AI materials. any progress recently?

gnoop

Does anyone know if anything cool appeared in AI  textures/ materials field?    Ai normal and height map  from photo?

I tried  recent Sampler and see no difference , still same  blurry , soft, unspecific normal maps .     Any better alternatives recently?      I sort of lost my hope and stopped to invest my time but still wonder if anyone  follow it up  closely.

  I have an impression  some  texture providers started to use  Sampler  instead of actual hi-res phtoscanning and you figure it out  only after buying.

Alex_J

Not AI, but did you know about this: Bounding Box Software - Materialize

gnoop

Alex_J said:

Not AI, but did you know about this: Bounding Box Software - Materialize

Based on unspecific  , rounded  bricks  image on their  front page  it looks no better than  same old crazy bump  where ugly normal maps are instantly revealed at any tangent light angle .   Isn't Unity  bought some  AI company  few years ago   and develops its own AI  tools as integrated something?        Have anyone  worked with them?

Anything new for materials  in recent Photoshops , beta Photoshop maybe?     I stopped  follow  and even hit update button  while still paying my subscribtion.  Last one was sbs  support in photoshop . Totally useless.

Alex_J

it says on the page the tool was developed and used for uncharted games and now is open source, haven't used it myself though

gnoop

Well . People in gamedev used crazybump a lot too  till  jobs rivalry  and  new visual standards for realistic graphics gradually weeded it out .  I still do same in Zbrush  although when an image  resembles   height/ depth  info.

zetheros

you can try Xnormal

gnoop

zetheros said:

you can try Xnormal

Anyything AI in xnormal?    Anything new at all?

ZacD

Those generate normal map from image tools are pretty much bumpy noise generators (which sometimes is all you really need). You'd think this is something AI image generation would be good at, and there's a ton of training material out there. 

Eric Chadwick

There are certainly tools out there using Stable DIffusion and the like to use neural networks to generate normal maps from photos.

The problem is they mostly rely on illegally-scraped datasets. If you're OK with the ethical and legal issues of that (I'm very much not cool with it) then have at it.
https://pypi.org/project/imaginairy-normal-map/
https://github.com/brycedrennan/imaginAIry?tab=readme-ov-file#depth-map-control

There are ethically sourced apps though, for example
https://hugotini.github.io/deepbump

SnowInChina

not sure, from what i have seen until now, every tool had massive problems to distinguish form from texture

you can see this in erics post on the reflections in the windows

every tool i've seen until now had the same problems with normalmaps and roughness

and i'm not really sure if the training argument counts for this kind of training? i may be wrong, but if they train the ai for stuff like normalmap generation from fotos, you're not really creating something out of the work from others, you just teach it what is what

if yoiu create materials with it, that would be a different story ? i'm not a legal expert but this is my opinion

gnoop

All the examples I saw so far looks like havin a sort of a bloom added to normal map making it looking blurry on actual material .  Except only bricks maybe. 

Eric Chadwick

SnowInChina said:

…

and i'm not really sure if the training argument counts for this kind of training? i may be wrong, but if they train the ai for stuff like normalmap generation from fotos, you're not really creating something out of the work from others, you just teach it what is what

if yoiu create materials with it, that would be a different story ? i'm not a legal expert but this is my opinion

Internet-scraped data is an infringement on copyright. I posted some info previously: https://polycount.com/discussion/comment/2785084/#Comment_2785084

Shrike

Is there really a large niche for AI materials?

1. Smart materials are king, and honestly the smaller and more selective your library, the better
2. You already have a infinite supply on (scanned) textures essentially, although some things might be rare

The one use I can see is a full texturing pass on an asset, but then again thats "standard workflow" which personally I think is outdated, dosnt scale with project size well and is unflexible. COD is now going 300 GB apparently and many games have to make a lot of cuts in memory

(On the other hand, we will see ML compression soon based on some research which might cut texture size by 5-10x apparently)

I assume the real niche with AI materials is in junior artists or programmers doing 1-2 man games and just cobbling something together with standard workflow or such. Although I could see some sort of pipeline where you generate a height map for a decal or something like that or starting point for a designer graph. Maybe I just dont see the vision yet, anything is possible, but the more I get better, the more I value extremely compact and extremely polished libraries, which feels like the opposite.

Edit: Ok, Hand painted or special styles, thats definitely a big case but then really requires consistency.

poopipe

for me the value is in improving on the algorithms for tasks like denoising, height map/normal map generation, light removal, edgedetection,  tiling etc. 

These are things that can't be done perfectly with generalised standard algorithms as they require intuition about the content of the image.  ML models can provide that intuition. 

the nice thing about reducing the problem space to focused tasks like this is that you can reasonably collect/create the training data required to make a decent model without stealing shit off people.   

gnoop

I am just looking for any new  tools  that could provide quick solutions .   I try Sampler  regularly   and I still see nothing really good except bricks surface  and some small pebbles maybe.    Still a bit inflated , not as  crisp in details and sharp edges  as hires photogrammetry .   Height and normal maps IMO are a key component for modern materials.   Color is not that much really   since I have yet to see any good delighting   AI.  Everything is derivative form height in a word.

Does anyone know something new and AI  in Unity realm ?    Our company once  ( few years ago) had been subscribed to  Artengine   but that time I found I could do it more easily in Photoshop and Designer.  AI height was awful.   Seams removal  and  chunks compositing  to compensate visible repeating  was easier or rather more controllable  to do in Photoshop or any non destructive vector editor .  

I know Unity bought them since and  I wonder if anyone have any good experience ? 

ps. IMO material AI should be trained on hi res photogrammetry  mostly.   Done by some drone that fly close to surface  and hi energy  flashes to override  sun light/shadows.  Just not a single 2x2 m super repeating piece like  Megascan  but rather many shots of a big span of same subject.  

dimwalker

Shrike said:

Is there really a large niche for AI materials?

Niche example: old-ish game moves to new engine or you are making enhanced edition. Doesn't happen that often, but when it does, you are sitting there with hundreds of diffuse textures that need normal maps now. Possibly also roughness, metallic etc.
Recreating them properly in designer would take too much time/money. Using something similar form a library that has full set of maps is not always an option, because you might have geometry cut to those specific wood planks, stones, bricks, cracks, trims etc.