality

[@Corruptinator]

I would say this is useful if you want to make reference meshes so then you can make a sculpted model based on that reference for further detailing/cleaning/passthrough to make your finalized mesh. Not an instant mesh creator but a time saving tool.

[@alek2341]

The issue I have with all these tools is the same issue I've had with "templates." It's never really quite right, and the people using this won't know how to "fix it" and thus will always have the "generated" look. The people who know how to fix the issue will just be making their own models.

[@trainee5471]

9:12 I find it impressive that the mesh was not generated using marching cubes, I don't see any signs of axis-aligned lattice typically seen with marching cubes Or maybe it was, but it was optimised with regular algorithms, which is considerably less impressive

[@InvasionAnimation]

First payment. They expect us to pay for ai generated stuff that barely takes any effort. And these are the same type of folks that say art is for everyone and they could use whatever they want in their models. Honestly humanity is getting to dang tiresome with their fake slop, and lack of skill. I hope this stuff gets good so I can take it into the woods , make my own cartoons, and leave this bull everywhere.

[@phrank1596]

Yeah, it takes barely any effort - they don't want you to pay with effort. Because it costs a ton of money to run such ai models. If you look at the hardware neccessary to run such stuff you get into hundrets of thousands of bucks and sucks electricity like a whole neighborhood. So yeah, giving that away for free wouldn't be a very sustainable business model. And yeah, neural networks need to be trained. The only way to protect the art of artists is to stab every humans eyes out so nobody can "steal" their art. But sadly it will never be "good", because of the nature of GANs. They are just text completion software on steroids. They will always lack in coherency and hallucinate stuff to fit their token size. They might be a funny tool to support creative work but will never fully replace it. At the end it is just another hype from silicon valley to make money out of bs. 😉

[@InvasionAnimation]

@@phrank1596 That's good actual talent won't be replaced. Still ticks me off that they are taking so many other peoples hard work without permission and expecting payment. It really makes you wish someone replaced all their money with fake money and see where mass producing lies gets them. If they trained it off of non copy right images and stuff like that then it would be better.

[@spacecityryder]

I think AI will make good, human art more valuable. We'll be flooded with AI garbage like what's already happening with videos & advertisements.

[@InvasionAnimation]

@@spacecityryder I agree, but it will be way harder to find it.

[@JoseKalatra]

@@InvasionAnimation There's the cope again. And i wish someone replaces your pencils with sticks! Ha ha! Take that! BTW how do you know they took anything without permission? How do you know they didn't use copyright free data? Is it just the cope speaking? BTW fuck copyright.

[@rashadfoux6927]

I think the problem with AI 3d model generation is that there are too many factor's for it to keep track of for an algorithm to do a good job at any one thing. Overall shape matching with a concept, rig ready polyflow, uv's, textures, are each extremely technical and deep disciplines that need years of training to master. The results of an algorithm vs even a beginner artist are always going to fall short because it has no ability to understand what it's doing. Alot of the optimization will be mitigated in the future by things like Nanite for Skeletal meshes from UE5, where the polycount won't really matter anymore. But matching to an image involves a general understanding of things not depicted in the image, along with being able to edit a mesh to fit to a specific production workflow. This might be solved when/if we have an AGI, but algorithm's aren't going to get us there.

[@InvasionAnimation]

AGI is basically a human it feels wrong to force it to work for us.

[@kingoietro99]

Great overview of the problem

[@user-rx3xl7zn1u]

It needs a retopology/optimization step before texturing. I assume exporting a high resolution mesh might give better sharp edges and detail and then retopologizing that mesh in Blender would make it actually game-ready.

[@biraj8008]

Maybe I can use it to make 3d low poly games something like those psx horror games

[@TheManOFCulture155]

Yeah it's actually very Good for Low Poly Models but for like Realistic Models no

[@andguy]

Tbh if that’s what you want to do you’d be way getter off learning blender so you can make exactly what you want. Watch “Character modeling for beginners - Blender” by Joey Carlino, he spends the first few minutes showing how quick and easy it is to model a psx style character.

[@ince55ant]

@@TheManOFCulture155 how is it good for low poly? it looks really bad for minimizing poly count

[@TheManOFCulture155]

@@ince55ant Yeah I Testet it out for myself and i Agree it doesn't look so Good

[@_KondoIsami_]

its not low poly tho, these assets are high poly and just kinda look low poly because their low quality. Your game would have bad quality assets with horrible performance if they all looked like this.

[@mmaaauuuuiiiii]

This will be really useful for creating prototype assets

[@9DDamian]

Guess now a lot of skills and talents are going down, but I prefer the good old classic way where I do it my way, am not that lazy to let an Ai do the thinking and working for me. And proof that we like it or not we are to have a cyber war in year to come.

[@nixellion]

Yeah, it's definitely rough, but I think it could be used for some types of games, either with some auto retopo tools. Maybe nanite for static meshes, but yeah. However, reading their website, it says they use a 1.5B model. It's actually very small, comparatively speaking. SD3 medium is 2B, and SD3 large is 8B. In the world of LLMs 8B is kinda the starting point of them getting coherent for general purpose use. Normally people prefer 30B or 70B models, if they have the hardware to run them. And 3D generation is clearly just the least demanded thing on the market right now. But I'm pretty sure there's a lot to achieve by either improving dataset quality, or model parameter size or both. At this rate I'd bet we'll get something really usable for 3d generation in maybe a year or so.

[@James-pr5kt]

AI asset generation needs to be modularized. Most serious devs/artists want to hand author the portion of the asset which allows them to express their creativity the best, then let AI fill in the supporting aspects. For instance you may want to quickly block out a mesh, but generating & mapping textures is something AI should be able to do well when provided with the mesh & a prompt. Yet afaik, this isn't something that can be done by any application with a decent interface... you're down to hacky uses of LLMs + Diffusion models or janky command line tools you'd need to spend dozens of hours troubleshooting & tuning to generate a mediocre output. Same for retopoing for performance, no one likes doing that, it's menial grunt work, yet the result you always want is to keep the "broad strokes" of the model while reducing the vertices. There's almost no room for creative expression. It's just a form of compression, sorting vertex data by importance, then representing the least important vertices with an approximation of those vertices using a smaller amount of vertices. Where AI can provide value is in determining the importance of vertices to the overall look of the asset so they can be accurately sorted. This should require little more than just feeding training data to the right ML model. Sourcing the example data of retopoing is probably the harder part, but it definitely exists and could be collected, probably not even at all that great of an expense.

[@lethn2929]

Hello fellow AI sceptic, the more they roll out AI for every damn thing in existence the more hilariously bad the results are going to be, it's a good thing you went with a box example because that's a good idea, even a newbie Blender artist is going to be able to do a much better job. There's also the issue of how the generation is not going to be able to do proportions, topology, UV editing all that extremely specific stuff that needs to be done neatly. Oh and good luck trying to create a properly functional rigged character if the mesh is completely borked from the beginning. Will that stop people from using it? Of course not, we're probably going to see a massive influx of scam games popping up on steam as a result of this as well as a few sneaky attempts by AAA studios to make their lives easier instead of doing a good job and morons won't be able to tell the difference or understand why any of this is bad. Expect to see games inexplicably popping up quite regularly with a 60gb+ install folder despite being only a couple of hours in gameplay time. I'm trying to be polite because youtube comments but I burst out laughing when I saw the wireframe, what you should do is show the AI guys what a proper low poly box should look like then maybe they can do something decent with it but this obsession with AI trying to take over everything is just hysterical. By the way, we see how badly AAA studios are doing with their own character models where they heavily rely on photogrammetry rather than any kind of solid modelling technique, it really starts showing in some of the latest titles they've been trying to push for.

[@JoseKalatra]

That's a lot of cope right here.

[@InvasionAnimation]

It will unfortunately get better. And now so many people are going to have to scam the rich since we are getting our jobs replaced.

[@InvasionAnimation]

@@JoseKalatra How is it cope? He is saying a buttload of AAA studios will use this as a shortcut, they will. That many scam games will come from this, they will. It is already in dang near everything these days.

[@katherine_elle]

@@JoseKalatra What do you mean cope? The topology is nightmare fuel for any 3D artist, there's really no comparison or competition. It's way faster to model these things from scratch than to fix that nightmarish topology and any generated UVs and textures. No self respecting dev would even use it in a game because of the number of polygons and when they do, it will be pretty obvious visually, and from the performance. Secondly, the more generated models that exist out there in the wild with bad topology, there more they're going be fed into training data, ruining the AI model even further. Do children like you only know to use the word cope these days without adding anything of value to the discussion?

[@JoseKalatra]

@@katherine_elle I don't care about you or adding to the discussion. You talk a lot about what's "GOING TO" happen, what people are "GOING TO" do, or about "training data, bad models ruin other models" like you know what you're talking about. That's just cope. Ya'll just sit in your echo chamber pacing back and forth, coping that "I-i-i it's never gonna get the topology and UVs right a-and no one is gonna use it. And if they do they're bad people. Not like us, the good people, drawing with pencil people." Y'all just spew your cope unprompted all over the internet in hopes of what? The AIs are gonna get worse or just disappear and people will salute all you brave fighters for defeating the planet? It's all just cope. It's literally the same thing that happend with image-gen AI. It's never gonna be this or that, but then it is and it turns out you were just coping and the AI is still here. And you going ad personam just proves my pont. You feel threatened, you cope, then you get mad at people for telling you that you're in your feelings.

[@thephilvz]

Why not making it from the start? Starting with this as a base for a retopo would be way faster than coming up with the shape from 2D concept art (for my limited artistic ability at least). You still gotta work but you can't say it has zero value. Not sure how salvageable the material would be once baked on the new topology, you could probably use it as a base layer and correct the flaws. Not sure how it would go.

[@neuplop]

Actually, the last model from the previous video seems to have the same stuff as this ai generated stuff, i wonder if you got scammed there.

[@watercat1248]

Yap I agree it better to spend few minutes or worse to 3D model or use free asset for using those, AI tools. The only AI tool that have use few times is. Imagine upscale and background remover. Those save my some time.

[@giggio1747]

People must understand better how to apply tools like these in their workflows. And companies must understand better how to sell these tools. They sell a very polemic dream of replacing real people tasks. And some of us take it too seriously… Come on, we are not that close from delegating 3D game asset creation to an autonomous AI like this…and that’s pretty obvious to me when comparing the needs of good games with what the tool delivers, in technical and artistic terms. However, we can’t deny how impressive this kind of tool is. And how powerful it may get in a few months. Will it be able to replace a good manual 3D art workflow? Probably one day, but not that soon. But despite all of this, I loved it. And I’m very excited to see where it’s going to. Because I’m no true 3D artist, and I a have a true difficulty for getting things out of my head, and turning it into 3D models. So, prototyping props, level design and even characters are painful to me because I know my skills are kind of limited in that tasks… every iteration takes forever. But a tool like that can help me quickly prototype and tests shapes, materials, sizes, details etc in a way I can keep making changes until I find the best approach to actually go after having a proper professional 3D model, done by a professional and human artist. So, currently I see it as just another tool to help my prototype workflow.

[@amdrewhamris]

How were you able to get the textures to work in Mixamo? When I upload the zip file download from Rodin (either obj or fbx) it's missing the textures.

[@stayathomedev]

Uploading the zip worked for me. Otherwise. Maybe repackage in blender?

[@jamesrivettcarnac]

I am with the sceptical view, but not the 'this will never work' It will, just take a look at the difference between 2 years ago and now. And it's worth learning how to use, at least you can choose not to.

[@NexusBaum]

thanks for the review! I'll learn the thing xD

[@queblegamedevelopment4143]

This is really cool! Obviously it's no where near like 'actually useable' in an out of the box way, but just like all these other AI tools, I'm excited to see where 3D generation gets in the future! Great video!!

[@BoopTheSynth]

Hate to break it to you but it's not going anywhere. 3d is just too specific and has too many rules that need to be followed for ai to even get close to a usable asset

[@warsang]

If you're doing a top view game where the level of detail doesn't matter as much, this is probably good enough; kzbin.info/www/bejne/iIjPkHmEhZugp5o has a video where you generate the concept art with Leonardo before passing it to Rodin. This works super well and mixamo auto-rigging it for you allows creating a ton of characters within minutes. Sure, if you're going for ultra-realistic high level of detail, this won't do but if you're an indie dev trying to get a quick proof of concept running or your game is from "far-away" ( RTS etc.) then this is probably exactly what you're looking for. Personally, this blows my mind

[@surplusking2425]

A box with mesh count 3K? You need RTX 4090 to run them properly for a Brawl Stars like stylized graphics mobile game

[@PretendCoding]

I think this, much like most AI, should be used as a starting point to be refined by actual artists. I use AI for code here an there, but it never does a good enough job to just "copy-paste" and have it work how I need it to.

[@CausticTitan]

It will actually take more time (like A LOT MORE TIME) to refine an existing model for something like this that it would be to start from scratch using platonic solids.

[@deadlock_problem]

AI garbage in AI garbage out

[@kurtu5]

Garbage UX. I tried the site for 2 minutes. I will never be going back.

[@armandadvar6462]

it is not free, only 10 credits not good

[@theoathman8188]

Results are garbage and not usable. but again, this is a developing field. AI might be able to generate very efficient 3D model in the future.

[@samamies88]

Depends on game and style. For example slap psx filters&shaders or make citybuilder where big houses u make by hand but small stuff such as barrels u can easily make with program like this. If u make it low quality then details matter less or if player views it from far away then details can't even be seen. I only agree with u if dev is wanting to show details to player. But not every game has those.

[@theoathman8188]

@@samamies88 I wouldn't recommend using these models raw. You could get inspired by it or edit them in blender,, because the vertex count is way too high

[@mehdiarjmandi5280]

:D

[@justagreenguy]

"BuT iT wIlL oNlY gEt BeTtEr"

[@AzoTheRed]

Did they say it was pronounced "Roh-DAN" or are you doing it that way to be a rebel...

[@stayathomedev]

It's named after the French sculptor. So row-DA(n) is right

[@DigitalCanineGames_]

FINALLY, a reason for me to not learn blender

[@kashifalam-hs7sx]

Hi , Your Videos are Amazing Do you edit Your Self or you have any Editor? We have a team of professional Video Editors From Different Countries, Would you like to see some of our Samples?

[@NeonfireStudio]

I created Most of the models by AI, (Meshy AI) both of this are nice!

[@SovereignVis]

Yeah, when making models for games, you always want to try to go with as few polys as you can get away with. This AI, doesn't really try to do that, and it looks like such a mess to clean up that you are better off just modeling it your self. 😅