Lambda

• Choose Lambda or local rendering
• Prerequisites
• Deploy
• Upload a reusable site
• Render one composition
• Render a JSONL batch
• Inspect progress
• Destroy the stack
• IAM policies
• State, cost, and cleanup

The basic lifecycle is:

Choose Lambda or local rendering

• Local render — dev-loop iteration, single host, anything under a few minutes at 1080p.
• lambda render — long videos, 4K, large parallel batches, or anything where local Chrome would time out / exhaust RAM. Pay-per-invocation, no idle cost.

For one-off short renders Lambda is not worth the deploy overhead.

Prerequisites

• AWS credentials configured (env vars, ~/.aws/credentials, SSO, or IMDS).
• AWS SAM CLI on PATH.
• bun on PATH (builds the Lambda handler ZIP).

Deploy

Builds packages/aws-lambda/dist/handler.zip and SAM-deploys the stack (Lambda + Step Functions + S3 + IAM). Idempotent — re-running on the same --stack-name is a no-op when nothing changed. Writes <cwd>/.hyperframes/lambda-stack-<name>.json so later subcommands don’t need to call describe-stacks.

Upload a reusable site

Tars + uploads <projectDir> to S3 with a content-addressed key. Returns a stable siteId you can reuse — re-renders of the same tree skip the upload.

Render one composition

Starts a Step Functions execution. Returns immediately with a renderId unless --wait is set, in which case the CLI blocks until completion and streams per-chunk progress lines. Add --json for machine-parseable output.

--width / --height footgun. Setting --width 3840 --height 2160 against a composition whose data-width="1920" silently produces 1080p — the runtime lays out the page at the composition’s authored dimensions and the CLI flags are ignored for layout. To actually output at 4K, use --output-resolution 4k (supersamples via deviceScaleFactor). The CLI now prints a warning when CLI dimensions disagree with the composition’s data-width / data-height and --output-resolution is not set; the warning is suppressed when --json is on or index.html isn’t on disk (--site-id flows).

For variable-driven templates, declare the schema in the composition and pass either --variables or --variables-file, never both. --strict-variables checks local project input before any render starts. Also read variables-and-media.md (opens in a new tab).

Render a JSONL batch

Use render-batch to upload one template once and start one Step Functions execution per nonblank JSONL line:

Each line must be an object with a non-empty outputKey. Choose unique keys to prevent outputs from overwriting one another. variables and executionName are optional:

Batch rules:

• The project is uploaded once unless --site-id reuses an earlier upload.
• --max-concurrent defaults to 50 and limits in-flight render executions. --max-parallel-chunks separately limits chunks inside each render.
• --strict-variables checks every entry, reports all variable issues, and aborts before AWS calls.
• --dry-run performs no upload or AWS render call. Every manifest row becomes would-invoke.
• The emitted manifest preserves input order and records inputLine, outputKey, executionArn, and status (started, would-invoke, or failed-to-start), plus an error when applicable.
• A per-entry start failure does not hide other rows. Human-output mode exits nonzero when a row fails to start. In --json mode the current CLI prints the manifest and exits zero, so gate on every row’s status, not the process code alone. Dispatch success is not render completion; inspect each execution with progress.

Inspect progress

Prints one snapshot — overall percent, frames rendered, Lambda invocations, accrued cost, and any errors. Accepts a bare renderId (resolved against the stack’s state-machine ARN) or a full SFN execution ARN.

Destroy the stack

Calls sam delete --no-prompts and drops the local state file. The render S3 bucket is configured Retain so it survives stack destruction — empty + delete it via the AWS console / CLI if you want the storage back.

Non-retryable errors

A subset of failures the Step Functions state machine short-circuits instead of running through its 4× 15-min retry budget. progress surfaces these immediately with the error class name; do not re-issue lambda render blindly when you see one.

• ChromeBinaryUnavailableError — @sparticuz/chromium returned an empty/missing executable path. A prior chunk hit Sandbox.Timedout mid-extraction and the warm instance is wedged until the execution environment recycles. Remedy: bump a Lambda env var (forces a new exec env) or lambda deploy again. Not a transient render failure; retries will burn budget on the same wedged instance.
• FFMPEG_VERSION_MISMATCH / PLAN_HASH_MISMATCH — planner / executor version drift. Re-deploy.

IAM policies

Print or validate the minimum IAM permissions the CLI needs.

validate reads a JSON policy doc and checks the union of its Effect: Allow actions (expanding s3:* / s3:Get* / * wildcards) against the CLI’s required action set. Missing actions print to stderr; the command exits non-zero. Wire it into CI to catch policy drift before the next deploy fails.

The default action set is deliberately broad (Resource: "*") because CloudFormation creates new ARNs on every adopter’s first deploy. Tighten Resource after that first run if security posture requires it.

State, cost, and cleanup

hyperframes lambda stores per-stack metadata under <cwd>/.hyperframes/lambda-stack-<name>.json (bucket name, state-machine ARN, region). Not secret, but AWS-account-identifying. Commit it to a repo or .gitignore it per your workflow.

• lambda destroy removes the SAM stack but leaves the S3 bucket (Retain). Delete it manually if you want the storage back.
• Lambda billing is per-invocation + duration. progress reports the accrued cost.
• --concurrency caps parallel Lambda invocations — keep it aligned with your account quota.
• --chunk-size and --max-parallel-chunks trade off per-chunk overhead against parallelism; larger chunks reduce coordinator overhead, smaller chunks parallelize more aggressively.