# Cloud GPUs (RunPod, Vast, Docker)

> Cloud licenses are not sold right now. This will change in the future.

You can run Jasna on a rented GPU. Use the Linux NVIDIA version from the
command line inside a supported container. Cloud activation is CLI-only in the
official Linux release; Windows and GitHub source checkouts do not support it.

## What you need

- **Cloud** has to be turned on for your License code. Ask where you got it.
- A normal container. RunPod and Vast work. A Docker container started with
  `--pid=host` doesn't.
- An NVIDIA GPU with driver **580 or newer**.
- At least **80 GB** of disk.
- `NVIDIA_DRIVER_CAPABILITIES=all` in your template or instance settings.
  Jasna can't read or save videos without it. On Vast, add
  `-e NVIDIA_DRIVER_CAPABILITIES=all` to the Docker options.
- Log in as `root`. RunPod and Vast do this by default.
- A fixed `/etc/machine-id`. Some containers (for example Modal or some Docker
  images) don't have one, and activation fails. Add one to your image.

## Install

Paste these lines into the terminal of your rented machine. If you use a
newer Jasna, change `0.11.1` to your version.

```bash
apt-get update && apt-get install -y zstd wget libgl1 libglib2.0-0

cd /root
for part in 000 001 002 003; do wget https://github.com/Kruk2/jasna/releases/download/v0.11.1/jasna-linux-nvidia-0.11.1.tar.zst.part$part; done

cat jasna-linux-nvidia-0.11.1.tar.zst.part* | zstd -dc | tar -xpf - --same-owner
cd jasna-linux-nvidia-0.11.1
```

Unpack in `/root` like above, as the same user that runs Jasna. Keep Jasna on
the machine's own disk (ext4, Btrfs, XFS, ZFS or the container's own
filesystem). Jasna won't start from `/workspace` or a network drive, and LTX
may refuse to load from a network share or network volume.

## Check the video encoder

Some rented machines have a GPU whose video encoder doesn't work. Check it
before you activate; it takes a second:

```bash
./tools/ffmpeg -loglevel error -f lavfi -i testsrc2=duration=1:size=1280x720 -c:v hevc_nvenc -f null - && echo OK
```

No `OK`? Delete the machine and rent another one. Jasna can't save videos
there.

## Activate

```bash
export XDG_CONFIG_HOME=/root/jasna-config
./jasna --license-code YOUR-LICENSE-CODE --cloud-workspace
```

After `installation verified` it can look stuck for up to a minute. Just
wait, don't press Ctrl+C.

RunPod's `/workspace` can ignore private file permissions. Keep activation in
`/root/jasna-config` as above. On other providers, you can use a persistent
folder that preserves private permissions.

Run the `export` line every time before starting Jasna. Before deleting the
Pod, download `jasna-config` privately to your PC. Restore it under `/root` on
the next Pod, keeping the folder private. This reuses the same activation slot.
If you lose it, you need a free slot to activate again.

## Restore a video

```bash
export XDG_CONFIG_HOME=/root/jasna-config
./jasna --input /root/video.mp4 --output /root/video-restored.mp4 --restoration-model-name ltx
```

The first time, Jasna asks to download the LTX model (15 GB). Type `y`.

On an RTX 50-series GPU, add `--ltx-fast` to restore about 1.4× faster
(it downloads about 8 GB more).

To send a video up and get the result back, use your provider's file
manager, or run this on your PC:

```bash
scp -P PORT video.mp4 root@HOST:/root/
scp -P PORT root@HOST:/root/video-restored.mp4 .
```

## Problems

| You see | Do this |
| ------- | ------- |
| `unsigned or modified Jasna runtime` | Unpack again in `/root` as `root`. |
| `libGL.so.1: cannot open shared object file` | Run the `apt-get` line from Install. |
| `Cloud workspace activation requires a supported Linux container` | Use the latest Linux release, a container without `--pid=host`, and config under `/root` on RunPod. |
| `Failed to open hevc encoder` | This machine's video encoder is broken. Rent another one. |
| `enable cloud activation` | Ask where you got your License code to turn on cloud. |
| `No activation found` or `Cloud workspace identity is missing` | Run the `export` line, then start Jasna again. |
| `unsafe Jasna launch environment` | Start it with `env -u PYTHONPATH -u PYTHONHOME -u LD_PRELOAD ./jasna ...` |

`Bit depth 10 ... not supported` is OK. Jasna reads that video a bit
slower, that's all.

Don't share your `jasna-config` folder. Anyone who has it can use your
activation.