FFmpeg API

Rendobar's FFmpeg API is hosted FFmpeg as a service. Send the command you already tested in one REST call, and it runs as written on FFmpeg 9.0, on CPU or GPU, and returns the output URL. Billed per second.

Your code, the CLI, a workflow tool or an AI agent sends FFmpeg jobs like the one above to Rendobar. They run in parallel across CPU and GPU machines, each output is delivered to your storage (Amazon S3, Cloudflare R2, Supabase Storage or any S3-compatible store), and job.deliveries_settled is posted to your webhook once every copy has landed.
Your toolsPOST /jobs
await rb.jobs.create({
type: "ffmpeg",
params: {
command: `ffmpeg -i ${url} -vf scale=1280:-2 out.mp4`,
},
});
~/videos $ rb ffmpeg -i in.mp4 -vf scale=1280:-2 out.mp4 ✓Uploading0.6s✓ExecutedExecuting1.9s ✓Saving0.2s →out.mp4

RendobarCreate job

Job Typeffmpeg

Commandffmpeg -i {{ $json.url }} -vf scale=1280:-2 out.mp4

in.mp4Can you make this video 720p?

submit_jobRendobar

Done, it's 1280×720 now: out.mp4

curl https://api.rendobar.com/jobs \
-H "Authorization: Bearer $RENDOBAR_API_KEY" \
--json '{"type": "ffmpeg", "params": {
"command": "ffmpeg -i https://…/in.mp4 -vf scale=1280:-2 out.mp4"
}}'
Our machines

Capabilities

What you can do with the FFmpeg API

Each task below is one FFmpeg command sent to POST /jobs, and each one opens the command we ran for it. No preset menu, no per-feature pricing.

GPU acceleration

Encode on GPU with NVENC

Change the encoder and the same command runs on an NVIDIA L4, L40S or RTX PRO 6000, with no instances to provision and no drivers to install. Leave compute on auto and any command that already uses NVENC or CUDA goes to a GPU on its own.

CPU to GPU, one line
ffmpeg -i in.mp4 \
--c:v libx264 -crf 23 \CPU, every plan
+-c:v h264_nvenc -cq 23 \GPU, Pro and Studio
out.mp4
hevc_nvenc and av1_nvenc work the same way. NVENC sets quality with -cq, where x264 uses -crf.

Billed per second at the GPU it ran on. For archival masters, software libx264 at a slow preset still wins on quality per bit, and it stays one flag away.

Pricing

FFmpeg API pricing and limits

A job pays for the seconds it runs, so the command you send sets the price. A short command costs a fraction of a cent and a long encode costs what it runs. Every plan runs the same FFmpeg 9.0, and the paid plans raise the limits.

Median FFmpeg job
$0.0025
Across the FFmpeg jobs we have measured
VP9 transcode
$0.0071
A 5-second clip, 6.1 s of compute
1,002-second encode
$1.02
About $0.001 for each second on CPU
Free credit
$5
No card, about 2,000 median jobs
PlanFreeProStudio
Price$0, no card$9 a month$29 a month
Credit included$5 once, at signup$5 a month$20 a month
Bonus on credit you buyNone25%25%
Longest job1 hour9 hours9 hours
Largest input500 MB10 GB20 GB
Jobs at once12050
Outputs kept7 days30 days90 days
GPU encoding (NVENC)CPU onlyYesYes

Every plan is on Rendobar pricing. For the measured jobs line by line, with GPU rates and what a failed command costs, read what each FFmpeg job costs.

Compare

Full FFmpeg, zero ops

Self-host FFmpeg and you own the servers, scaling, and security. Use a preset-based API and you lose raw control. Rendobar's FFmpeg API runs every FFmpeg command and leaves you nothing to run.

Compared onRendobarSelf-hosted FFmpegPreset-based APIs
Run any FFmpeg command Yes Yes Presets only
Servers to run None You manage them None
Time to first job One POST Hours to days Minutes
Pay only for what runs Per compute second Servers bill 24/7 Often GB-metered
Sandboxed per job Yes, isolated Your responsibility Not exposed
Async webhooks + realtime Built in Build it yourself Varies
Native MCP for AI agents Yes No No
Zero egress on output Yes Depends on host Often charged
FFmpeg upgrades + security Managed Your job Managed
Free to start $5, no card Pay for infra Card required

Compared against hosting FFmpeg yourself and against preset-based APIs.

FAQ

Frequently asked questions

What is an FFmpeg API?

An FFmpeg API is FFmpeg as a service: you send an FFmpeg command and its input URLs over HTTPS, the command runs on servers you do not manage, and you get the output file's URL back. FFmpeg itself has no HTTP interface. It ships a command-line tool and C libraries such as libavcodec and libavformat, and its own docs call those the API. Rendobar's FFmpeg API is the hosted kind. It runs the command as you wrote it, so any filter graph FFmpeg supports works, with no preset list in between.

Does Rendobar change my FFmpeg command?

No. The job runs your argv as written on FFmpeg 9.0, on CPU and GPU alike, and reports the exact build in its ffmpegVersion field. Reference each input by the name in the command, and every file the command writes comes back under the name you gave it. The one thing refused at submit is shell syntax such as a pipe or a redirect, because no shell runs around the command. A failed job returns FFmpeg's own first error line, so you debug against FFmpeg.

How much does the FFmpeg API cost?

A job is billed for the seconds it runs, about $0.001 a second on CPU, so the command you send sets the price. A GPU job bills at the rate of the GPU it ran on. A failed job pays only for the seconds it ran, which for most FFmpeg errors is under a second, and a job that never starts costs nothing. If our infrastructure caused a failure, support refunds it. Pro is $9 a month and Studio $29, each with credit included.

Is there a free FFmpeg API?

Yes. Every Rendobar account starts with $5 of free credit and needs no card. On the Free plan a job runs up to 1 hour and each input can be up to 500 MB. The median FFmpeg job we have run cost $0.0025, so the credit runs about 2,000 jobs of that size. The free FFmpeg API guide runs real jobs against that credit.

How do I get an FFmpeg API key?

Sign up at app.rendobar.com, open API keys and create one. It is shown once, so store it as an environment variable. Send it in the Authorization header as a Bearer token, or pass it to the Rendobar SDK. The quickstart runs a first FFmpeg command with it.

Is the API async? How do I get the result?

Yes. POST /jobs returns a job ID at once and the command runs in the background. Rendobar sends a signed webhook to your URL when the job finishes, or you poll GET /jobs/{id}, or the SDK waits for you and hands back the output URL. From a terminal or CI, the Rendobar CLI runs rb ffmpeg and downloads the file for you.

How long can a single job run?

Free plan jobs run up to 1 hour. Pro and Studio raise that to 9 hours per job. That is a limit on processing time, not on how long your output can be. There is no cap on output length, so a long render is fine as long as it finishes inside the budget.

Can I run FFmpeg on Cloudflare Workers, Vercel, or AWS Lambda?

Not the way FFmpeg runs on a server. A Worker can't start the FFmpeg binary or a second thread, and everything shares 128 MB. A Vercel function takes a request body up to 4.5 MB and runs up to 800 seconds on Pro, and Lambda stops at 15 minutes with no GPU. From any of them, call Rendobar's FFmpeg API with one HTTPS request and public or presigned input URLs, and the job runs on its own machine. The Vercel, Cloudflare Workers and AWS Lambda guides show the code.

Can Claude Code or Cursor run FFmpeg through the API?

Yes. Rendobar runs an MCP server, so Claude Code, Cursor and other MCP clients can upload a file, submit an FFmpeg job and read the result inside a conversation, with no FFmpeg on your machine. The MCP server page has the install step for each client.

What can't Rendobar's FFmpeg API do?

It makes files. It does not ingest a live stream, record a screen or host a player that streams video to viewers. A job reads its inputs from URLs or uploads and runs one FFmpeg command with no shell around it, so you can't pipe one process into another. GPU encoding needs Pro and Studio. The free plan runs on CPU, one job at a time.

Your command. Our infrastructure.

Bring your own commands. Run them on scalable CPU and GPU infrastructure, and pay only for compute time.