> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voyager.so/llms.txt
> Use this file to discover all available pages before exploring further.

# Render with Vork on Voyager cloud GPUs

> Run Vork renders on a Voyager cloud GPU: which GPUs work, vork setup, the push-render-pull loop, render profiles, long renders, and a copy-paste prompt for your agent.

[Vork](https://github.com/voyager-so/vork) renders scenes on a ray-tracing GPU. Voyager runs Vork on its cloud GPUs, so you or your agent can render from any computer, including one without a GPU.

> [!NOTE]
>
> **Before you start:** a paid Voyager workspace (Starter or above) and the `voyager-cloud` CLI ([quickstart](https://docs.voyager.so/api/machines/quickstart.md)). The machine bills by the minute until you release it.

## Which GPUs run Vork

Vork needs ray tracing.

| Preset | GPU | Use it for |
| - | - | - |
| `render` (default) | RTX PRO 6000 | Final renders |
| `render-lite` | L4 | Drafts and previews, for less |

The `compute` preset (H100) and A100 GPUs have no ray-tracing cores and **can't run Vork**. Use them for CUDA compute, not renders.

## Give this to your agent

Paste this into Claude Code, Codex or any coding agent. The same text is on [voyager.so/vork/cloud](https://voyager.so/vork/cloud) and in `voyager-cloud help vork`.

```text theme={"system"}
Render with Vork (https://github.com/voyager-so/vork) on Voyager cloud GPUs.
Needs a paid Voyager workspace (Starter or above). If access is refused, show me the billing link from the error.
1. Install the CLI (if voyager-cloud isn't installed yet):
   macOS / Linux:  curl -fsSL https://voyager.so/cloud/install.sh | sh
   Windows (PowerShell):  irm https://voyager.so/cloud/install.ps1 | iex
2. Sign in: voyager-cloud login (opens my browser). If VOYAGER_API_KEY is set, use it instead. Never ask me to paste a key.
3. Read the guide: voyager-cloud help vork
4. Lease one GPU and wait until it is ready: voyager-cloud lease render --wait --json  (render-lite is cheaper for drafts)
   Keep the machine id. Reuse this one machine for the whole session; don't lease one per render.
5. Once per machine: voyager-cloud exec <machine> -- vork setup   (up to a few minutes with no output; it is not hung)
   Then voyager-cloud exec <machine> -- vork help   for the scene format.
6. Loop: edit the scene locally → voyager-cloud push <machine> scene/ →
   voyager-cloud exec <machine> -- vork render scene/shot.json --profile std/previs --out out/preview.png
   (exec waits and prints the output; a non-zero exit means the render failed) →
   voyager-cloud pull <machine> out/ → look at the image → revise.
   Use std/previs or std/draft while iterating; std/final (or std/path) only for the final render.
   Long renders: voyager-cloud exec --detach <machine> -- vork render … prints a job id;
   then voyager-cloud logs <job> --follow and voyager-cloud wait <job> before pulling.
7. When done: pull the final outputs, then voyager-cloud release <machine>. It bills until it is released or paused
   (idle machines pause by themselves after 15 minutes). Tell me the cost (release prints it).
Using MCP instead? Add https://mcp.voyager.so/mcp/machines to your agent; it has the same steps as tools.
```

## Do it yourself

**1. Lease a GPU**

```sh theme={"system"}
voyager-cloud lease render --name vork --wait
```

Use one machine for the whole session. Leasing a new one for each render repeats the setup and costs more.

**2. Set up Vork, once per machine**

```sh theme={"system"}
voyager-cloud exec vork -- vork setup
voyager-cloud exec vork -- vork help
```

`vork setup` can take a few minutes with no output; it isn't stuck. `vork help` describes the scene format. To skip setup next time, save a [snapshot](https://docs.voyager.so/api/machines/machines.md#snapshots) afterwards and lease from it.

**3. Render, look, revise**

```sh theme={"system"}
voyager-cloud push vork scene/
voyager-cloud exec vork -- vork render scene/shot.json --profile std/previs --out out/preview.png
voyager-cloud pull vork out/
```

`exec` waits and prints Vork's output. A non-zero exit means the render failed; the output says why. Edit the scene on your computer and repeat. `push` sends only what changed.

**4. Final render and release**

```sh theme={"system"}
voyager-cloud exec vork -- vork render scene/shot.json --profile std/final --out out/final.png
voyager-cloud pull vork out/
voyager-cloud release vork
```

Release prints what the session cost.

## Render profiles

| Profile | Use it for |
| - | - |
| `std/previs` | Fast checks of framing, layout and motion while you iterate |
| `std/draft` | Better-looking drafts, still quick |
| `std/final` | The final render |
| `std/path` | An alternative final-quality profile; `vork help` explains when to pick it |

Iterate on `std/previs` or `std/draft`, on `render-lite` if you like, and switch to `std/final` or `std/path` on `render` only for the last render.

## Long renders

```sh theme={"system"}
job=$(voyager-cloud exec --detach vork -- vork render scene/shot.json --profile std/final --out out/final.png)
voyager-cloud logs $job --follow
voyager-cloud wait $job
voyager-cloud pull vork out/
```

A running render keeps the machine awake. When nothing is running, it pauses itself after 15 minutes.

## If something went wrong

> [accordion] Vork fails to start on the machine
>
> Check the preset with `voyager-cloud status <machine>`. If it is `compute` (H100) or another GPU without ray tracing, release it and lease `render` or `render-lite`.

> [accordion] `vork setup` prints nothing for minutes
>
> That's normal on a new machine. Let it finish. If it fails, run it again and read its error output.

> [accordion] A render exits with an error
>
> `exec` exits with Vork's exit code and prints its error. Fix the scene and render again. For a detached render, read the error with `voyager-cloud logs <job>`.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.