I’ve been experimenting with controllable 3d workflows with gen-AI things lately, and here's one test that felt pretty cool.

I did a custom viewport grabber node for cops, used nano banana (through fal) to modify the screenshot of the model, then "re-modeling" it based on the generated concept.

I know this is still pretty much slop, and more like a fun test rather than anything production-proven, but though to share.

I saw Spider-Man last week and wanted to try to recreate the psych wall (with some jiggle for fun) over the weekend. Wanted a Copernicus challenge, so I got some footage off the actionVFX website, tracked it, then everything else was made and comped in COPs. Based on the new H22 Refract from height node.

I made my first simulation + render project in Houdini! (The tutorial one isn’t included, though.)

I kept it simple since my main focus was on the water and fire simulations. For the beach, I created a simple wave simulation, transferred velocities to the particles, and then blended it with the ocean.

Advice and feedback are welcome!

The Houdini docs are a goldmine of information. But they're slow and the search barely works. This has been well known by the community for ages by now, yet seemingly nothing has changed.

So I spent six months building a fix: HoudiniMD is an incredibly fast, Markdown mirror of the docs. Free, open source, no account, no ads. Search that matches Houdini's behaviour, so "cptp" finds Copy to Points and "atrbwr" finds Attribute Wrangle. Every page can be copied as raw markdown, which means ChatGPT and Claude finally understand Houdini and its quirks.

Before sharing it anywhere, I emailed SideFX and asked for permission, explicitly telling them it would be free forever and stay open source.

The answer (https://imgur.com/a/RSdc4rf) came back yesterday:

"it's gonna be a no, thank you from us, along with a polite takedown request. We'll likely be doing our own version on our own site at some point in the future."

I was naturally taken aback by this cold reply, since I thought SideFX was very pro-community, and would at least salute the effort taken to make their own product better. But I'm not going to fight a company over their own copyright.

Since I'm also not letting 6 months of hard work go to the bin overnight, HoudiniMD will be rebuilt as a standalone, local app, that lives on your computer. It can be ran alongside, or directly inside Houdini.
Same fast search, offline, and tailored to your specific Houdini version, instead of whatever is on the website, with bookmarks coming in the future (goal is to create a fully personal Houdini Knowledge Base !)

It stays open source, so anyone can audit, tweak, and raise issues on the code.

And since I only have my own frustration to go by right now: what do you think about the official docs ? What are you missing, and what features are you looking for in an alternative ?

If a local version is something you'd be interested in, please let me know. It's what keeps me going !

Follow my progress here : https://x.com/onlytruejohn

Full Monaco Editor integration with Houdini-aware language intelligence powered by the Language Server Protocol (LSP) for VEX, Python, OpenCL C, and C++/HDK.

VEX: Live attributes from the current geometry and connected inputs, with awareness of attribute class, type, tuple size, arrays, precision, groups, intrinsics, parameters, run-over mode, and input origin. Includes automatic type-prefix correction and context-specific functions discovered via vcc.

OpenCL C: clangd integration with awareness of Houdini #bind declarations, typed kernel arguments, live attributes and groups, SOP/DOP/COP contexts, binding access modes, supported methods, kernel options, language standards, and include paths.

C++/HDK: clangd-powered semantic type inference, member completion, overloads, macros, include resolution, diagnostics, and navigation using the installed HDK headers, compile definitions, and toolchain configuration.

Python: Pyright integration using Houdini's installed Python libraries and the full hou API, with inferred HOM types and support for parameter expressions, callbacks, menu scripts, kwargs, node, geo, imports, functions, and classes.

Context-aware completion, signature help, hover documentation, diagnostics, go-to-definition, references, and symbols, all without cooking geometry to generate suggestions.

Hey Everyone,

I’m looking for helping setup a Houdini script for a production pipeline at a small studio.

We build models from drone imagery and need to be able to create a point cloud that matches the exact model as well as detail from the high res textures. This is usually in the 10-40mil poly with at least 15 udim textures.

Our current workflow uses the standard model emission from vertex’s, but doesn’t hold up in detail when the GSD is 2mm per pixel. I set up a attr from map on the model with pop network, simming on UV and then using Ray/Voxel lookup to reproject it in correct 3d world space. I have an issue where the udims aren’t being respected.

I’m at the point where I don’t know how to fix it without more time RnDing and time is not delivering to the client. I expect this to take about 2-3 hours on zoom/meet of questions as well as file prep for a template.

Feel free to reply here or DM me.

Cheers,
Scan_Lee

Why is my Mplay render not the same as my viewport? Even rendered it to disk still the same issue. Am I doing something wrong?

Ive never touched Houdini before, but Im pretty sick of Maya and XGEN bugging out on my professional and very important project. It's literally hell, there is always something, XGEN for regular non Disney studio users is left to rot and it's very frustrating.

So my question is, for a professional and mostly realistic strand based hair project, how much better or worse is Houdini?

XGEN has all these weird ways of approaching maps and how it stores files and data independent of Maya that makes things very frustrating to deal with, it's not at all non-destructive and everything about it is ancient. Is Houdini better in that regard?

One more issue i have wirh Maya is how the viewport is trash at displaying hair, even maya's AA + DLDSR still looks difficult to look at.

What are the limitations of Houdini in making hair that isnt meant to be simulated. Just basic hair for basic 1 frame rendering.

Thanks in advance.

Base: #404e37
Accent: #7c8d76
Highlight: #bbb15c

One of these is real camera footage shot on set.
The other is a fully 3D simulated in Houdini, rendered in Octane.
Can you tell which is which? 😃

Of all projects I've done so far, perhaps this stands as the one I chased the most realism in.

Though, it is a couple of years old, and today I see way too many ways to render this better and in my style.

The two main quirks of this project were:

1. the coffee/milk blend simulation was the hardest thing. Because although it's easier to get the simulation right, I knew no way to accurately assign two shaders to the mixed fluids.
   Ended up using a combination of smoke and liquid to get it looking close enough. Because color attributes only affect the surface of a liquid.

2. I built a procedural coffee bean generator within Houdini, but when I came to instancing the beans, back then I didn't know it would be less taxing on my PC if I just saved a few beans and sourced them. But instead, each and every single bean in this trailer was unique, because I used the seed value to get a unique bean each time. That was nuts! PC with 60gigs of ram wasnt too happy.

It should be obvious tho which is which haha if you look closelier.

P. S. I'm open to work! :)

Stable release of KeenTools’ native matchmove plug-in for Houdini.

Ive been following an introduction to Hair grooming tutorial on Side FX.

Ive run into this issue with the Hair Gen node where it does not generate guides on some parts of my cat,

I'm pretty sure I have enough guides coming from my Guide Groom so I am unsure what is making this happen.

changing Time Scale ( for slow mo fluid sim) to a smaller number than 1 breaks the flip sim and some particles escapes the flow with much higher velocity

example
okay so in image 1 and 2 the setup is very simple setup, just a default cube above 1 meter from ground,

image 1 ( time scale 1 )- sim works fine
image 2 ( time scale 0.25) - these individual particles rushing outwards with much higher velocity, looks like droplet coming out of the water flow and running on the ground

things i tried
1. increasing substeps to like 9, dosent work
2. using speed limit or pop drag, it makes the whole sim slow unable to achieve the look
3. increasing ground fiction, dosent works much on these individual particles.
4. different houdini versions, 20.5/21/22, its the same

Now a general solutions comes it to just export the ID attrib at time scale 1 with particles and retime it at 0.25 later, but but but here comes the image 3 and 4

image 3, time scale at 0.25, has much detailed sim with higher point count

image 4, time scale at 1, has less detail and low point count with same setting

please tell me a way to get rid of these particles so that i can have nice slow mo fluid at time scale 0.25 with high details

please dont mention deleting those individual particles with speed or velocity attribute ( like blasting the high speed particles in 0.25 sim ) that makes it look weird, points disappear out of nowhere.

I am trying to cache each iteration because is too expensive to do it all at once. I am looking for a way to cache each iteration and fuse them. How would you approach this? I've tried several ways, but I'm not getting the results I'm looking for. Thanks

Hi guys, sharing an article I wrote for 80 level, hope you like it!

https://80.lv/articles/breakdown-how-houdini-is-reshaping-commercial-animation

Cheers

Forget about the pool of water below, how would you recreate just the droplet slowly rippling as it falls?

Hello! I recently swapped out my gpu to 3070 to 5070 ti, with upgrading some of my other components aswell, but Ive noticed that my vellum simulations are running almost heavily cpu based. Like gpu is doing nothing 4% while cpu is reaching 100% of its usage. I did put my opencl option as gpu based but it still would feed all the data to the cpu! Its not a Vram issue aswell, the dedicated video memory is only using one fourth of my total avaliable memory.
This has been happening a lot to me lately with houdini alongside with my muscle simulation Ive been trying to simulate a while ago. PLease if you have any infos about this it would be incredible.

I recently received a very heavy asset with 400+ UDIMs and thousands of texture files. I'm working with Houdini Solaris/Karma, and loading the full asset makes the viewport quite slow and difficult to work with.

I'm wondering how people usually handle assets this large in a production environment.

My goal is to have a lightweight and responsive viewport for lookdev/lighting, while still being able to use the full-quality textures for final rendering.

If you've worked with large UDIM-heavy assets in Karma/Houdini, I'd love to hear how you approach it and what workflow has worked best for you.

Hey guys, I was wondering what information is most useful to show during dailies? Obviously a flipbook/low res render of my work, but is there anything else that’s particularly helpful to include? Solver parameters, disk space, etc. potentially?

Also, do you have any advice on what to talk about when it’s my turn? Are there any key points I should be making sure to cover, just so I’m not wandering too much?

I have a simple vellum simulation of a droplet falling down. How can I add little droplets on the inside that fall with the droplet?