& VFX TOOLS

Production-born AI research published openly for the VFX and animation industry.

Our research starts on the production floor, inside real visual effects work, real animation pipelines, and real immersive builds at the highest level. When we hit a problem the industry hasn't solved, we solve it, and every method, every tool, and every finding gets published openly, because the field advances when serious practitioners share serious work.

04
Papers published
06
OSS tools
Production ResearchOpen Source ToolsMethod DevelopmentField PublicationLab ExperimentsPublic Documentation

how the work gets made

Our research does not begin with a paper. It begins with a shot that will not cook in time, a scene graph that will not hold the data, or a pipeline step an artist has run by hand four hundred times. When we hit a problem the industry has not solved, we stop and solve it properly, then put the result back into the productions that raised it.

Crucible came out of exactly that. It is an open-source Hydra scene index plugin that cooks Houdini HDAs lazily at Karma render time: author a CrucibleProcedural prim in USD and the referenced HDA is evaluated on Houdini's main thread with its output mesh injected straight into the Hydra scene, with no pre-cooking, no LOPs recook, and no HAPI session to manage. houdini_usd_gsplat came from the same floor, bringing 3D Gaussian Splatting into Houdini Solaris natively through a first-class USD schema, a UsdImaging adapter, and a CPU Hydra render delegate, so splats behave as proper USD primitives inside normal LOP workflows.

The method is deliberately unglamorous. Identify the problem inside live production, run controlled experiments against it, put the result through peer review, then release the tool and the method openly. We publish because the field advances when practitioners share real work, and because open research is the only honest proof that the AI and pipeline technology we claim to build in-house actually exists. You can read the code.

PyTorch + JAXHoudini + SolarisUSD / Hydra / KarmaCUDA kernelsPublic GitHub repos

what we believe

WE RESEARCH WHAT WE CANNOT FIND ELSEWHERE.

WE BUILD WHAT THE INDUSTRY HASN'T MADE YET.

WE PUBLISH EVERYTHING.

THE FIELD ADVANCES WHEN THE WORK IS VISIBLE.

how we work

01

PROBLEM IDENTIFICATION INSIDE LIVE PRODUCTION

02

CONTROLLED RESEARCH AND EXPERIMENTATION

03

PEER REVIEW AND REFINEMENT

04

OPEN-SOURCE RELEASE AND PUBLICATION

who this is for

The lab is for people who read the source before they trust the claim.

  • VFX supervisors and pipeline TDs evaluating USD, Hydra, and Houdini approaches who want working code rather than a conference slide.
  • Studios and technical directors looking to adopt Gaussian splatting or procedural USD workflows without building the integration layer themselves.
  • Researchers and technical peers in the USD, Solaris, and real-time graphics community who want methods documented well enough to reproduce.
  • Producers assessing whether a studio's AI capability is real, and what it means for schedule and cost on their project.

the outcome

What every project in this discipline delivers.

  • Research publications
  • Open-source tools
  • Lab notes and method documentation
  • Production-ready implementations