Where do you get your graphic assets from?

[curry]

I'm working on an isometric RPG in Unity but unfortunately I'm just a humble coder. I'd appreciate if someone could help me find some sweet character sprites and shit.

 

[RPK]

https://www.gamedevmarket.net/ is one option

 

[RobotSquirrel]

Itch.io sells assets
Humble occasionally (but you'll have to wait for sales) the ones for Unity are usually a must buy if they have things like weapons, buildings and vehicles as these can all be kitbashed.
Blender Market sells meshes - but its tools are much better value if you know how to use em.
Artstation has some very high quality stuff, but I find their stuff is usually quite expensive for the good stuff - really amazing for kitbash kits and brush alphas.
Unity Asset store, I've bought some assets there in the past but I'd wait for the big sales normally. The Unity assets you have to be careful with due to the licensing.
MakeHuman is an amazing tool and comes with a whole stack of free to use assets. I also like that its low-poly as well because Metahuman and Clone3d(cc) aren't suitable for that kind of thing. Keep in mind the assets tend to look ugly (we're talking Oblivion FaceGen tier) so you'll need to do a lot of tweaking - also clothing options all mostly suck. But It's useful as an asset if you want to trace over the mesh to get isometric perspective perfect on your drawings - this is how I've used it in the past but mostly now I just use it as a placeholder for testing character meshes. It is however extremely useful when combined with Stable Diffusion as you can prompt the outputs towards what generally you want the final asset to look like great for concept ideas.
Mixamo is good if you don't have the skills to animate - but its inferior in every way to having your own motion capture solution.

 

[Bester]

Use stable to generate sprites. It's the future, learn now.

 

[RobotSquirrel]

Use stable to generate sprites. It's the future, learn now.

there's a legal problem with that, as long as you transform the output with your own work you should be fine but if you just dump the raw outputs into your game engine you could be sued as you'd be classified as a "prompter" and not the original artist. This is why I'm being super super careful and only using SD for references and concepts - with concepts I still try to do some transformative work on them in order to "make them my own" most of concept art is just photobashed anyway and by law this too would fit photobashing, a lot of the industry does very sketchy shit when it comes to references anyway and they get away with it, Stable Diffusion funnily enough for concept art anyway is less sketchy hence why I like it.

 

[Bester]

They can only tell it's stable when it's waifus with weird hands. So your post is incorrect.

 

[Wayward Son]

I get mine off of RPGMaker’s RTP.

 

[S.torch]

I draw all of them myself.

 

[The_Sloth_Sleeps]

AI art is cringe, especially with isometric. It reminds me a lot of the Unity shovelware craze of 2012, people will recognize AI art instantly and cringe and you've lost a sale. In short AI art is a dead end. And while we are on the topic of shit things do yourself a favour and ditch Unity.

Anyway, for my project I do all the art myself but I get a lot of help, I beg, borrow, steal and buy. But I always change the form in some way. I use a combination of reference photos, purchased assets, 3d models that I render in Isometric. I can pretty much make anything but it takes me a while compared to a real artist.

I recommend getting at least good with Krita and/or Gimp. And by that I don't mean just blindly practicing. What I actually mean is you probably have to change your way of thinking about the art creation process. For many years I tried to create art under delusions of how art is created, which was frustrating.

Actually for reasonable results, you don't need to be artistically talented but you do need to get some strategies under your belt for how to create. And those strategies have to match your ability. I know that sounds a bit abstract, but once you dive and start creating in that may or may not make more sense. I would advise building off something already existing to start with. Learn how to use layers. Don't hand draw too much, use the line snap tool for free drawing, etc, etc.

 

[puur prutswerk]

I have spent pretty much my entire disability benefits on Unreal marketplace assets for years. It helps me regain my imagination which left me on my 25th birthday.

An actual game by me will probably take a very long time.

 

[what am i doing]

Use stable to generate sprites. It's the future, learn now.

there's a legal problem with that, as long as you transform the output with your own work you should be fine but if you just dump the raw outputs into your game engine you could be sued as you'd be classified as a "prompter" and not the original artist. This is why I'm being super super careful and only using SD for references and concepts - with concepts I still try to do some transformative work on them in order to "make them my own" most of concept art is just photobashed anyway and by law this too would fit photobashing, a lot of the industry does very sketchy shit when it comes to references anyway and they get away with it, Stable Diffusion funnily enough for concept art anyway is less sketchy hence why I like it.

If you have an Adobe CC subscription, Firefly's been trained only on public domain material and Adobe's own stock repositories.

 

[The_Sloth_Sleeps]

Use stable to generate sprites. It's the future, learn now.

there's a legal problem with that, as long as you transform the output with your own work you should be fine but if you just dump the raw outputs into your game engine you could be sued as you'd be classified as a "prompter" and not the original artist. This is why I'm being super super careful and only using SD for references and concepts - with concepts I still try to do some transformative work on them in order to "make them my own" most of concept art is just photobashed anyway and by law this too would fit photobashing, a lot of the industry does very sketchy shit when it comes to references anyway and they get away with it, Stable Diffusion funnily enough for concept art anyway is less sketchy hence why I like it.

If you have an Adobe CC subscription, Firefly's been trained only on public domain material and Adobe's own stock repositories.

Does this really help...at all? AI art generation is a dead end. Yes you can generate an image of bear on a bicycle, so what?

But can it assist with making proper isometric assets? Based on the lack of evidence, no. All we get in the form of a solution are the "any day now" or "your prompt is wrong" quips. Or more bears on bicycles.

 

[what am i doing]

Use stable to generate sprites. It's the future, learn now.

there's a legal problem with that, as long as you transform the output with your own work you should be fine but if you just dump the raw outputs into your game engine you could be sued as you'd be classified as a "prompter" and not the original artist. This is why I'm being super super careful and only using SD for references and concepts - with concepts I still try to do some transformative work on them in order to "make them my own" most of concept art is just photobashed anyway and by law this too would fit photobashing, a lot of the industry does very sketchy shit when it comes to references anyway and they get away with it, Stable Diffusion funnily enough for concept art anyway is less sketchy hence why I like it.

If you have an Adobe CC subscription, Firefly's been trained only on public domain material and Adobe's own stock repositories.

Does this really help...at all?

Did you read our posts... at all? He's remarking that there are legal issues with a lot of ML art (because most models are trained on copyrighted art to which the trainers did not have rights) and I am bringing up a model which has no legal issues due to being trained solely on art which is either public domain or owned / licensed by the company training the model.

We are not discussing whether you like the output or not.

 

[The_Sloth_Sleeps]

Use stable to generate sprites. It's the future, learn now.

there's a legal problem with that, as long as you transform the output with your own work you should be fine but if you just dump the raw outputs into your game engine you could be sued as you'd be classified as a "prompter" and not the original artist. This is why I'm being super super careful and only using SD for references and concepts - with concepts I still try to do some transformative work on them in order to "make them my own" most of concept art is just photobashed anyway and by law this too would fit photobashing, a lot of the industry does very sketchy shit when it comes to references anyway and they get away with it, Stable Diffusion funnily enough for concept art anyway is less sketchy hence why I like it.

If you have an Adobe CC subscription, Firefly's been trained only on public domain material and Adobe's own stock repositories.

Does this really help...at all?

Did you read our posts... at all? He's remarking that there are legal issues with a lot of ML art (because most models are trained on copyrighted art to which the trainers did not have rights) and I am bringing up a model which has no legal issues due to being trained solely on art which is either public domain or owned / licensed by the company training the model.

We are not discussing whether you like the output or not.

The thread is about where/how we make isometric assets.

 

[what am i doing]

Use stable to generate sprites. It's the future, learn now.

there's a legal problem with that, as long as you transform the output with your own work you should be fine but if you just dump the raw outputs into your game engine you could be sued as you'd be classified as a "prompter" and not the original artist. This is why I'm being super super careful and only using SD for references and concepts - with concepts I still try to do some transformative work on them in order to "make them my own" most of concept art is just photobashed anyway and by law this too would fit photobashing, a lot of the industry does very sketchy shit when it comes to references anyway and they get away with it, Stable Diffusion funnily enough for concept art anyway is less sketchy hence why I like it.

If you have an Adobe CC subscription, Firefly's been trained only on public domain material and Adobe's own stock repositories.

Does this really help...at all?

Did you read our posts... at all? He's remarking that there are legal issues with a lot of ML art (because most models are trained on copyrighted art to which the trainers did not have rights) and I am bringing up a model which has no legal issues due to being trained solely on art which is either public domain or owned / licensed by the company training the model.

We are not discussing whether you like the output or not.

The thread is about where/how we make isometric assets.

And our posts specifically are discussing legality, not whether ML art makes you seethe.

 

[RobotSquirrel]

If you have an Adobe CC subscription, Firefly's been trained only on public domain material and Adobe's own stock repositories.

I don't have a CC sub (from my cold dead hands etc etc lol), I'm still using CS6, though have migrated to Krita and Substance for most of my workflows now. I only use Photoshop now for UI design and Branding/Logos (mainly because paths are awesome and combined with smart objects gives you super crisp non-destructive detail). I will say I'm pretty impressed with Krita. Also I'm on Linux and can only run Photoshop in a VM. I'm slowly migrating all my workflow over to Linux fully.

 

[puur prutswerk]

Use stable to generate sprites. It's the future, learn now.

there's a legal problem with that, as long as you transform the output with your own work you should be fine but if you just dump the raw outputs into your game engine you could be sued as you'd be classified as a "prompter" and not the original artist. This is why I'm being super super careful and only using SD for references and concepts - with concepts I still try to do some transformative work on them in order to "make them my own" most of concept art is just photobashed anyway and by law this too would fit photobashing, a lot of the industry does very sketchy shit when it comes to references anyway and they get away with it, Stable Diffusion funnily enough for concept art anyway is less sketchy hence why I like it.

If you have an Adobe CC subscription, Firefly's been trained only on public domain material and Adobe's own stock repositories.

Does this really help...at all? AI art generation is a dead end. Yes you can generate an image of bear on a bicycle, so what?

But can it assist with making proper isometric assets? Based on the lack of evidence, no. All we get in the form of a solution are the "any day now" or "your prompt is wrong" quips. Or more bears on bicycles.

Honestly, if your game doesn't have enough bears on bicycles then you can kiss the entire Amsterdam gay scene goodbye as an available market.

 

[The_Sloth_Sleeps]

So for context, I made this in two hours. I could do a lot more with it if I wanted to spend, say 6 hours.

I used the aforementioned principles (buying, borrowing, stealing, copying). I used a random image off google as a template (pictured right), set semi transparent layer, drew over, acquired textures, some rotating, cut n pasting, and applied isometric lighting as layers on top and then added features, and decals by hand using brushes. Whats important is to pick the "right" textures otherwise your final look will be off nomatter what.

I am not a skillful enough artist to create my own textures or just start drawing like that...so I use these principles. Its not even really art its just mechanical manipulation in Gimp.

 

[RobotSquirrel]

So for context, I made this in two hours. I could do a lot more with it if I wanted to spend, say 6 hours.

And this is what the AI did instantly using ControlNET with Reference. Using the blurry picture as the reference.

problems to note, - the background isn't clean (this could be fixed easily with a better reference image, especially if it was a 3d mesh render).
Quality - there's a noticeable lack of texture.
There is a lack of consistency on the building.
Pretty certain as well the image won't tile well and will need modification.
Note as well, if using a model trained on Isometric images you'd likely get a better result, this was using a model that wasn't trained on isometric images.
Sample was DIMM and 20 Steps.

Also note, if you had more ControlNETs the result probably would be better and more accurate.
I will say doing it manually would be superior, but as a tool to idea generate AI is pretty useful, just don't expect game ready assets from it you'd need to put in a lot more work to make it worthwhile, what's the point when manually you could get a better result faster.

Its not bad, but its not perfect and for isometric games it needs to be perfect otherwise it won't tile correctly and will generally look bad. That said I've seen some Isometric AI generated art that looks amazing but there's no publicly available workflows to replicate it which makes me suspicious as to how they pulled it off.

Textures.com is a good free resource if you need textures FYI, I used to use it all the time until I discovered Substance Painter, now I just paint everything.
Photopea is a program I'm probably going to move to for UI and Branding instead of Photoshop tbh, its basically the same program. Krita can do the rest.

 

[The_Sloth_Sleeps]

So for context, I made this in two hours. I could do a lot more with it if I wanted to spend, say 6 hours.

And this is what the AI did instantly using ControlNET with Reference. Using the blurry picture as the reference.

problems to note, - the background isn't clean (this could be fixed easily with a better reference image, especially if it was a 3d mesh render).
Quality - there's a noticeable lack of texture.
There is a lack of consistency on the building.
Pretty certain as well the image won't tile well and will need modification.
Note as well, if using a model trained on Isometric images you'd likely get a better result, this was using a model that wasn't trained on isometric images.
Sample was DIMM and 20 Steps.

Also note, if you had more ControlNETs the result probably would be better and more accurate.
I will say doing it manually would be superior, but as a tool to idea generate AI is pretty useful, just don't expect game ready assets from it you'd need to put in a lot more work to make it worthwhile, what's the point when manually you could get a better result faster.

Its not bad, but its not perfect and for isometric games it needs to be perfect otherwise it won't tile correctly and will generally look bad. That said I've seen some Isometric AI generated art that looks amazing but there's no publicly available workflows to replicate it which makes me suspicious as to how they pulled it off.

Textures.com is a good free resource if you need textures FYI, I used to use it all the time until I discovered Substance Painter, now I just paint everything.
Photopea is a program I'm probably going to move to for UI and Branding instead of Photoshop tbh, its basically the same program. Krita can do the rest.

Honestly I still have not seen any evidence AI produces anything of value. Like take the above. I don't see the point...its useless. Maybe I could grab part of that door and it might make an ok texture but thats about it. AI art is a proven dead end. I certainly won't waste any time on it.

Regarding textures - Textures.com is of course good but a bit expensive for me currently. Are there are any free alternatives? Do you know of any texture generators that can produce 2D textures (or if somehow 3D can be adapted to 2D, IDK if possible).

 

[RobotSquirrel]

Textures.com is of course good but a bit expensive for me currently.

it has a free section thought you have a download limit, I honestly never worked fast enough to exceed the limits and never needed the higher resolutions.
https://texture.ninja/ is also a free alternative it used to have a different name but yeah I've been using these textures for decades.

 

[RPK]

https://freepbr.com/ is a good place for textures too.

 

[The_Sloth_Sleeps]

https://freepbr.com/ is a good place for textures too.

I suppose to use it in 2D it must be put into a 3D program and rendered to image.

 

[RPK]

https://freepbr.com/ is a good place for textures too.

I suppose to use it in 2D it must be put into a 3D program and rendered to image.

probably. the only thing I've used stuff from there for is, for instance, applying a stone texture to parts of a mesh to replace a crappy stone texture on an existing 3d model.

 

[RobotSquirrel]

I suppose to use it in 2D it must be put into a 3D program and rendered to image.

Just take the Albedo and it'll be fine, its basically what a Diffuse used to be its just that it won't look as nice because all the other texture maps are meant to modify the shader so it looks nicer. You'll still get the details but minus the reflections and depth.