Hi everyone,
I’m Rabia Rahou, founder of Roji Research,
I’m developing Roji, an end-to-end OpenUSD environment that brings scene assembly, look development, lighting, interactive review and rendering into one connected workflow across local workstations and cloud infrastructure.
Roji is built so artists can work end to end in OpenUSD, from look development and lighting through rendering, without relying on external, non-specialized applications or separate render engines.
What Roji brings together
- Graph-based USD scene assembly for organizing assets, layers, variants and production scenes.
- Interactive 3D review for working directly with complex USD stages.
- Shaderverse for look development, materials and shading workflows.
- Lighting and camera workflows inside the same USD environment.
- Roji Render for live and final-frame rendering.
- AI-assisted workflows designed to support artists and technical teams rather than replace their decisions.
Roji XPU
One of our current research areas is a unified CPU and GPU rendering architecture.
Roji XPU does not divide the image into separate CPU and GPU regions. Every available processor contributes samples to the same continuous frame.
We have been validating renders between Embree on the CPU and OptiX on the GPU, comparing geometry, normals, materials, light transport and AOVs to establish one shared BSDF contract across both backends.
The current architecture includes:
- Dynamic CPU-to-multi-GPU sample scheduling
- Unified BSDF behaviour across Embree and OptiX
- Live USD and H.U.G.E geometry parity
- Sparse RTEX texture streaming
- Adaptive interrupt-and-resume rendering
- Production AOVs, custom LPEs and light groups
- Resumable Cryptomatte and Deep data
- Memory safeguards that keep large scenes responsive under RAM or VRAM pressure
Why we are sharing this here
We are preparing the September closed beta and would value feedback from artists, TDs, supervisors and engineers already working with USD in production.
In particular, we are interested in:
-
- Where USD workflows still create friction between assembly, look development, lighting and rendering
- How production teams organize layers, variants, materials and overrides
- Which lighting workflows, lighting huge scene per usd stages, should receive priority
- How large production scenes should be managed and streamed under RAM or VRAM pressure
- Which CPU/GPU parity cases are most important to validate
- Which AOV, LPE, Deep and Cryptomatte workflows should receive priority
- What an end-to-end USD application must support before it can be trusted in production
We are not presenting Roji as a finished answer. The purpose of the beta is to test these ideas against real production experience and determine what needs to change.
You can read more about the project and request access to the closed beta here:
Feedback is welcome even if you are not interested in joining the beta.
Rabia Rahou
Founder, Roji Research
