Houdini Karma Rendering: Solaris to Frames

This Houdini Karma tutorial takes a scene from Solaris setup to a saved image sequence. You will choose the correct render engine, create render settings, preview the result, reduce noise, and render final frames without guessing which node does what.

Lesson 8 of the Houdini curriculum. Next up: Caching, Versioning, and Production Hygiene.

Updated August 4, 2026 against SideFX’s current Karma documentation, rendering guide, and Karma XPU hardware notes.

What is Karma rendering in Houdini?

Karma is Houdini’s physically based USD renderer. It is integrated with Solaris, Houdini’s scene-building and look-development context, and can render interactively in the viewport or write final frames through a USD Render ROP. For new projects, SideFX recommends Karma instead of Mantra.

The short workflow is: import or build the scene in /stage, add lights and a camera, create a Karma Render Settings LOP, preview with a Karma viewport delegate, then connect a USD Render ROP to save the final image sequence.

Karma CPU vs Karma XPU

Choose the engine before tuning the render because CPU and XPU are not guaranteed to produce pixel-identical output.

On a Mac, Karma CPU is the clearest default. XPU can run there, but it does not use the Apple GPU, so do not expect the NVIDIA acceleration described in many XPU tutorials. Once you choose an engine for a production render, keep the same engine locally and on the render farm.

Step-by-step: render a scene with Karma

1. Bring the scene into Solaris

Switch to the Solaris desktop or open the /stage context. If your geometry was built in /obj, add a Scene Import LOP to bring the object-level scene into USD. Confirm the geometry, camera, and lights appear in the Scene Graph Tree.

2. Add a camera and lights

Create or import a camera and at least one USD light. A simple first setup is a Dome Light for environment illumination plus a Rect Light for direction. Make sure the camera path is the one you intend to render.

3. Add Karma Render Settings

Create a Karma Render Settings LOP after the scene-building nodes. Set the resolution and camera, then choose the CPU or XPU parameter set. For a fast first preview, keep sampling modest; increase it only after the composition and lighting are approved.

4. Preview in the Solaris viewport

Use the viewport’s render-delegate menu to choose Karma CPU or Karma XPU. Let the image resolve, then adjust lights and materials interactively. Save useful comparisons to the Render Gallery instead of judging changes from memory.

5. Set sampling and denoising

For Karma CPU, start around 4×4 Pixel Samples for previews and raise the value for finals only when visible noise remains. Enable denoising for a cleaner preview, but do not use it to hide missing samples, fireflies, or poorly lit areas. Check AOVs to identify whether noise comes from direct light, reflections, refractions, or volumes before increasing every sample control.

6. Add a USD Render ROP

Connect a USD Render ROP after the render-settings node. This writes temporary USD and launches husk to produce image files. Point it at the correct render-settings primitive and set an output path that includes a frame token, for example:

$HIP/render/karma/beauty.$F4.exr

$F4 creates zero-padded frame numbers such as beauty.0001.exr. Without a frame token, an animation can overwrite the same file on every frame.

7. Match and render the frame range

Set the render range to the frames actually cached by your simulation. Render one representative frame first, open the output file, and verify the camera, resolution, color, and AOVs. Then render the full range through the USD Render ROP.

Recommended output settings

• Use OpenEXR for animation, compositing, high dynamic range, or multiple AOVs.
• Use PNG only for simple finished stills that do not need compositing flexibility.
• Put each shot or version in its own output directory to avoid overwrites.
• Render a short frame range before committing a long simulation to final quality.
• Inspect the Log Viewer when husk fails or an expected file is missing.

Why is my Karma render noisy?

Do not raise every setting at once. First inspect the noisy component in the Render Gallery or AOVs. Improve the light setup, then increase the sampling control connected to the noisy path. Denoising is useful for the final residual noise, but excessive denoising can smear texture and fine geometry.

For systematic diagnosis, SideFX’s Refining Karma Renders course covers AOVs, light path expressions, CPU and XPU sampling, and denoisers with current Houdini workflows.

Common Karma rendering problems

The viewport renders, but no file is saved

The viewport delegate is only an interactive preview. Add and execute a USD Render ROP, verify its render-settings path, and set a writable output image path.

Every animation frame overwrites the previous one

Add $F4 or another frame token to the output filename. Test frames 1 and 2 and confirm that two separate files appear before launching the full render.

Karma XPU is not using the GPU on a Mac

That is expected. SideFX currently supports NVIDIA GPUs for XPU acceleration; Apple Silicon falls back to the Embree CPU device. Use Karma CPU for a straightforward Mac workflow, or render XPU on a supported NVIDIA machine.

Materials look different between CPU and XPU

XPU supports MaterialX, USD Preview Surface, and Karma-specific nodes, while legacy shader support is more limited. Use MaterialX for an XPU pipeline and avoid switching delegates mid-production.

Do it hands-free with Skilly

Reading steps is one thing; finding the right Solaris node while Houdini is open is another. Skilly is a voice-first tutor for macOS: ask “where do I add the USD Render ROP?” and get guidance against the Houdini window already on your screen. Start free with no card.

Keep going

This is one stage of the full Houdini beginner curriculum. Continue with Caching, Versioning, and Production Hygiene.