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

# Video clipping

> Enable YouTube video clipping on a self-hosted instance

<Snippet file="audience/self-host-only.mdx" />

Video clipping turns a long YouTube video into captioned vertical clips and
draft posts. It is off until every piece below is configured, and the clipping
tools do not appear in the agent or in MCP while it is off.

<Note>
  What clipping does from a user's point of view is covered in
  [Video clipping](/general/clipping). This page is only the setup.
</Note>

## What it needs

Postiz does not download or render video itself. It hands the heavy work to
[postiz-uploader](https://github.com/gitroomhq/postiz-uploader), an open source
media worker you deploy on [RunPod Serverless](https://docs.runpod.io/serverless/overview),
and passes files around as presigned URLs.

| Piece                                        | Used for                                                                                            |
| -------------------------------------------- | --------------------------------------------------------------------------------------------------- |
| Cloudflare R2 storage                        | Presigned upload and download URLs for the worker and the transcriber. Local storage cannot do this |
| A postiz-uploader CPU endpoint               | Ingest jobs: the video's details, captions, audio, and the window of each clip                      |
| A postiz-uploader GPU endpoint               | Clip jobs: cutting, fitting to vertical and burning in the captions                                 |
| An Oxylabs account, configured on the worker | Fetching from YouTube. Postiz never talks to YouTube directly                                       |
| Deepgram                                     | Transcribing videos that have no usable captions, and word timing for the captions                  |
| OpenAI                                       | Picking the clips and writing their titles and post texts                                           |
| Temporal                                     | The clipping runs as a workflow in the orchestrator                                                 |

## Setup

<Steps>
  <Step title="Switch storage to Cloudflare R2">
    Follow [Cloudflare R2](/self-host/configuration/r2) and set
    `STORAGE_PROVIDER="cloudflare"`.
  </Step>

  <Step title="Deploy postiz-uploader twice on RunPod">
    Create two serverless endpoints from the
    [postiz-uploader](https://github.com/gitroomhq/postiz-uploader) image: a CPU
    endpoint for ingest jobs and a GPU endpoint for clip jobs. Its README covers
    the image, the deployment and the worker's own variables.

    On the ingest endpoint, set the worker's `OXYLABS_USERNAME`,
    `OXYLABS_PASSWORD` and `OXYLABS_STORAGE_URL`. Without them every clipping
    fails at the first step.
  </Step>

  <Step title="Create the API keys">
    * A RunPod API key with access to both endpoints.
    * A [Deepgram](https://deepgram.com/) API key.
    * An OpenAI API key, if the instance does not have one already.
  </Step>

  <Step title="Add the variables to Postiz">
    ```env theme={null}
    STORAGE_PROVIDER="cloudflare"
    RUNPOD_API_KEY="your-runpod-api-key"
    RUNPOD_INGEST_ENDPOINT_ID="cpu-endpoint-id"
    RUNPOD_CLIPPER_ENDPOINT_ID="gpu-endpoint-id"
    DEEPGRAM_API_KEY="your-deepgram-api-key"
    OPENAI_API_KEY="your-openai-api-key"
    ```

    All six have to be set. If one is missing, clipping stays off and nothing
    is logged about it.
  </Step>

  <Step title="Restart Postiz">
    <Snippet file="env-restart.mdx" />

    The backend and the orchestrator both read these variables, so they need
    the same values if you run them as separate processes.
  </Step>
</Steps>

## Check that it works

Connect an MCP client and list the tools, or open the Agent section. When
clipping is on, `clippingTool` and `clippingStatusTool` are there. Ask for a
clipping of a short public YouTube video and watch it move through its steps.

## Limits on a self-hosted instance

There are no plans, so no clipping minutes are counted. These caps still apply
and are not configurable:

| Limit            | Value                         |
| ---------------- | ----------------------------- |
| Longest video    | 180 minutes                   |
| Clips per video  | 10                            |
| At the same time | One clipping per organization |
| Per day          | 20 clippings per organization |

A clipping that has not finished after 4 hours is closed as failed the next
time anybody looks at it.

## Costs to expect

Every clipping spends money at four providers: RunPod seconds, Oxylabs
requests, Deepgram minutes (only for videos without usable captions, or captions
without word timing) and OpenAI tokens for one prompt carrying the transcript.

## Troubleshooting

| Symptom                              | Cause                                                                                                                         |
| ------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------- |
| The clipping tools are missing       | One of the six variables is not set, or storage is not `cloudflare`                                                           |
| `Clipping is not available`          | Temporal is not reachable from the backend, or the workflow could not start                                                   |
| Every clipping fails while analysing | The ingest endpoint has no Oxylabs credentials, or the RunPod key cannot reach the endpoint                                   |
| Cropped clips look soft              | The worker caps downloads at `OXYLABS_MAX_HEIGHT` (720 by default). Raise it on the ingest endpoint to let `crop` fetch 1080p |