Cloud

When to use managed cloud, Lambda, Cloud Run, or local

• hyperframes render (local): fastest iteration loop, use while authoring.
• hyperframes cloud render: zero-infra. HeyGen runs the render and you pay per credit. This is the default answer to “render in the cloud” when you don’t want to manage Chrome/FFmpeg/AWS.
• hyperframes lambda render: bring-your-own-AWS distributed rendering with chunked parallelism. Only worth it when you’ve already invested in AWS (see lambda.md).
• hyperframes cloudrun render: bring-your-own-GCP distributed rendering through Cloud Run and Workflows. Use only when GCP ownership is explicit (see cloudrun.md).

Authentication

Cloud rendering needs a HeyGen credential, stored at ~/.heygen/credentials (0600) and shared with the heygen CLI (opens in a new tab): sign in with one and the other picks up the session.

Credential resolution order (first match wins): HEYGEN_API_KEY, then HYPERFRAMES_API_KEY, then ~/.heygen/credentials. Point at a different backend with HEYGEN_API_URL (default https://api.heygen.com).

The render pipeline

cloud render runs end-to-end:

1. Resolve the project: a local directory (default .), or skip the upload with --asset-id / --url.
2. Auto-detect aspect ratio from the entry HTML’s data-width/data-height.
3. Zip the project (same ignore set as hyperframes publish, including .hyperframesignore).
4. Upload the zip through the direct-to-S3 asset flow, yielding an asset_id.
5. Submit the render to POST /v3/hyperframes/renders, yielding a render_id.
6. Poll GET /v3/hyperframes/renders/{id} until it completes or fails (skip with --no-wait).
7. Download the signed video URL to disk.

Archive size and .hyperframesignore

The direct-upload limit is 200 MB. HyperFrames automatically excludes root-level renders/ and snapshots/, along with its existing development exclusions such as .git, node_modules, dist, .next, coverage, and dotfiles. Add project-specific gitignore-style rules to <project>/.hyperframesignore when other generated or intermediate assets are not required at render time. The same rules affect hyperframes publish.

Inspect the exact archive without authenticating, uploading, spending credits, or starting a render:

The result reports compressed size_bytes, file_count, the 200 MB limit, and the ten largest included files.

When a cloud upload reports a size-limit error, agents must use this workflow:

1. Run the dry-run command and inspect the largest included files and directories.
2. Classify obvious generated outputs first: old renders, extra snapshot/contact-sheet directories, caches, exported previews, and source media used only to produce final assets.
3. Before excluding anything else, search src, href, url(), data-composition-src, JavaScript strings, manifests, and variable-driven paths across every HTML, CSS, and JavaScript entry.
4. Preserve existing .hyperframesignore comments and rules. Add the narrowest verified-unneeded root-relative paths; prefer an exact directory or file over a broad wildcard.
5. Never ignore index.html, the selected composition, mounted sub-compositions, fonts, images, audio, video, scripts, or manifests merely because they are large. Never ignore all of assets/.
6. Rerun dry-run until the archive is below the limit, then run npx hyperframes check. Remember that check sees the source directory, so it cannot prove a dynamically computed asset path remains in the filtered archive; the reference audit is still required.

Example:

Rules support comments, globs, and negation. A later rule can override a default, for example !/snapshots/ when that directory intentionally contains render inputs.

Render options

--resolution 4k cannot combine with --format webm/mov: the 4k supersampling path has no alpha channel. Render 4k as mp4, or render alpha at native resolution.

Templates and variables

Cloud rendering supports composition variables (opens in a new tab): declare data-composition-variables on the composition, then fill them at render time.

For a local project the CLI validates --variables against the declared schema before uploading. For --asset-id/--url the schema lives server-side, so mismatches surface as a hyperframes_project_invalid API error.

Upload once, re-render many is the idiomatic template loop: render a local project to get its asset_id, then re-submit against that asset with new values (no re-zip, no re-upload).

For high-volume personalized batches, both self-managed paths provide JSONL fan-out: AWS Lambda (lambda.md) and Google Cloud Run (cloudrun.md). The full variables schema (types, declarative bindings, sub-composition overrides, precedence) lives in the hyperframes-core skill.

Fire-and-forget and webhooks

By default the CLI blocks, polls, and downloads. Combine --no-wait (submit and exit with just the render_id) with --callback-url (HTTPS webhook on terminal status) for true fire-and-forget:

Managing renders

video_url and thumbnail_url are short-lived presigned URLs, so re-fetch with cloud get rather than caching them.

Safe retries

The CLI transparently retries a 401 by force-refreshing the OAuth token and replaying. That’s harmless for reads, but the zip upload (POST /v3/assets) is not idempotent: a blind retry creates a duplicate asset and bills twice. Pass --idempotency-key so retries are safe:

The key is forwarded to both upload and submit (the server scopes idempotency per-endpoint, so reusing one value is safe). Use any opaque string in [A-Za-z0-9_:.-], 1–255 chars.

Full flag reference: docs /deploy/cloud and /packages/cli#hyperframes-cloud.