Sculpting in Houdini, this is not necessary; what is necessary is the expansion of KineFX by means of Character Animation tools etc which we are already seeing. As for FX, I believe SideFX is on that road very well so it knows what it needs to improve on, certainty Houdini doesn't need to become like Blender that is a wrong path to follow.

As with rendering, I have used Mantra; although not in a while I use to be very effective in Mental Ray I decided to make Octane my render of choice over the rest it's the off ramp from the technical back to creative process when it comes to rendering. Therefore I can't speak on Karma other then I've read it still has issues and doesn't seem as polished as Mantra was, that is just an opinion and how I see it from others who use it.

There are plenty of renders that exist that if you feel as if Karma just isn't giving you the results you want, choose another render that is dedicated to rendering and learn it.

Looks like I discovered XPU's Achilles' heel, it isn't able to manage the GPU memory the same as Arnold, I used a scene with around 97 MaterialX materials, 41 of these have 4x 2k textures which were created with Substance Painter and the Dome light has a 16k hdri on it, Karma XPU cannot render the scene with the GPU's, it only uses the CPU in XPU mode, but Arnold has no problem rendering the scene, because of this XPU takes 15 mins where Arnold takes just over a min to render. I really hope you guys can get this sorted in H20 as now I have to go back to Arnold for now.

Are you able to describe the scene to us at all?
eg
- just lots of textures + geo?
- perhaps lots of instancing?
- curve-heavy?

If rendering in the viewport, you can open the render-stats on screen
https://www.sidefx.com/docs/houdini/solaris/karma_xpu.html#howto [www.sidefx.com]

The memory report for the GPU will contain its last known state before running out of memory and can inform us of what it was that blew out the memory.

If texture-heavy, you could try and clamp textures to 1k
export KARMA_XPU_MAX_TEXTURE_RES=1024
https://www.sidefx.com/docs/houdini/solaris/karma_xpu.html#textures [www.sidefx.com]

GDay Brians, The scene is basically a vehicle (Car) with the bonnet up exposing all the engine and running gear, all the details you would expect of a car with the bonnet up are there, there are lots of geometry for the running gear which all contains textures, there are 41 MaterialX materials each containing 4 textures (colour, roughness, metallic and height for the normal), no curves or instancing, if I clamp the textures to 1k then the Dome light hdri is too low resolution to be usable. The viewport is set through karma settings at default (720) using XPU at 5 samples, just for quick feedback, but even with these low settings the render only uses 1 GPU with the CPU and errors the second GPU (The system has 2x rtx2080ti cards). It is because of the number of textures I have on the running gear, if I hide some of the engine geometry the scene will render fine.

Are both GPUs the same spec? I'm curious as to why one would fail but the other work fine.
What do the memory stats look like in the viewport? Are they similar on both GPUs?

The Houdini OpenGL viewport might be filling up one of the GPUs with its own copy of the textures + geo

Hi Brians, yep, both GPU's are exactly the same, I set the environment variable KARMA_XPU_MAX_TEXTURE_RES to 1024 and it renders as you would expect with XPU on both GPU's, but the Dome light HDRI is also set to 1K which renders very blurred and low res.
After a few more test renders, it appears that the Dome HDRI is what needs the resolution set so high, I set the KARMA_XPU_MAX_TEXTURE_RES to 4096 which renders on both GPU's most of the time, but sometimes just on one, the Dome light Environment HDRI is still blurred, but it's not noticeable as the close up shots under bonnet is expected to have an out of focus background.

Perhaps I should exclude domelight textures from any form of resolution clamping?
Or perhaps have a secondary environment variable (for sanity) set to 8k (or something)?

That would be fantastic, a clamp for the Dome light and another for the Materials.

Stephen Davidson
HOUDINI TO ZBRUSH (Much more than GoZ)
An advanced Houdini to ZBrush workflow. leverage Houdini for it's strengths in proceduralism, object and asset management, object versions, posing models with KineFX,and leverage ZBrush for its sculpting tools. A FACS workflow and ability to sculpt animated models. It would make it possible to create a Houdini scene for exporting displacement maps using Houdini instead of ZBrush, which is more production friendly. Houdini would carry everything and only fingers ZBrush's brain for sculpting. (I cant tell you how frustrating ZBrush is in production, particularly when working on assets between artists and particularly in ensuring standards when exporting displacements). Bringing most of that work to Houdini will make it manageable in a studio environment.

No idea if this is doable but it would be nice if creased edges/data from Zbrush could get exported as crease data in Houdini 20.

I really hope Houdinid to get some sculpting and painting ability. Especially sculpting. I understand it'll never catch Zbrush or even Blender in terms of sculpting, but when making corrective blend shapes, importing/exporting between Houdini and Blender is really no solution. That severely limits the use of KineFX to me.

As a game dev, a customizable real-time viewport shading system will be tremendously helpful as well. But I know the chance that Houdini gets it is less than zero.

At least some sculpting, please?

Speaking of H20… how much longer do we have to wait SideFX? :-)

I would love to see those 2 tiny tiny add-ons on UI ^_^

1) Really love the karma/solaris workflow so I hope the Karma CPU and XPU will be released in production mode (for example light & shadows linker working in xpu).
2) efficient COPs

I love the new Karma XPU renderer, but I hate the fact that every time I want to edit the geometry in any way, the viewport switches out of the Karma render view and you have to get back to the Karma render view after you have edited or moved the geometry, I know that you can do the transformations in Solaris, but I like to do old school, but I suppose I could just use the Karma rop instead.

My kingdom for a hotkey in MPlay that mirrors/flips the canvas left and right...

🎡Systems

• APEX COPs Revamp + ML Image Processing nodes + Real-time Compositor Viewport
  
• Develop your own equivalent file format to .SBSAR to generate textures, layer them with mtlx and embed everything into an USD file.
  
• Karma Baker with auto setups values using intersection analysis between the LP and HP and automatic adaptive cages
  
• Start building a more extensive noise library for mtlx, cops and heightfields as we will need them to create materials. Afterwards, start building a smart materials and generators library.
  
• After the aforementioned, lay the foundations for a node-based texture painting workflow that rivals Substance Painter.
  
• APEX Scene Modeling&Sculpting. A good reason for sculpting is that now we are going to get texture synthesis in COPs we can create our own custom brushes and decals. Btw you should be able to use as a brush what you sculpted on a plane, that's a common and pretty straightforward workflow in blender.
  
• Sculpting directly on normal map (Cozy Blanket/Uniform). No high poly meshes, no retopo, no baking, big deal.
  
• Custom COPs brushes support for Path SOP, with some sidefx presets to paint stitches, zippers, fabric and welding seams (SP).
  
• More appealing subd modeling. Idk why but setting the render flag on the subd node while templating the unsubdivided geo isn't comfortable. Maybe an interactive mode in the subd node to edit the "cage" would do the trick.
  
• Specific nodes to wrap and encourage the use of SDF modeling and kitbashing like the Dominik Lange plugin
  
• Procedural NURBS Modeling. Once someone in the world develops such a thing, even AI will have a hard time surpassing it. Furthermore, parametric modeling is less destructive than polygonal, minimizes manual tweaking which is closer to Houdini's idea. A workflow example would be: convert to nurbs > bevel > convert back to poly, this way u would steal the most envied feature of Plasticity (which is the NURBS beveling). Another killer feature to start with would be nurbs lofting and bridge edges for surface modeling.
  
• Pretty standard surface mat shader like the blender one also helps with the modeling and feeling.
  
• Shortcut Concatenation (blender) would be a good idea to reduce friction between the user and the viewport. eg: typing g+x+value(move+axis_constraint+amount) or R+R to rotate with the trackball.
  
• Component sensitive nodes.
  
• More snapping presets for the pie menu would also help improve the modeling mode.
  
• Interactive flag (maybe replacing the not so used frozen flag?). Why? (1) To know a node has an interactive mode and which ones do not. (2) To indicate that you are currently in the interactive mode of that node. (3) To avoid the need to lift your left hand to press Enter or Escape. (4) To feel in control of your process.
  
• Holding alt+mmb while orbiting to snap to different orthographic views (plasticity). Or even better, 'snap to closest orthographic view', which allows you to map all planes to one key (MODO)
  
• Symmetry selections and select similar.
  
• Group aliases. eg: select ring, fill selection, select from relative bbox (from top/mid/bot), select line (which is a select loop but smarter enough to stop at corners)...
  
• Seamless interoperability between solaris and the other contexts (Kill OBJ level).
  
• Seamless interoperability with UE5 through USD and Mtlx (mtlx should translate to substrata definition on import).
  
• Vulkan viewport gold and default.
  
• Vulkan ROP
  
• Further expand the component builder workflow.
  
• Layout node performance improvement.
  
• Pure Mtlx (non-karma specific) skin, ocean, white water, pyro, cloud and toon/anime shaders.
  
• ML ROM to train flesh simulations (Maya).
  
• ML and GPU Multithreading Powered Solvers.
  
• Multisolver SOP.
  
• UI Refresh: more responsive, modernized skin and dynamic tab arrangement.
  
• More Gaea like nodes for the terrain suite. eg: shorelines, craters...(btw Jake Rice has the code for the Aperature node on his git)
  
• Scales tools like the ones from Wout Tengrootenhuysen but leveraging the new feather's guide system and painting on UVs
  
• Ivy tools like the ones from MCworldKit
  
• Clothing tools for mimicking the most common workflows of Marvelous Designer and help with garment creation.
  
• Add a Braid SOP to the grooming suite (blender)
  
• Update Hair Card Generate SOP to work with feathers. Additionally, explore UE Nanite Spline Meshes for nanite feathers.
  
• Continue to bring valuable things hidden in networks to SOP level. Blender users are using VOPs in SOPs without even realizing it. That's the key, make people use it without telling them they are using it. A lot of hidden gems are not used simply because they are buried. An example is to use a simple geometry CHOP you have to create a CHOPnet, a channel node, dive inside, use a Geometry node (specify the SOP path) and then use the actual node (...) 4 nodes and one lvl down to do a noise filter... Therefore, bring geometry CHOPs into SOPs for straightforward use.
  
• Refurbish hidden features to bring them back to life as if they were brand new. Help users rediscover these valuable tools that have been neglected or overlooked.
  
• Get rid of redundant nodes. eg: a shape node gathering all primitives (default prims, platonics, superformula, labs star, labs shapes...)
  
• USD Cache SOP like the hda from Ayman Abolila
  
• Watertight Collision, like the hda from attila sipos. In fact, that's the output you would expect from the collision source node.
  
• Better particle fluid surface node, the result of the actual one is very poor. I always see people coming up with some workaround because the output of this node is not what they expect.
  
• More aliases, including presets for point and group expressions and other wrangle snippets. Also, list them somewhere pls, either in the docs or by pressing tab and typing "aliases". Aliases were a great introduction to avoid the creation and maintenance of new nodes that do not deserve a node by themselves.
  
• Straightforward Lattice node, the current one requires many nodes to make it work and is not that good after all.
  
• A basic Radial Array SOP with both the bend360 with auto capture length and 360/ncopies methods, including an interactive mode for easily placing the pivot center of the radial array leveraging the snapping options (plasticity).
  
• Slide SOP (to slide edges and points).
  
• Edge Flow SOP like the one from Tom Jerry
  
• Limit Bevel (plasticity)
  
• Extend Sheet Trim (Plasticity)
  
• Project Curve Curve (plasticity)
  
• Wrap Face and Unwrap Face SOPs (plasticity)
  
• Sketchup Extrude
  
• Geometry Layout SOP like the stone wall idea from iansmithartist
  
• Art-directable point velocity SOP like the BoomBox and Easy Debris hda from Jason Harmon
  
• Add an option in the boolean node to enable infinite slicing with the ability to choose whether or not to affect backfaces, eliminating the need to thicken the cutting shape beforehand. That way, we can replicate the imprint curve tool from plasticity by setting the operation to 'shatter'.
  
• Expand Match Size capabilities to unrotate and align to components like the 'align to component' shelf tool, right now feels like match size only does half of the job.
  
• Start passing the shelf tools to nodes, there are many attractive functionalities going unnoticed by being there. For example: I don't think most people know that after your grain simulations you have to use "upres sand" to invoke a combo of 4 nodes that does the rest of the work for you.
  
• Multiple outputs split node, so that we can split as many items as we need, instead of being limited to two.
  
• Bring Crowd Solver to SOP lvl.
  
• The trigger node for crowds is fantastic, just like the ones in game engines, thanks to its overridable functions. It's a shame that it is for crowds only. I would like you to bring it to the surface for triggering events in other workflows as well.
  
• Help Feike to push out the APEX Flora tools to match or exceed the quality of Nils Westfelt's Simple Tree Tools and speed tree.
  
• Make a fine selection of the best nodes from third party libraries like Ql, AE, EZ, MOPs, Modeler and make your own take. People hate relying on addons, and that's one of Houdini's strengths, as it sets it apart from the competition, which handles this matter poorly.
  
• APEX Auto Rig Component for reverse foot ik
  
• AI assisted keyframing + help for physics and secondary motion (cascadeur)
  
• UE5 queen and manny APEX rig ready to animate (cascadeur).
  
• Adding a tangentu attr with a polyframe set to Style=MikkT speeds up the bonedeform like 500%. There is also an HDA called "Wrap Deformer" from Momme which leads to better results than the official point deform node.
  

🧰Misc

• The groups widget is kind of annoying. It should be great to visualize groups through the info panel by clicking on the group names (like you do with attrs).
  
• In the info panel, add support to quickly change the attr visualization type (float grayscale, vector, str...) just by pressing rmb over the attr names.
  
• Provide some kind of visual feedback for parameters that differ from their default values. Also add the option to the cogwheel to show only params that have been tweaked. This would help navigate through others files or showcasing breakdowns and tuts.
  
• rmb/assign shortcut... for nodes, options, ui buttons and window tabs (plasticity)
  
• Drag and drop nodes to pie menus to simple replace or assign nodes to empty or occupied slots. Same thing on new empty windows to drag and drop nodes and make your own tool palettes to be used as pop-ups, binding them to a shortcut (rmb/assign shortcut... on its header)
  
• Fix redirections/dependencies script
  
• An option to stop nodes that are taking too long to cook in order to prevent hangs.
  
• Make attrs work directly in params without having to use the point/prim() functions, as they do in the primitive properties node for example, or like the context option editor in Solaris, where global variables can be declared and used in any parameter.
  
• Focus on elements when double-clicking them in the spreadsheet (unreal PCG).
  
• Support drag and drop in the spreadsheet to drag the attr names or values and drop them into any param field
  
• Better xform gizmos.
  
• Add a compute range button to all fit parameters (not all of them has it).
  
• Be able to set the render flag in mtlx and VOP nodes.
  
• Do you know when you have to set the resolution in a pyro/flip parameter and link it to every other resolution parameter? pls, save us from that redundancy somehow, so we can just set it once and pass it through.
  
• More user-friendly vellum params like the Cloth Builder from Daniel Bystedt to help us to work faster reducing the need to obsessively fine-tune parameters within vast decimal ranges.
  
• Add inputs and outputs info to nodes documentation, and make sure to specify which of them are optional.
  
• GPT, Mixtral, Bing or Bard powered documentation, including Houdini Engine Doc in the same site.
  
• Houdini 20.5 Crash Course. A fast adoption guide to all the new features and workflow changes.
  
• Houdini 20.5 Foundations book.
  

🧪R&D

• ML assisted sculpting and painting
  
• Develop USD-HDAs with the ability to contain a procedural mtlx which adapts to geometry changes in real time when used in game engines (may only be achievable in Substrata real-time BSDF).
  
• AI-powered nodes (starting for Auto UV and Capture Biharmonic)... Also similar nodes to Runway or InstaMAT like "color grade from prompt", "color grade from reference", "tracking from target", "automatic subject masking", "image2material", "magnific", "color ramp from prompt", "image to svg", "style transfer", "super-resolution", "inpainting"...
  
• Explore Neural VDBs, Neural Physics and Material Super-resolution.
  
• COPs SDF editor.
  
• Copilot wrangles for vex, python and USD with option for user feedback to refine the model.
  

...
"I never saved anything for the swim back" Gattaca (1997)

nicholasralabate
My kingdom for a hotkey in MPlay that mirrors/flips the canvas left and right...

in 19.5 and onwards 'o' is the hotkey for flipping left and right in mplay

Carlos_Rivadulla
-Revamped Substance Designer like COPs. Packing an MTLX within an USD-HDA has a lot of potential.
-Sculpting basics (you don't need to be the best). I don't care if it's an all-in-one node or a suite of nodes. I'm dying for grid remeshins>sculpting>quadremeshing. Aren't you?
-More appealing direct modeling.There is too much friction between the user and the viewport.
-More appealing subd modeling.
-Kill OBJ level in favor of LOPs.
-Kill Mantra in favor of Karma.
-try to end redshift's hegemony in favor of materialx
-GPU Multithreading pyro solver.
-cascadeur like Physics anim.
-Vulkan viewport.
-Continue bringing valuable things hidden in networks to SOP level. Blender users are using VOPs in SOPs, and they don't even know it. That's the key, make people use it without telling them they are using it. There are a lot of things that are not used simply because they are hidden. Maybe geeks like that, but it's not reasonable, accessible, intuitive, or user-friendly at all. Constraint Relationships for mechanical stuff is a good example of what I'm not using because it's out of scope. Another example is to use a simple geometry CHOP you have to create a CHOPnet, a channel node, dive inside, use a Geometry (specify the path) and then use the actual node (...) 4 nodes and one lvl down to do a noise filter really?.
-Get rid of redundant nodes (clean the house, tidy up your room, throw away the old furniture)
-Add the whitewater nodes to the new FLIP suite.
-A shortcut like R+R in Blender to rotate with the trackball should be great to have. Especially when rotating bones. Typing in the viewport like g+x+value (move+axis_constraint+amount) is also a good idea to reduce friction.
-Symmetry selections and select similar.
-The groups widget is kind of annoying. It should be great to visualize groups through the info panel by clicking on the group name, just like you do with attrs.

Dreaming is free section
-AI-powered nodes (starting with Auto UV as an experimental field). Similar to Runway tools "color grade from prompt", "color grade from reference", "auto-tracking from target". They have 30 one-click tools that could perfectly be Houdini nodes.
-Neural VDBs
-AI Accelerated karma
-Convert to nurbs > bevel > convert back to poly. To leverage the most envied feature of software like Plasticity (which is the NURBS beveling).

Some Nice Words
Don't rush to release this update. Take as much time as you need to make it as impactful, disruptive, loud and groundbreaking as possible, so that it shakes up the industry and claim your throne as industry standard. If you have to delay it 1,2,3 months, fuck people really.
...
"I never saved anything for the swim back" Gattaca (1997)

These are my hopes and dreams too! lol
I hope sidefx is listening, I’m doing a lot of non-procedural modeling in houdini atm and the software almost seems like it’s begging me not to do so! the viewport interaction with modeling tools is a hit and miss situation. The node graph sometimes drops edit nodes in weird places in the node graph and totally f’up the model; it’s just way more frustrating than it needs to be! The object level has always been the worst thing about houdini for me and I truly hope I no longer have to use it, my start up currently defaults to stage context.

The houdini openGL view might just be the worst one I’ve use on any 3D software in terms not updating properly; a lot of times I have to close the viewport panel and open a new one to kick in updates, I don’t know if that’s houdini or GPU drivers.

UV tools need to be better!

I’m really looking forward to H20 whenever it comes out! I too hope SESI takes their time and give us a great update!

traileverse

Carlos_Rivadulla
-Revamped Substance Designer like COPs. Packing an MTLX within an USD-HDA has a lot of potential.
-Sculpting basics (you don't need to be the best). I don't care if it's an all-in-one node or a suite of nodes. I'm dying for grid remeshins>sculpting>quadremeshing. Aren't you?
-More appealing direct modeling.There is too much friction between the user and the viewport.
-More appealing subd modeling.
-Kill OBJ level in favor of LOPs.
-Kill Mantra in favor of Karma.
-try to end redshift's hegemony in favor of materialx
-GPU Multithreading pyro solver.
-cascadeur like Physics anim.
-Vulkan viewport.
-Continue bringing valuable things hidden in networks to SOP level. Blender users are using VOPs in SOPs, and they don't even know it. That's the key, make people use it without telling them they are using it. There are a lot of things that are not used simply because they are hidden. Maybe geeks like that, but it's not reasonable, accessible, intuitive, or user-friendly at all. Constraint Relationships for mechanical stuff is a good example of what I'm not using because it's out of scope. Another example is to use a simple geometry CHOP you have to create a CHOPnet, a channel node, dive inside, use a Geometry (specify the path) and then use the actual node (...) 4 nodes and one lvl down to do a noise filter really?.
-Get rid of redundant nodes (clean the house, tidy up your room, throw away the old furniture)
-Add the whitewater nodes to the new FLIP suite.
-A shortcut like R+R in Blender to rotate with the trackball should be great to have. Especially when rotating bones. Typing in the viewport like g+x+value (move+axis_constraint+amount) is also a good idea to reduce friction.
-Symmetry selections and select similar.
-The groups widget is kind of annoying. It should be great to visualize groups through the info panel by clicking on the group name, just like you do with attrs.

Dreaming is free section
-AI-powered nodes (starting with Auto UV as an experimental field). Similar to Runway tools "color grade from prompt", "color grade from reference", "auto-tracking from target". They have 30 one-click tools that could perfectly be Houdini nodes.
-Neural VDBs
-AI Accelerated karma
-Convert to nurbs > bevel > convert back to poly. To leverage the most envied feature of software like Plasticity (which is the NURBS beveling).

Some Nice Words
Don't rush to release this update. Take as much time as you need to make it as impactful, disruptive, loud and groundbreaking as possible, so that it shakes up the industry and claim your throne as industry standard. If you have to delay it 1,2,3 months, fuck people really.
...
"I never saved anything for the swim back" Gattaca (1997)

These are my hopes and dreams too! lol
I hope sidefx is listening, I’m doing a lot of non-procedural modeling in houdini atm and the software almost seems like it’s begging me not to do so! the viewport interaction with modeling tools is a hit and miss situation. The node graph sometimes drops edit nodes in weird places in the node graph and totally f’up the model; it’s just way more frustrating than it needs to be! The object level has always been the worst thing about houdini for me and I truly hope I no longer have to use it, my start up currently defaults to stage context.

The houdini openGL view might just be the worst one I’ve use on any 3D software in terms not updating properly; a lot of times I have to close the viewport panel and open a new one to kick in updates, I don’t know if that’s houdini or GPU drivers.

UV tools need to be better!

I’m really looking forward to H20 whenever it comes out! I too hope SESI takes their time and give us a great update!

I mainly do modeling and then look dev and I can say that Houdini is one of the best, I started with 3D Studio on to Max then Softimage then most of the available 3D softwares like Blender, Lightwave etc before landing on Maya for many years and now Houdini. In my experience the best was Softimage XSI, Maya inherited most of what XSI had, but Houdini is my favourite, sure it needs a bit more functionality with non procedural modeling, but what is there works well.
True, from time to time the viewport needs to be reset and the zoom sometimes goes out of whack and some of the functionality could be more accessible, but I like it.
If Polydraw had a mirror function built in and more of what Edit has, then I would be like a pig in mud.
T.

Tser
I mainly do modeling and then look dev and I can say that Houdini is one of the best, I started with 3D Studio on to Max then Softimage then most of the available 3D softwares like Blender, Lightwave etc before landing on Maya for many years and now Houdini. In my experience the best was Softimage XSI, Maya inherited most of what XSI had, but Houdini is my favourite, sure it needs a bit more functionality with non procedural modeling, but what is there works well.
True, from time to time the viewport needs to be reset and the zoom sometimes goes out of whack and some of the functionality could be more accessible, but I like it.
If Polydraw had a mirror function built in and more of what Edit has, then I would be like a pig in mud.
T.

Yes! Boosted Polydraw node!

My two cents:
Edit SOP functionality in it.
Inset, extrude.
Smooth selected Edge aka Labs Edge Smooth SOP.
Drop primitives.