# Render USD scenes with Houdini Husk on Deadline Cloud
<a name="examples-jb-houdini-husk-usd"></a>

The [houdini\_husk\_usd\_render](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/job_bundles/houdini_husk_usd_render) job bundle renders a USD scene using the Houdini [husk](https://www.sidefx.com/docs/houdini/ref/utils/husk.html) CLI. By default, Husk renders with the Houdini Karma renderer, but Husk supports any Hydra-compatible USD render delegate. The bundle expands to a task per frame using the Frames job parameter and limits execution to Linux workers through host requirements.

To render with the `KarmaXPU` GPU engine, uncomment the `amount.worker.gpu` host requirement in the job template so that the job is scheduled on a worker with a GPU. The Karma renderer requires a Karma license to run. Service-managed fleets provide usage-based Karma licenses automatically.

This bundle also shows how to write a custom asset introspection tool for job attachments. The included `generate_usd_job.py` script inspects a USD file, finds all referenced asset files, and generates a job bundle with job attachment references for those assets. To use the script, install the `usd-core` and `deadline` Python libraries and run:

```
python3 generate_usd_job.py {{my_scene.usd}}
```

The bundle also supports V-Ray and Redshift renderers through their Hydra render delegates. To use these renderers, build a conda package with the [houdini-vray-7](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/conda_recipes/houdini-vray-7), [houdini-redshift-2025](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/conda_recipes/houdini-redshift-2025), or [houdini-redshift-2026](https://github.com/aws-deadline/deadline-cloud-samples/tree/mainline/conda_recipes/houdini-redshift-2026) conda recipes.

The bundle includes a `sample.usda` file with a cube and sphere that has no external dependencies and can be rendered without attaching any other files.