Deploy distributed HyperFrames rendering to AWS Lambda and drive renders from a laptop or CI.
HyperFrames ships a first-class AWS Lambda deployment: one Lambda function fronts a Step Functions standard workflow that fans renders out across many parallel chunk workers, with intermediate artifacts in S3. End-to-end is three commands once your AWS credentials are configured.
Templates with variables work on the same Lambda stack — declare data-composition-variables on the composition, then pass values per render with --variables or fan a whole batch out with lambda render-batch. See the Templates on Lambda guide for the personalised-render pipeline (single render, batch from JSONL, programmatic SDK) and the 256 KiB Step Functions execution-input cap.
┌──────────────────────────────────────────────────────────────────┐│ Step Functions state machine ││ Plan → Map(N) RenderChunk → Assemble │└──────────────────────────────────────────────────────────────────┘ │ dispatches by event.Action ▼┌──────────────────────────────────────────────────────────────────┐│ One Lambda function (packages/aws-lambda/dist/handler.zip) ││ handler.mjs ││ ├─ Action="plan" → @hyperframes/producer/distributed ││ ├─ Action="renderChunk" → @hyperframes/producer/distributed ││ └─ Action="assemble" → @hyperframes/producer/distributed ││ bin/ffmpeg — ffmpeg-static ││ node_modules/@sparticuz/chromium/ — Lambda-optimised Chromium │└──────────────────────────────────────────────────────────────────┘ │ pure functions over local paths ▼┌──────────────────────────────────────────────────────────────────┐│ S3 bucket — plan tarball + per-chunk outputs + final mp4 │└──────────────────────────────────────────────────────────────────┘
The Lambda handler is a thin dispatch: parse the Step Functions event, download inputs from S3 into /tmp, call the OSS primitive from @hyperframes/producer/distributed, upload outputs back, return a small JSON result. Everything heavy — capture, encode, audio mix — happens inside the OSS primitives.
The published CLI loads the AWS adapter only for Lambda commands. Each release pins a prebuilt Lambda handler ZIP (a GitHub release asset); lambda deploy downloads it, checks its SHA-256 and deploys it with the adapter’s SAM template. No repo checkout needed.
npm install -g @hyperframes/aws-lambda alongside the CLI, at the same version
HyperFrames repo checkout (optional)
Inside a checkout, or with HYPERFRAMES_REPO_ROOT pointing at one, lambda deploy builds the handler ZIP from source instead (needs bun). Use it to deploy unreleased changes, --chrome-source=chrome-headless-shell, or an adapter release older than the prebuilt handler. A checkout deploy on arm64 also needs the video image built from that checkout (HYPERFRAMES_LAMBDA_VIDEO_IMAGE), or --video-arch x86_64.
git clone https://github.com/heygen-com/hyperframes, and npm install -g bun or bun.sh
By default the stack reserves no concurrency. Each render reads the account’s
unreserved Lambda concurrency when it plans, and runs up to 90% of it at once
(at most 256 chunks), leaving the rest for other functions in the account. The
plan logs the number it chose. Pass --concurrency=N to reserve N executions
for the function instead; every render then runs at most N chunks at once.Upgrading: redeploying an existing stack without --concurrency keeps the
reservation it already has, and deploy prints it. Stacks deployed by earlier
CLI versions reserved 8 by default, so they stay at 8 until you pass
--concurrency=N. Pass --concurrency=none to remove the reservation so each
render sizes itself to the account.deploy records the stack’s bucket and state machine in .hyperframes/lambda-stack-<stack-name>.json under the current directory. Every other lambda command reads that file, so run them from the same directory with the same --stack-name, or they stop with “no stack state”. Redeploying to upgrade rewrites the same file.After deploy, render anything with:
The --wait flag blocks and streams per-chunk progress + accrued cost; drop it to fire-and-forget, then poll with hyperframes lambda progress <renderId> on your own cadence.See the CLI reference for full flag documentation.
Every lambda render call without --site-id hashes the project tree and checks each file against the bucket. For tight inner loops (CI smoke jobs, prompt iteration in a demo flow), pre-stage the project once and reuse the upload:
The siteId is content-addressed via a SHA-256 of the project tree. A site is a small sites/<siteId>/files.json that lists each file’s path and SHA-256. The files themselves live once per bucket under cas/sha256/, shared by every site and render, so a file another project already uploaded is never uploaded again. Pass the same --site-id to as many lambda render calls as you like.The stack’s bucket expires objects under cas/sha256/ 30 days after they were last written (the ExpireSharedContent lifecycle rule). Each render or sites create that reuses a file at least 7 days old rewrites it in place, so files in use never expire. If a site’s files did expire, lambda render --site-id re-uploads them from the project directory, if its copies still match. Otherwise, or when renderToLambda gets a siteHandle but no projectDir, the render stops with SITE_FILES_MISSING before starting; run sites create again to restore them.
lambda deploy now runs the plan and the render chunks on an arm64 container-image function by default; audio and assemble stay on the x86 function. Redeploying an existing stack without --video-arch moves it to arm64 and adds the <stack>-video ECR repository. To keep a stack on x86, pass --video-arch x86_64 on every deploy. --chrome-source chrome-headless-shell also needs --video-arch x86_64, because the arm64 image bundles sparticuz Chrome only.
Upgrading from a version before shared content storage
Redeploy the stack before upgrading the CLI or SDK that starts renders. A stack from an earlier version cannot read files.json sites. The CLI and renderToLambda check the state machine first and stop with STACK_TOO_OLD before uploading anything; run hyperframes lambda deploy (or redeploy your SAM template or CDK construct from the same version) and render again.
Existing site ids keep working. When sites/<siteId>/files.json is missing, the render uses that site’s project.tar.gz from the earlier version. If neither exists the render stops with SITE_NOT_FOUND, naming both keys; run sites create again.
Every project uploads once more. The site id scheme changed (it now also covers symlinked files), so sites create or the first lambda render of each project after the upgrade returns a new site id and uploads its files once.
If you want to read the CloudFormation before you deploy, or you need to customise the topology (extra alarms, SNS subscribers, KMS keys, …), invoke SAM directly against the template at examples/aws-lambda/template.yaml:
The template emits three CloudFormation outputs you’ll need to invoke renders:
RenderBucketName — S3 bucket for plan tarballs + per-chunk outputs + final renders.
RenderStateMachineArn — the Step Functions standard workflow that orchestrates Plan → Map → Assemble.
RenderFunctionArn — the single Lambda function the state machine dispatches against.
The SAM template’s default for ReservedConcurrency is -1 (unreserved), the same as the Path 1 CLI: each render runs up to 90% of the account’s unreserved concurrency. Set it to bound how many chunks run at once.
For users already running CDK, the @hyperframes/aws-lambda package exports a HyperframesRenderStack L2 construct that emits the same topology as the SAM template:
import { App, CfnOutput, Stack } from "aws-cdk-lib";import { HyperframesRenderStack } from "@hyperframes/aws-lambda/cdk";const app = new App();const stack = new Stack(app, "MyApp");const render = new HyperframesRenderStack(stack, "Render", { projectName: "hyperframes", lambdaMemoryMb: 10240, reservedConcurrency: 8, chromeSource: "sparticuz",});new CfnOutput(stack, "RenderBucketName", { value: render.bucket.bucketName });new CfnOutput(stack, "StateMachineArn", { value: render.stateMachine.stateMachineArn });
aws-cdk-lib and constructs are declared as optional peer dependencies of @hyperframes/aws-lambda, so consumers who only need the SDK don’t pay the CDK import cost.The construct exposes .bucket, .renderFunction, and .stateMachine so you can wire dashboards, SNS topics, or other AWS resources alongside it without re-deriving ARNs.
The CLI ships a built-in IAM bootstrap to avoid the “User is not authorized to perform iam:CreateRole” first-deploy trap:
# Print an inline policy doc to attach to the IAM user that runs the CLI.hyperframes lambda policies user# Print { TrustRelationship, InlinePolicy } for a CloudFormation service role.hyperframes lambda policies role# Validate a checked-in policy still covers the CLI's needs (exit non-zero on missing).hyperframes lambda policies validate ./infra/iam/hyperframes-deploy.json
Code that calls the SDK’s renderToLambda with its own role needs states:DescribeStateMachine on the state machine besides states:StartExecution and S3 read/write on the bucket: each render reads the definition to confirm the stack is new enough.The generated documents grant Resource: "*" for the CLI’s required action set. After your first successful deploy you can narrow Resource to the deployed ARNs — predictable per the CloudFormation outputs above. Adopters running the CLI in CI typically check the policy doc into source control and run policies validate as a pre-deploy step to catch drift.
Lambda renders are billed by GB-seconds (Lambda billed duration × configured memory) plus a tiny per-state-transition fee for Step Functions standard workflows. hyperframes lambda progress exposes the running tally:
The IAM credential running lambda deploy doesn’t have permission to create the service role CloudFormation needs. Run hyperframes lambda policies user and attach the printed policy to your IAM user (or take the policies role output and have your admin create a deploy role).
Step Functions invoked a renderChunk with a plan hash that didn’t match the planDir on S3. Almost always means the producer version differs between the local plan() build and the deployed Lambda ZIP. Re-run hyperframes lambda deploy (which rebuilds the ZIP) and re-render.
The handler launched Chromium but the runtime probe found a non-SwiftShader GL backend. Hardware GL is non-deterministic across chunk boundaries, so distributed renders refuse it at the runtime-image / launch-flags layer (not at the composition layer). Rebuild the handler ZIP and redeploy:
bun run --cwd packages/aws-lambda build:ziphyperframes lambda deploy --stack-name=<your-stack>
The build pipeline pins @sparticuz/chromium + the Chrome flags (--use-gl=swiftshader --use-angle=swiftshader) so a fresh deploy almost always resolves this. If it persists, your stack’s Lambda function is pointing at a stale handler ZIP from a previous deploy — lambda deploy always rebuilds, so re-running unsticks it.
Most often a Lambda cold-start chain on a many-chunk render. The Map state’s reserved concurrency caps how many chunks can run in parallel — if you set --concurrency=4 and your render has 16 chunks, the state machine processes them in batches of 4. hyperframes lambda progress <id> shows how many invocations are in flight.If progress doesn’t advance for >10 minutes, check the Step Functions execution in the AWS console — failed Lambda invocations include the typed error name (FONT_FETCH_FAILED, FFMPEG_VERSION_MISMATCH, etc.) which short-circuits the state machine.
Check the failed invocation’s CloudWatch logs to distinguish Plan from RenderChunk. In current versions, VIDEO_EXTRACTION_FAILED with ffmpeg_timeout means extraction stopped producing new frames for the configured interval, which defaults to 300,000 ms. Frame progress resets that timer, so an extraction can run for longer than five minutes while it keeps producing frames. Normal frame extraction splits long ranges into segments of up to two minutes, including variable-frame-rate sources.Plan v2 leaves eligible constant-frame-rate SDR extraction to chunk workers after a short timing probe. Variable-frame-rate sources and required HDR color conversions still extract during Plan; the deprecated v1 path also extracts videos during Plan. Smaller chunks or a higher maxParallelChunks can reduce each chunk’s work, but do not divide extraction that still happens during Plan.If the logs show that a slow extraction needs a longer gap between frames, and the work fits within the Lambda invocation budget, increase FFMPEG_PROCESS_TIMEOUT_MS on the deployed function that runs the failing phase. The default arm64 deployment uses the video function for Plan and RenderChunk. Setting the variable only in the terminal running lambda render does not configure the remote function. CDK users can configure the construct’s render function before deploying. Here, render is the HyperframesRenderStack instance created in the CDK example above:
This allows extraction to wait ten minutes without a new frame. Each Plan, RenderChunk, and Assemble invocation must still finish within its own configured Lambda timeout, at most 15 minutes, including downloads, processing, and uploads, and fit its /tmp storage. Measure the failing phase’s elapsed time and storage use in CloudWatch.If planning cannot fit, split the composition into shorter, frame-aligned projects, render each as a separate job, and combine the results. Choose part lengths from measured planning times. There is no fixed maximum video length: resolution, frame rate, codec, and Lambda resources affect extraction time.
The render bucket is created with CloudFormation Retain on delete — hyperframes lambda destroy (or sam delete) tears the function + state machine down but the bucket survives. This is intentional: it protects final-rendered MP4s from being lost when you re-deploy. To fully reclaim storage, empty + delete the bucket via the AWS console / aws s3 rb.A stack deployed on arm64 (the default) also has an ECR repository, <stack>-video, holding the video function’s image. hyperframes lambda destroy deletes it after the stack.
The stack was deployed by an earlier version that cannot read this version’s sites. Redeploy it (hyperframes lambda deploy), then render again. See Upgrading.
SITE_FILES_MISSING: some of the site’s files expired from cas/sha256/ after 30 days unused. Render from the project directory the site was created from, or run sites create again. SITE_NOT_FOUND: the bucket has no site under that id in either format; run sites create again and use the id it prints.
There is no completion webhook. Poll with hyperframes lambda progress or
watch the Step Functions execution.
There is no Lambda-specific composition-discovery verb. Point
lambda render at the project directory containing index.html.
Each --region is an independent stack. Cross-region failover is not built
in.
The distributed AWS path is SDR-only.
Plan v2 leaves eligible constant-frame-rate SDR video extraction to each
render chunk after a short timing probe. Each chunk extracts only the frames
it shows. On a 10,240 MB function, a 26-minute 576×1024 24 fps H.264 clip
plans in about 20 seconds. Each chunk reads only the byte ranges of each
video it shows, straight from the bucket. Variable-frame-rate sources and
required HDR color conversions still need extraction during Plan. Normal
frame extraction uses segments of up to two minutes, including variable-frame-rate
sources; FFMPEG_PROCESS_TIMEOUT_MS limits how long extraction can go without
producing a new frame. See long-video planning timeouts
for timeout configuration and splitting jobs that cannot fit.
Plan, each RenderChunk, and Assemble must finish within their own configured
Lambda timeout, at most 15 minutes, and fit the function’s /tmp storage.
For long renders, raise --max-parallel-chunks (maxParallelChunks in the SDK) so each chunk covers
fewer frames.
