USD: Export Cycles Shader networks as USD material networks. #119180

Charles Wardlaw · 2024-03-07T20:08:55+01:00

Charles Wardlaw commented

2024-03-07 20:08:55 +01:00

By default the networks are not exported, but are available now as an option on the export operator.

This is only the Export side of this patch, but it should allow for the HDCycles delegate to pick up these networks for rendering.

By default the networks are not exported, but are available now as an option on the export operator. This is only the Export side of this patch, but it should allow for the HDCycles delegate to pick up these networks for rendering.

Charles Wardlaw added 1 commit 2024-03-07 20:09:05 +01:00

5b66a37c5e USD: Export Cycles Shader networks as USD material networks.

By default the networks are not exported, but are available now as an option on the export operator.

Charles Wardlaw requested review from Michael Kowalski 2024-03-07 20:09:24 +01:00

Charles Wardlaw added the

Interest

USD

label 2024-03-07 20:09:32 +01:00

Charles Wardlaw added this to the Pipeline, Assets & IO project 2024-03-07 20:09:39 +01:00

Charles Wardlaw modified the project from Pipeline, Assets & IO to USD

2024-03-07 20:09:45 +01:00

Charles Wardlaw self-assigned this 2024-03-07 20:09:56 +01:00

Charles Wardlaw added 1 commit 2024-03-07 20:10:40 +01:00

6bbbc57b99 Merge branch 'main' into feature/usd_cycles_shaders

Charles Wardlaw added 2 commits 2024-03-07 23:53:16 +01:00

2c2db1e1f9 Moving Cycles code to separate file for ease of future updates at Michael's request.

28fcea9d4c Merge branch 'feature/usd_cycles_shaders' of projects.blender.org:CharlesWardlaw/blender into feature/usd_cycles_shaders

Michael Kowalski reviewed 2024-03-07 23:58:20 +01:00

source/blender/io/usd/intern/usd_writer_material.cc

						
				@ -90,1 +111,4 @@

				static const pxr::TfToken bsdf("bsdf", pxr::TfToken::Immortal);

				static const pxr::TfToken closure("closure", pxr::TfToken::Immortal);

				static const pxr::TfToken vector("vector", pxr::TfToken::Immortal);

				}  // namespace cyclestokens

Michael Kowalski commented

2024-03-07 23:58:20 +01:00

I believe the cyclestokens can be removed as they are used in usd_writer_cycles.cc.

I believe the `cyclestokens` can be removed as they are used in `usd_writer_cycles.cc`.

CharlesWardlaw marked this conversation as resolved

Charles Wardlaw added 1 commit 2024-03-08 00:42:59 +01:00

f21f3a6d9e Removed extraneous tokens.

Brecht Van Lommel commented

2024-03-08 15:27:23 +01:00

The shader node export code is not very generic, it requires work to maintain this as we add or change nodes. If we are going to have to maintain this going forward, I would look to have a good use case for this, and at the moment it's not clear to me who this is for.

By itself it's not that useful to be able to export to a USD file and then only render that in HdCycles, since you could have just rendered from Blender directly. If there was some way to edit these outside of Blender it could make some sense, but I don't know of concrete plans for someone to make a good editor for that?

So I would like to better understand which concrete use cases this is meant to address.

The shader node export code is not very generic, it requires work to maintain this as we add or change nodes. If we are going to have to maintain this going forward, I would look to have a good use case for this, and at the moment it's not clear to me who this is for. By itself it's not that useful to be able to export to a USD file and then only render that in HdCycles, since you could have just rendered from Blender directly. If there was some way to edit these outside of Blender it could make some sense, but I don't know of concrete plans for someone to make a good editor for that? So I would like to better understand which concrete use cases this is meant to address.

Michael Kowalski commented

2024-03-26 14:58:28 +01:00

@brecht I suppose another option might be to implement Cycles to USD conversion as a Python export hook, but I haven't thought through all the pros and cons of doing it that way.

Brecht Van Lommel commented

2024-03-26 15:42:39 +01:00

But can you explain who this functionality is for, in which workflow it is useful? As in, are there real world cases where people are using Cycles shader nodes in other software, or planning to do so?

Esteban Covo commented

2024-03-26 17:20:37 +01:00

First-time contributor

Hi @brecht , we are a 30 people studio that use primarily Blender and Houdini, since a few months back we've been starting to adopt Gaffer, this was a no brainier since it now uses Cycles, we developed our own python exporter to be able to write cycles materials, but it is less than ideal.
What I've seen so far is that the user profile that gets exposure to Cycles in the Gaffer community brings a complementary perspective that I feel benefits Cycles, everything that makes Cycles more accessible from outside independently to Blender I think is just going to push this amazing tool forward in ways that are not evident inside the Blender bubble.

Answering your question more concretely, for us it would be super useful to be able to pull Cycles materials from a USD file to convert the materials to Gaffer (for this I think Alex Fuller proposed a schema ), I know there is software in development right now which uses Cycles and it would also be useful among other things to make bridges or translators to look dev apps, painting apps, etc.

Hi @brecht , we are a 30 people studio that use primarily Blender and Houdini, since a few months back we've been starting to adopt Gaffer, this was a no brainier since it now uses Cycles, we developed our own python exporter to be able to write cycles materials, but it is less than ideal. What I've seen so far is that the user profile that gets exposure to Cycles in the Gaffer community brings a complementary perspective that I feel benefits Cycles, everything that makes Cycles more accessible from outside independently to Blender I think is just going to push this amazing tool forward in ways that are not evident inside the Blender bubble. Answering your question more concretely, for us it would be super useful to be able to pull Cycles materials from a USD file to convert the materials to Gaffer (for this I think Alex Fuller proposed a schema ), I know there is software in development right now which uses Cycles and it would also be useful among other things to make bridges or translators to look dev apps, painting apps, etc.

matt estela commented

2024-03-26 21:14:36 +01:00

First-time contributor

We'd also find this useful at our workplace.

We have some folk using houdini+hdcycles, others using vanilla blender, and others running hdcycles via usd python. We keep getting surprised that although we're all targeting the same renderer (cycles vs hdcycles hair splitting aside), its not easy for us to share materials.

Right now the solution is very manual; get blender and houdini open side by side, and if we manually replicate material nodes one by one we can get something working. I'd planned to write a little script to speed this up, but if this functionality already exists there's no point reinventing the wheel!

For us its also about the being able to use the best tool for the job. Houdini and hdycles is better for procedural shot building. Blender is better for interactive lookdev for artists. Usd and python is better for developers and pipeline/farm focused workflows. Being able to fluidly move cycles materials between those 3 use cases would hugely boost our productivity.

As an aside while materialx sounds great, its still feels at least a year or two away from being production ready, vs moving this feature from the usd branch and into the main blender line would help with production issues right now.

We'd also find this useful at our workplace. We have some folk using houdini+hdcycles, others using vanilla blender, and others running hdcycles via usd python. We keep getting surprised that although we're all targeting the same renderer (cycles vs hdcycles hair splitting aside), its not easy for us to share materials. Right now the solution is very manual; get blender and houdini open side by side, and if we manually replicate material nodes one by one we can get something working. I'd planned to write a little script to speed this up, but if this functionality already exists there's no point reinventing the wheel! For us its also about the being able to use the best tool for the job. Houdini and hdycles is better for procedural shot building. Blender is better for interactive lookdev for artists. Usd and python is better for developers and pipeline/farm focused workflows. Being able to fluidly move cycles materials between those 3 use cases would hugely boost our productivity. As an aside while materialx sounds great, its still feels at least a year or two away from being production ready, vs moving this feature from the usd branch and into the main blender line would help with production issues right now.

Brecht Van Lommel commented

2024-03-28 19:50:36 +01:00

Ok, good to know there are some real world users for this.

However the implementation should be easier to maintain, I don't think we would accept something that is structured like this. Adding a shader node to Blender currently requires changing many files, we are trying to reduce that and this would go in the opposite direction.

Ideally we'd have a situation where conversions Blender -> Cycles, Blender -> USD, USD -> Cycles, , USD -> Blender are just dealing with generic nodes and data types, rather than each needing to be updated whenever a node is added or changed.

We could change the Cycles sockets to better match Blender sockets, properties and enum items (we rather not change the latter for blend file and API compatibility). There can be some generic method to loop over the relevant RNA properties on the Blender node.

When a node needs specialized code should also be handled through a callback on the node type info, see the way this was done for MaterialX.

Ok, good to know there are some real world users for this. However the implementation should be easier to maintain, I don't think we would accept something that is structured like this. Adding a shader node to Blender currently requires changing many files, we are trying to reduce that and this would go in the opposite direction. Ideally we'd have a situation where conversions Blender -> Cycles, Blender -> USD, USD -> Cycles, , USD -> Blender are just dealing with generic nodes and data types, rather than each needing to be updated whenever a node is added or changed. We could change the Cycles sockets to better match Blender sockets, properties and enum items (we rather not change the latter for blend file and API compatibility). There can be some generic method to loop over the relevant RNA properties on the Blender node. When a node needs specialized code should also be handled through a callback on the node type info, see the way this was done for MaterialX.

Esteban Covo commented

2024-04-10 17:46:12 +02:00

First-time contributor

Right now we loop through all the nodes and write them to a USD file and then solve discrepancies with an importer on the other side, it is slow and in very few cases prone to errors but gets the job done.

This pull request has changes conflicting with the target branch.