# Cloud GPU

Source: https://frameleaf.app/docs/cloud/cloud-gpu/

Cloud GPU runs heavy jobs, such as descriptions, enhancement, restoration, captions and voice-overs, on fast remote hardware, so you get results quickly without buying a graphics card. You see what will be sent and what it will cost before each job starts, and you only pay for the work that's done.

Prefer to keep everything at home? Local processing is always there. As long as the blue Cloud option appears in a tool, you can use it.

## Before you start

An administrator sets it up once:

1. Open **Settings**, then **Frameleaf Cloud**, then **Cloud processing**.
2. Enable the Cloud destination.
3. Review the processing consent and choose who may spend.
4. Add credit to the AI Wallet if your account asks for it.

You need a live licence activated on the server that sends the job. A licence on another server doesn't cover it. You don't need a Cloud plan for AI: you pay with AI credit. Buying AI credit also needs a live licence activated on a server linked to your account. If the licence lapses, new purchases and jobs pause; your earlier charges and wallet history stay as they were.

If a particular model can't run in the Cloud, the tool tells you why and you can run it locally instead.

## Choose what leaves your server

Each Cloud job sends only the inputs that task needs, and the dialog shows you exactly what that is before you confirm. Different tools send different things: a description preview, restoration inputs, transcription audio or a voice-over script.

Text recognition and Studio's final export always stay on your network. Smart search and face detection stay at home too, unless an administrator chooses Frameleaf Cloud for them in **Where work runs**, on the **Computers** page. The Cloud then reads a metadata-stripped preview of each photo with the same model this server uses; grouping faces into people, naming them and reading what you search for stay on your server. See [Search and faces on Frameleaf Cloud](https://frameleaf.app/docs/admin/workers/#search-and-faces-on-frameleaf-cloud). Nothing else goes to the Cloud unless you start a Cloud job.

### Each person decides for their own photos

Turning Cloud processing on for the server doesn't decide it for everyone. Each person chooses whether AI work on their own photos may use Frameleaf Cloud, in **Settings**, then **Your preferences**, then **Where your photos are processed**:

- **At home or on Frameleaf Cloud** follows the administrator's choices.
- **Only at home** keeps that person's photos on your server and the computers at home. Work set to Frameleaf Cloud runs at home for their photos instead, or waits, and a tool that asks each time never offers Frameleaf Cloud for them.

The choice travels with the photo: a photo in a shared album, or seen by a partner, follows its owner's choice, whoever looks at it. Administrators can see everyone's choice under **Each person’s photos** on **Computers**, but only that person can change it. Work that already started on Frameleaf Cloud finishes there. See [Where your photos are processed](https://frameleaf.app/docs/using/account-settings/#where-your-photos-are-processed).

Items protected by [Locked encryption](https://frameleaf.app/docs/using/locked-encryption/) never go to the Cloud. A Cloud job prepares its inputs and keeps its results on the server's disk, so it can't take an encrypted item: the tool says **Frameleaf Cloud can't work on an encrypted Locked item; restore it on this server instead**. Run that tool on your own server; it works on encrypted items in memory.

## Run a job

1. Open the description, enhancement, restoration, caption or voice-over tool.
2. Choose a model and the Cloud destination.
3. Check what will be sent, the estimated cost and your spending limit.
4. Confirm the job.
5. Follow progress in **Recent cloud jobs** or **Activity**, then see the result in your library.

The model list and dialog show the current input limits and any quality choices for your release. Start with a small selection, then run the rest once you like the results.

## Captions in Studio

Studio can send one clip's sound to Frameleaf Cloud to be turned into captions, which helps with a long video or an older computer. In the captions tool, choose **Frameleaf Cloud**, then **See the price**; nothing is sent until you confirm. The job shows in **Activity**, and the lines are added to the clip when it finishes. Cloud captions arrive in whole lines, so word-by-word caption styles need **This device** instead. See [Captions on Frameleaf Cloud](https://frameleaf.app/docs/studio/captions/#on-frameleaf-cloud).

## How you pay

When you confirm, the estimate is reserved in your AI Wallet, not spent. You're charged only for the work actually completed, up to the reserved amount and within the terms you accepted. The rest of the reservation comes back to your available balance, and a job with no billable usage releases all of it. Work that finished before a failed item or an interruption is still charged.

Your wallet statement shows the final charge and the amount released for each job. To get unused purchased credit back on your payment card, send a refund request. Supporter discounts don't apply to AI credit, GPU time or start fees.

If a job is held by your balance, consent or spending limit, the dialog tells you which one; fix it and try again. A job only ever runs where you chose to send it.

## Photo locations

**Identify photo locations** is another Cloud job paid from AI credit, though it needs no GPU: it names the café, park or venue where each photo was taken, at $5 per 1,000 photos checked. Only GPS coordinates are sent, never the photos. Start it from **Library care**; see [Name the businesses and venues in your photos](https://frameleaf.app/docs/using/places/#name-the-businesses-and-venues-in-your-photos).

## Automatic descriptions

Turn on **Describe new photos automatically** to have each new upload described in the Cloud without asking each time. It's a separate opt-in with its own budget and consent. Describing your existing photos is a separate backfill. Turn it off at any time to go back to approving each request yourself.

## Review the result

AI descriptions can get things wrong, and restoration can invent detail. Check a few results before applying changes across a large collection. Your originals are always kept, and [Library Care](https://frameleaf.app/docs/using/library-care/) and [backup verification](https://frameleaf.app/docs/admin/backup-and-restore/) keep protecting your library whatever you process.
