Procedural material library creation starting from real world

davide445

For a simulation project need to create an accurate library of materials based on a specific building, that can later re-used and modified for later usage. 

Testing the photogrammetry path I find difficult and not scalable to just edit the taken photos to generated what needed.

So was thinking if make sense to just use the photo as a reference, extract the parametric data, and next generate the material using some non destructive  scriptable tool such as Filter Forge or Substance Designer. 

There is a way to extract let's say the noise pattern parameters from a photo, so to next reuse them programmatically? Or there is some reliable database where all the needed parameters are set for the various materials? 

oglu

Thats not an easy task. Thats the reason why there are libraries like megascan or tools like alchemy.

I would search for existing materials in those libraries and match them. 

davide445

Thanks @oglu for your answer from experienced pro perspective.

Since my project involve also some quantitative evaluation of different parameters I will need not only to just use visually similar materials, but being able to programmatically generate new versions with parameters set in between the existing one.

So I need both to have a reference table of the basic parameters from whom the material was created, and also being able to manipulate it to generate a new one.

In your experience Substance or Filter Forge will allow to achieve that?

musashidan

Substance Alchemist might be perfect for this. Considered it?

poopipe

Alchemist is a little limited for this I suspect - you'll be able to make the materials but applying any controllable logic to generation en masse is out of it's scope

If you need to generate the data programmatically then designer and the automation toolkit are where you probably want to start looking.  

davide445

From what I have read and a bit tested with their trial, comes down to Filter Forge or Substance Designer.

All the others (Quixel Mixer, Substance Alchemist, etc) are not scriptable nor node based. Dind't know if I miss some other app worth considering.

Problem for FF is didn't play well with current Affinity Photo (crashing), meaning I can work only on the whole image since the selection can be made only in an external application.

Designer it's a bit scary, for his complexity and fear to be locked in with increasing costs, vs tons of free filters from FF. Did anyone have experience with both?

poopipe

I've not used filter forge, I just did some digging though and while it looks pretty decent I don't see filter forge having any functional benefits over designer and it seems to lack all the scripting capabilies.

It also seems to be not cheap-you need the pro version to handle large 16 bit images so you immediately need to spend $400
The subscription cost for individuals with designer is way less

davide445

Evaluating both FF ad SD, they seems to have different strengths and weakness.

Both have scripting capabilities (FF in LUA, SD in Python), both are node based, with a vast library of pre-made textures (free in case of FF).

SD appear to be more streamlined in managing real world images to create materials, FF is probably more oriented in generate the textures. FF does have a more easy to manage UI, SD is more complete.

FF does have a 80% off campaign running as soon you are 2 days on their website, and you can renew it deleting the cookies. In this way the price for the Pro comes down to less than SD perpetual license.

poopipe

One of the major advantages of designer is it's place as a pipeline management tool - this is overlooked by most small operations as its just not that important to them but if you're proposing to build a massive library it may well be a consideration.
Where designer's weaknesses lie are that you cannot develop your own atomic nodes and you cannot effectively iterate across an image. This makes blurs and other multisample type filters difficult to create. 


For transparency...
 I'm a huge fan and advocate of Allegorithmic pipelines - I build them for a living so I have to be 
And
As I said above, I know nothing much about FF 

but

 if I were  faced with the choice I would be looking at how FF dealt with large scale automation, modularity and change propagation rather than the type of image it's geared to create - the art part is essentially an implementation detail.