The worker image

What it is

When you choose to rent a GPU, Likenor creates an endpoint in your own RunPod account and names the container image it should run. That image is the worker image. It holds the Likenor worker, the Python that prepares photographs and renders scenes, and the runtime those need. The image is a few gigabytes.

It is published on GitHub's container registry as a public package named likenor-worker. Each tag names the commit whose worker source it carries.

Why it is public

Your RunPod account pulls the image, not Likenor's. A public image needs no shared secret and no registry credential of Likenor's inside your account.

The source code is private. The image is not. Anyone can pull it and list what is inside.

What it does not contain

No model weights. Every build is checked by name against the list of model files and refused if one is inside. The models a run needs, about 60 GB, are fetched at the start of the job, onto the machine you rented.

No photographs and nothing of yours. The image is the same for every owner.

No RunPod key and no account credential. Your RunPod keys stay in a protected store on this computer: the operating system's own protection on Windows and macOS, an encrypted file on Linux.

Who can run a job on it

Anyone with a RunPod account can start the image. That yields nothing. The worker refuses a job that carries no Likenor pass. The models are not in the image either.

Where your photographs go

Your photographs stay on this computer unless you choose to rent a GPU, and a rented run leaves this computer for a third-party provider Likenor does not own.

When you rent a GPU, the photographs you send go to the GPU you rented, in your own RunPod account, and nowhere else. Likenor holds no key to your account and receives no copy.

You pay RunPod directly for the GPU. Likenor charges for the software and never for the computing that runs it. See Terms and Privacy.