heygen-com--hyperframes/docs/guides/deploy.mdx

---
title: Deploy
description: "Run a Hyperframes preview + render API in the cloud from an official template."
---

Hyperframes ships three official deployment templates that wrap a composition in a small web app: an in-browser preview and a `/api/render` endpoint that produces an MP4 server-side. All are open source, Apache-2.0 — Vercel and Cloudflare deploy from a single button, Modal from a single `modal deploy`.

| Template | Compute | Storage | Deploy |
|----------|---------|---------|--------|
| [Vercel](https://github.com/heygen-com/hyperframes-vercel-template) | [Vercel Sandbox](https://vercel.com/docs/vercel-sandbox) (Firecracker microVM) | [Vercel Blob](https://vercel.com/docs/vercel-blob) | [vercel.com/templates/ai/hyperframes-on-vercel](https://vercel.com/templates/ai/hyperframes-on-vercel) |
| [Cloudflare](https://github.com/heygen-com/hyperframes-cloudflare-template) | [Cloudflare Containers](https://developers.cloudflare.com/containers/) (Workers + Durable Object) | [R2](https://developers.cloudflare.com/r2/) | One-click button in the repo README |
| [Modal](https://github.com/heygen-com/hyperframes-modal-template) | [Modal Functions](https://modal.com/docs/guide/webhooks) (serverless containers, scale-to-zero) | [Modal Volume](https://modal.com/docs/guide/volumes) | `modal deploy src/app.py` |

All three templates use the same shape:

- **Preview** the bundled `ui-3d-reveal` composition in the browser via the [`<hyperframes-player>`](/packages/player) web component.
- **Render** to MP4 by POSTing to `/api/render`. The handler ships the composition to a sandboxed runtime that has Chromium, FFmpeg, and `hyperframes` pre-installed, then streams the MP4 back to storage and returns a URL.
- **Author locally**, deploy the preview + render API. Compositions are still built on your machine with `npx hyperframes init`, then dropped into the template's compositions directory.

## Choosing a template

<Tabs>
  <Tab title="Vercel">
    Pick this if you already deploy on Vercel, want zero-config Blob storage, or want to reuse Vercel's CI/preview environments.

    [![Deploy with Vercel](https://vercel.com/button)](https://vercel.com/new/clone?repository-url=https%3A%2F%2Fgithub.com%2Fheygen-com%2Fhyperframes-vercel-template&stores=%5B%7B%22type%22%3A%22blob%22%2C%22access%22%3A%22public%22%7D%5D)

    **What you get**

    - A Next.js app with `<hyperframes-player>` preview and a `POST /api/render` route.
    - A pre-baked Vercel Sandbox snapshot built during `next build` — cold renders skip the Chromium/FFmpeg install and restore from snapshot in ~100 ms.
    - A Vercel Blob store provisioned automatically on deploy. `BLOB_READ_WRITE_TOKEN` is injected for you.

    **Performance**

    Renders run on `standard-4` (4 vCPU). With `--workers auto`, three parallel Chrome workers cut render time meaningfully vs. single-worker. Concrete render time depends on composition length, complexity, and asset size.

    **Pricing**

    Vercel Pro plans include Sandbox credit each month. See [Vercel Sandbox pricing](https://vercel.com/docs/vercel-sandbox/pricing) for current per-vCPU and per-GB rates and the up-to-date credit allowance.

    <Note>
      Vercel Functions cap at 300s and a 50 MB compressed bundle, which can't fit Chromium + FFmpeg. The template uses Vercel Sandbox specifically because it's the purpose-built primitive for this workload — up to 5 hours of runtime and up to 8 vCPUs per render.
    </Note>
  </Tab>
  <Tab title="Cloudflare">
    Pick this if you're already on Cloudflare Workers, want R2's free egress, or want full control over the renderer image.

    [![Deploy to Cloudflare](https://deploy.workers.cloudflare.com/button)](https://deploy.workers.cloudflare.com/?url=https://github.com/heygen-com/hyperframes-cloudflare-template)

    **What you get**

    - A Worker that serves preview HTML and forwards `/api/render` requests to a `RenderContainer` Durable Object.
    - A pre-built OCI container image with Chromium + FFmpeg + `hyperframes` baked in — no install at request time.
    - An R2 bucket (`hyperframes-renders`) provisioned automatically on deploy.

    **Performance**

    Renders run on `standard-4` (4 vCPU, 12 GiB). With `--workers auto`, three parallel Chrome workers cut render time meaningfully vs. single-worker. Container instances sleep after 10 minutes of inactivity, so the next request after a quiet period pays a cold-start penalty.

    **Pricing**

    Cloudflare Containers bills per-10ms for memory, CPU, and disk; R2 storage has no egress within Cloudflare's network. Requires a [Workers Paid](https://developers.cloudflare.com/workers/platform/pricing/) plan. See [Cloudflare Containers pricing](https://developers.cloudflare.com/containers/pricing/) and [R2 pricing](https://developers.cloudflare.com/r2/pricing/) for current rates.

    <Note>
      Cloudflare's hosted [Browser Rendering](https://developers.cloudflare.com/browser-rendering/) API can't install FFmpeg — that's why the template uses Cloudflare Containers, which gives you a real OCI container in a Worker-bound Durable Object with up to 4 vCPUs and 12 GiB RAM.
    </Note>
  </Tab>
  <Tab title="Modal">
    Pick this if you already run on Modal, want containers that scale to zero with per-second billing, or want the render to run as a spawned function decoupled from the request.

    Modal deploys from the CLI rather than a button:

    ```bash Terminal
    uv sync
    source .venv/bin/activate
    modal setup                 # one-time auth
    modal deploy src/app.py
    ```

    **What you get**

    - A FastAPI app (`@modal.asgi_app`) that serves the `<hyperframes-player>` preview and a `POST /api/render` route.
    - A Modal image with Node.js 22, Python 3.12, FFmpeg, `hyperframes`, and `chrome-headless-shell` baked in at build time — no install at request time.
    - A Modal Volume (`hyperframes-renders`) that persists the MP4, served back from the `/renders/:name` endpoint.

    **Performance**

    First deploy takes ~2 min to build the image; subsequent deploys are ~2s. Renders cold-boot in ~1s and containers scale to zero when idle. With `--workers auto`, three parallel Chrome workers cut render time (GPU is disabled via `--no-browser-gpu`). A 12s 1080p30 composition renders in roughly 10–90s depending on complexity.

    **Pricing**

    You pay per second of render time and nothing while idle, since containers scale to zero. See [Modal pricing](https://modal.com/pricing) for current per-CPU and per-GB rates.

    <Note>
      Modal web endpoints cap at 150s per request, so `POST /api/render` returns a `call_id` immediately and the render runs in a spawned Modal Function (up to 15 min, `timeout=900`). Poll `GET /api/render/:id` — it returns `202` while the render is still running, then the finished MP4.
    </Note>
  </Tab>
</Tabs>

## Architecture

All three templates follow the same flow: the browser plays a preview locally, then POSTs to a render endpoint that delegates to a sandboxed runtime with Chromium + FFmpeg.

```
 Browser                    Edge / Function              Sandboxed renderer
┌──────────────────┐       ┌────────────────────┐       ┌──────────────────────────┐
│ <hyperframes-    │ ────▶ │ /api/render        │ ────▶ │ hyperframes render       │
│  player>         │       │  ship composition  │       │  (Chromium + FFmpeg,     │
│ preview iframe   │       │  → renderer        │       │   pre-installed)         │
│                  │ ◀──── │  ← stream MP4      │ ◀──── │                          │
│                  │  url  │  → upload to blob  │  mp4  │                          │
└──────────────────┘       └────────────────────┘       └──────────────────────────┘
                                    │
                                    └─▶ Vercel Blob / Cloudflare R2 / Modal Volume
```

The key cost-saver in all three templates is **pre-baking the renderer**. Installing Chromium system libraries plus `chrome-headless-shell` takes 30–60s, which would dominate every cold render. Vercel's template snapshots the sandbox at build time; Cloudflare's and Modal's templates bake everything into the container image. All restore in milliseconds and let you spend the entire request budget on actual rendering.

## Swapping the composition

All three templates ship with one bundled composition (`ui-3d-reveal`). To use your own:

<Steps>
  <Step title="Author locally">
    Compositions are HTML — author them on your machine with the [CLI](/packages/cli):

    ```bash Terminal
    npx hyperframes init my-video
    cd my-video
    npx hyperframes preview
    ```
  </Step>
  <Step title="Drop the bundle into the template">
    Copy your composition into the template's compositions directory — `public/compositions/<your-name>/` for Vercel and Cloudflare, `compositions/<your-name>/` for Modal.
  </Step>
  <Step title="Point the template at it">
    - **Vercel**: edit `PREVIEW_COMPOSITION_DIR` at the top of `lib/preview.ts` and the dimensions in `app/page.tsx` if it isn't 1920×1080.
    - **Cloudflare**: set `PREVIEW_COMPOSITION_DIR=compositions/<your-name>` when running `npm run deploy`, or edit the default in `scripts/build.mjs`. Update player dimensions in `public/index.html` if needed.
    - **Modal**: set `PREVIEW_COMPOSITION = "<your-name>"` in `src/app.py`. Update the `<hyperframes-player>` dimensions in `web/index.html` if it isn't 1920×1080.
  </Step>
  <Step title="Deploy">
    ```bash Terminal
    # Vercel
    vercel deploy

    # Cloudflare
    npm run deploy

    # Modal
    modal deploy src/app.py
    ```
  </Step>
</Steps>

## When to use a template vs. roll your own

Templates are optimized for **a single render endpoint behind a preview UI**. They're the fastest way to get a hosted Hyperframes render API running. If you need:

- A **render queue** with retries, deduplication, or priorities — start from a template, then add your own queue (e.g. Vercel Queues, Cloudflare Queues, SQS).
- **Multi-tenant rendering** with per-user composition uploads — start from a template, replace the bundled composition with a runtime-fetched one.
- **Self-hosted** rendering — see the [Rendering guide](/guides/rendering) and run `hyperframes render --docker` on your own infrastructure.

For everything else, the templates are the recommended starting point.

## Next Steps

<CardGroup cols={2}>
  <Card title="Rendering" icon="film" href="/guides/rendering">
    Render compositions locally or in Docker
  </Card>
  <Card title="Player package" icon="play" href="/packages/player">
    Embed `<hyperframes-player>` in any HTML page
  </Card>
  <Card title="Vercel template" icon="github" href="https://github.com/heygen-com/hyperframes-vercel-template">
    Source on GitHub
  </Card>
  <Card title="Cloudflare template" icon="github" href="https://github.com/heygen-com/hyperframes-cloudflare-template">
    Source on GitHub
  </Card>
  <Card title="Modal template" icon="github" href="https://github.com/heygen-com/hyperframes-modal-template">
    Source on GitHub
  </Card>
</CardGroup>