BUG: BakeMaster doesn't inspect node groups to get data to bake

[KirilStrezikozin]

This bug report is:

• [x] not a duplicate
• [x] fixed
• [x] ~after rebasing: https://github.com/KirilStrezikozin/BakeMaster-Blender-Addon/issues/73#issuecomment-2236074891~ No longer relevant.

Describe the bug soeinfeuerball (Discord) found out that BakeMaster doesn’t look inside node groups. This bug is also related to sovanima (Discord), who mentioned that BakeMaster may not work properly for complex materials. The latter is a separate issue (actually this bug fix may have fixed that one now), this bug report focuses on what soeinfeuerball faced.

To Reproduce Steps to reproduce the behavior:

1. Add an object where materials have a node group plugged directly into material output nodes.
2. Bake.
3. See wrong or missing colors on image textures.

Expected behavior BakeMaster should inspect node groups and not just treat them as a single node.

Screenshots soeinfeuerball's screenshot with the original model on the right and the baked one on the left:

Desktop (please complete the following information):

• OS Windows
• OS Version - Windows 11
• Blender Version 4.0
• BakeMaster Version 2.6.0

Additional context This is about reworking mat_replug

[KirilStrezikozin]

TODOs:

• [x] ~Rewrite relink_materials in utils.py to recursively spin nodes and look inside node groups (later)~ the new mats_replug will replace relink_materials preview feat.
• [x] Rewrite mats_handle_replug to recursively sping nodes and look inside node groups

[KirilStrezikozin]

[KirilStrezikozin]

A spreadsheet with all ShaderNode types. Green cells are the ones BakeMaster replugs for baking and map preview.

[KirilStrezikozin]

makes the full list:

ShaderNodeBsdfAnisotropic
ShaderNodeBsdfDiffuse
ShaderNodeBsdfGlass
ShaderNodeBsdfGlossy
ShaderNodeBsdfHair
ShaderNodeBsdfHairPrincipled
ShaderNodeBsdfPrincipled
ShaderNodeBsdfRefraction
ShaderNodeBsdfToon
ShaderNodeBsdfTranslucent
ShaderNodeBsdfTransparent
ShaderNodeBsdfVelvet
ShaderNodeDisplacement
ShaderNodeEmission
ShaderNodeNormalMap
ShaderNodeVectorDisplacement

[KirilStrezikozin]

bruh pushed to master, not dev-2.6.1. master has a breaking change.

[KirilStrezikozin]

The new replugging algo is now used for non-cycles, non-custom map previews and baking, such as AlbedoM, AlbedoS, Metalness, Material Displacement, Decal Normal etc.

[KirilStrezikozin]

Time to test now.

[KirilStrezikozin]

• [x] TODO: when turning on preview, BakeMaster should somehow store what was the active Material Output. Because when there're many, after Preview is turned off some random one gets active, thus material gets messed up. This has to be added to custom-nodes-mats too.

[KirilStrezikozin]

• [x] TODO: restore_ii replug restore instructions from replug_make_nodes are wrong

[KirilStrezikozin]

this is like 90% done

p.s (after 3 days) more like 50% - the system is awesome but raw

[KirilStrezikozin]

damn ShaderNode.label is 64 characters max. Not even enough for a node and link restore instruction :((

[KirilStrezikozin]

same for ShaderNode.name

[KirilStrezikozin]

I have a feeling that we can change our restore instructions to fit withing the 64 characters. Let's count. A full-blown maximum length instruction is something like the following (Specular map with old existing links):

(header[0], 13 characters) BM_maRestore\n

scenario 1:

-----
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
-----
node
node
node (5x nodes)
node
node
-----
node (1x node)
~link~ (deleted on node delete)
-----

scenario 2:

-----
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
~link~ (deleted on node delete)
-----
node
node
node (5x nodes)
node
node
-----
link (1x link)
-----

Which makes it either 6x nodes or 5x nodes + 1x link to restore (ignoring useless restore links instructions used between created nodes, they are deleted anyway when those nodes are deleted)

[KirilStrezikozin]

BM_matRestore\n can be reduced to BMmaR\n (6 characters). We're left with 57. node remove instruction can be described at least with n! and a node index, \n. link instruction can be described at least with l!, node index, !, output index, !, node index, !, input index, \n.

For an instruction scenarios above, this is either:

1. 24 characters + 6 node indexes
2. 27 + 7 node indexes + 2 socket indexes

Where unknown numbers are up to 33 characters in total for 6 (1) and 30 for 9 (2). Makes it an average 5.5 characters for (1) and 3.33 for (2).

[KirilStrezikozin]

ok resolved it we having a custom str prop in bm_props for restore instructions

[KirilStrezikozin]

• [x] displacement node replug is not ready, what if we didn't replug anything, we should be just black?

[KirilStrezikozin]

• [x] check default values for each unfound socket to replug, like opacity shouldn't be black if shader node doesn't have opacity socket, it should be white.

[KirilStrezikozin]

• [x] normal map doesn't want to get replugged using l_socket from node group input node

[KirilStrezikozin]

imo a good name for this thing I'm developing is: map-material preview-replugging

[KirilStrezikozin]

btw, I get sockets by doing shadernode.outputs.get(key_str, default_any), where the returned nodesocket has identifier prop like Socket_5. Just sayin in case you somehow need to retrieve nodesockets index only having its reference. Ahh well just use ...outputs.find(key_str) - this returns an index

[KirilStrezikozin]

• [x] If we catch NodeGroupInput node, we should go to parent node_tree and continue searching from the nodegroupinput's input of index we stopped at inside. ~This is applicable only to displacement and normal maps.~

[KirilStrezikozin]

• [x] check line 4694 (seems displacement preview doesn't worky worky)

[KirilStrezikozin]

kromantiking (Discord) reported this too, AlbedoM comes out black when material setup uses nested node groups. I might need the reference below for future testing of a solution for this issue:

(kromantiking from Discord) It is based on this material from Blenderkit: https://www.blenderkit.com/get-blenderkit/44c1d3c0-c8a5-459f-b86b-f779c5238963/

[KirilStrezikozin]

• [x] #74 found that when baking Emit, image transparency is discarded. This should be checked here.

[KirilStrezikozin]

• [x] em_strength is used for Emission maps, but value 0 is set when emission strength input socket has a link. Solution: allow to pass em_strength_from_socket.

[KirilStrezikozin]

• [x] Emission nodes are muted for non-emission map preview and replug is not done (don't remember the reason, found in first-try-written https://github.com/KirilStrezikozin/BakeMaster-Blender-Addon-Dev/commit/ebd8b7dc915ec0c4249243f577cd5b707e9c8395).

[KirilStrezikozin]

• [x] Initially muted nodes are replugged anyway.

[KirilStrezikozin]

• [x] Preview only active Material Output, because whenever a nested node group is encountered and a replug is done there, it overwrites parent material outputs, so inactive material output at the highest level might contain the final preview.

[KirilStrezikozin]

• [x] Top-level inspection for material output nodes in material node trees should inspect node groups if found.

[KirilStrezikozin]

• [x] When two node groups are plugged into a mix shader node, the socket preview from the first node group might be the final output, ignoring the second node group. This is a possible limitation of the currently implemented preview technique. A fix will probably be about modifying node group outputs to pass the required replugged sockets up the hierarchy in the node tree (this is for Normal and Displacement maps that have their data from nested node groups)

[KirilStrezikozin]

• [x] When Material Output node's Surface socket has no links, a preview with the default blank emission color is not added. [all maps]

[KirilStrezikozin]

• [x] ~Check if there is a way to skip already linked nodes after or before restore.~ hard because this will break my restore function's universal approach.

[KirilStrezikozin]

• [x] Reroute nodes are not supported by replug. AssertionError below:

[KirilStrezikozin]

• [x] Default blank map preview color is expected on muted links with no valid preview data endpoint. The solution involves recursive calls to valid and not muted (from 2.93 onward) links.

[KirilStrezikozin]

• [x] ~Use ShaderNodeTree.get_output_node(sc_render_engine) instead of a for loop to retrieve active material output nodes.~ It doesn't matter, because we still need a for loop to find unconnected nodegroups.

[KirilStrezikozin]

• [x] Ensure that all possible scenarios when preview should be blank work.

[KirilStrezikozin]

• [x] TODO (probably for the future): how big a NoteTree my recursive algorithm can handle to refactor it into an iterative algorithm? stack limit in blender is set to 2000. The current deepest stack travel if the required node is right after material output is 10. For each next node in the current node tree, add +2, in a ShaderNodeGroup, add +1 with no subsequent reduction in depth. So even for a 1000 nodes after material output (and it will make to a 2000 recursive calls only if these nodes in between material output and the ShaderNodeShader are MixShader, AddShader, Reroutes, NodeGroups, which is baffling for a general usage of blender shaders.

[KirilStrezikozin]

• [x] TODO: add a bm_prop toggle to tell this to inspect node outputs for sockets that look like map names and prioritize them.

[KirilStrezikozin]

Map preview stability check:

• [x] AlbedoM
• [x] Metalness
• [x] Roughness
• [x] AlbedoS
• [x] Specular
• [x] Glossiness
• [x] Opacity
• [x] Emission
• [x] Normal
• [x] Displacement
• [x] BSDF Pass

[KirilStrezikozin]

• [ ] TODO: alter default values for different shader node types to return more corrected ones. For example, Glossy BSDF can return default metalness as (1, 1, 1, 1) instead of (0, 0, 0, 1) since it is a BSDF for metallic surfaces. Similarly, Glass BSDF can return default transmission as (1, 1, 1, 1) instead of (0, 0, 0, 1) because it is a BSDF with a transmissive layer (both reflects and transmits light). This has to be tested before implementing actually to be true. This could allow replicating complex mixed BSDF Shaders of different natures into one PBR-compatible Shader like Principled BSDF.

  Corrections:

  1. Glossy BSDF -> metalness is 1.0;
  2. ...

  Progress:

  • [x] Glass BSDF via Principled BSDF;
  • [ ] Diffuse BSDF via Principled BSDF, impossible due to light distribution difference?
  • [x] Glossy BSDF via Principled BSDF;
  • [ ] Transculent BSDF;
  • [ ] Transparent BSDF;
  • [ ] ...

  Useful links:

  1. node_principled_bsdf.osl

  DO THIS IN THE FUTURE.

[KirilStrezikozin]

• [x] TODO: test & fix named outputs
• [x] TODO: use map preview-replug in baking
• [x] TODO: check this: Replace Emmision node replugs with Diffuse + Mix Multiply (strength).
• [x] TODO: replace prints() with debug() and a global DEBUG bool var.
• [x] TODO: test Normal replug for PASS map.

[KirilStrezikozin]

• [x] BUG: wrong colors for sRGB bakes #74 found that when baking Emit, image transparency is discarded. This should be checked here.

Using Combined bake type + Emit pass filter in the latest commit. This preserves transparency.

[KirilStrezikozin]

socket_replug(...) targets with Diffuse BSDF nodes are tracked up until this moment. Next commits mentioned here will not include these alternative targets (reason - unused).

[KirilStrezikozin]

• [x] TODO: Displacement and Normal (exact targets: DISPLACEMENT, NORMAL_POST) maps do not capture ems_socket & ems_value properly and when these values are evaluated in a different node tree than where Emission node is placed (fix: use mix node and multiply data by node strength).
• [x] ~TODO: when rebasing 2.6.3*, collapse custom nodes according to the bm_props property.~ 'collapse nodes' collapsed nodes in baked materials, this doesn't affect previews.

[KirilStrezikozin]

• [x] Doubled strength is applied for Disp mat at least: on math node and on post node.
• [x] Remove initial displacement links for Disp replug to have a clear view on the displacement texture.