Media Converter for AI Agents (FFmpeg)

A pay-per-file FFmpeg API for AI agents and automations: send a public audio or video URL and get back an MP3, a Whisper-ready 16 kHz mono WAV, a web-ready MP4, a GIF or a thumbnail, with no ffmpeg install, no Docker and no server to run.

What this FFmpeg API does

You give it a media URL and a target format. It downloads the file, runs FFmpeg, stores the result in your run's key-value store, and returns one result row per file with a download link (outputUrl) and technical metadata.

Target (target)	What you get
wav16k	Mono 16 kHz WAV, ready for Whisper and other speech-to-text models
mp3, wav, flac, ogg, m4a	Audio extracted from video, or audio transcoded between formats
mp4	H.264 MP4 with faststart, for web playback (max 1080p)
webm	VP9 WebM (max 1080p)
gif	Animated preview, 8 fps, 360 px wide, no audio (billed per output minute like audio/video)
thumbnail	A JPEG frame: the first frame, or the frame at startSeconds

On top of that:

• Trim: set startSeconds and durationSeconds next to url (they also act as the default for every files item), or on individual files items, to convert only that part.
• Batch: pass a files array to convert many URLs in one run, each with its own target and trim.
• Metadata: each row includes the source's container format, duration, size, video codec and resolution, and audio codec, sample rate and channels (from ffprobe).
• Failed files are free: an unreadable, unreachable or rejected file comes back as a status: "error" row with an error message. You aren't charged for it and the run keeps going.

Why use it instead of running FFmpeg yourself

• Nothing to install. No ffmpeg binary, no Docker image, no codecs to manage, no worker server to keep alive.
• One call, one outcome. {"url": "...", "target": "wav16k"} returns a Whisper-ready 16 kHz mono WAV. No flag soup.
• Low risk. You pay per converted file plus output minutes. No subscription, no idle server, and failed files cost nothing.
• Works where your automation already lives. Use it from MCP clients like Claude and Cursor (through Apify's MCP server), n8n, Make, Zapier or a plain HTTP call.
• Agent-friendly output. One structured row per file, with a download URL your agent can pass to the next step (transcription, upload, storage).

Use cases

Whisper audio preprocessing: 16 kHz WAV in one call

Speech-to-text models such as Whisper expect 16 kHz mono audio. Convert a recorded call, a webinar video or a podcast to a speech-ready file before you transcribe it.

{
  "url": "https://example.com/recordings/team-call.mp4",
  "target": "wav16k"
}

Convert video to MP3 / extract audio from video

Pull the audio track out of a video for podcasts, audiograms or archives. Use m4a, ogg or flac if you need a different container or lossless output.

{
  "url": "https://example.com/videos/interview.mp4",
  "target": "mp3"
}

Video to GIF

Make a short animated preview (8 fps, 360 px) for a social post, a product page or a chat message. Trim it so the GIF covers only the moment you want.

{
  "files": [
    { "url": "https://example.com/videos/demo.mp4", "target": "gif", "startSeconds": 5, "durationSeconds": 6 }
  ]
}

Video thumbnail generator

Grab a JPEG frame for a catalog, a CMS or a video card. Leave out startSeconds to get the first frame.

{
  "files": [
    { "url": "https://example.com/videos/episode-12.mp4", "target": "thumbnail", "startSeconds": 42 }
  ]
}

Trim audio API

Cut a clip from a long recording, for example the first 10 minutes of a meeting to send to a summarizer. You pay only for the output minutes.

{
  "url": "https://example.com/audio/all-hands.m4a",
  "target": "mp3",
  "startSeconds": 0,
  "durationSeconds": 600
}

Batch conversion for n8n / Make pipelines (n8n ffmpeg, make ffmpeg)

Convert a whole list in one run. The top-level target, startSeconds and durationSeconds are defaults, and each file can override them. Files are processed one after another.

{
  "target": "wav16k",
  "files": [
    { "url": "https://example.com/podcast/ep-101.mp3" },
    { "url": "https://example.com/podcast/ep-102.mp3" },
    { "url": "https://example.com/podcast/ep-103-trailer.mp4", "target": "thumbnail", "startSeconds": 3 }
  ]
}

FFmpeg MCP tool for AI agents

Give Claude, Cursor or another MCP client a media tool. Your agent sends a URL and a target and gets a converted file link back. See Use from MCP.

{
  "url": "https://example.com/uploads/voice-note.ogg",
  "target": "wav16k"
}

Web-ready video (MP4 / WebM)

Re-encode uploads into H.264 MP4 with faststart, or VP9 WebM, so they play smoothly in the browser. Video that is already web-ready is copied without re-encoding and billed at the normal media-minute rate. Re-encoding above 720p (1080p-class) is billed per HD re-encode minute; see Pricing.

{
  "url": "https://example.com/uploads/raw-clip.mov",
  "target": "mp4"
}

Pricing

Pay per event. You pay only for successful output:

Event	Price
File converted (per successful output file)	$0.005 ($5 per 1,000 files)
Media minute (per started minute of output audio, GIF, stream-copied video, or video re-encoded at 720p or lower; thumbnails excluded)	$0.002
HD video re-encode minute (per started minute when mp4/webm video is actually re-encoded above 720p, i.e. 1080p-class after the 1080p cap; charged instead of the media minute, never both)	$0.04
Actor start (per run, one event per GB of memory, minimum one; default 1 GB = 1 event)	$0.00005 per event

Failed files are free. Each output file pays exactly one per-minute rate. Every result row shows which one in minuteEvent.

Why HD re-encoding costs more: encoding 1080p video takes about 8× real time on the default 1 GB run, which costs far more in compute than $0.002 a minute covers. To stay on the cheap rate, send H.264/AAC (for mp4) or VP9/Opus (for webm) at 1080p or below without trimming, so it's copied rather than re-encoded. You can also downscale to 720p first.

Worked cost examples

• 100 one-hour podcasts to Whisper-ready WAV (wav16k):
  • File events: 100 × $0.005 = $0.50
  • Media minutes: 100 × 60 min × $0.002 = $12.00
  • Total: $12.50, or $0.125 per episode (plus $0.00005 actor start for one batch run at the default 1 GB).
  • Minutes are counted per started minute of output, so a 60 min 30 s episode bills 61 minutes ($0.127).
• 1,000 video thumbnails: 1,000 × $0.005 = $5.00. Thumbnails have no minute charge.
• A 6-second GIF preview: $0.005 + 1 × $0.002 = $0.007.
• One 10-minute podcast to wav16k: $0.005 + 10 × $0.002 = $0.025.
• A 30-second 1080p H.264 MP4 to mp4 (already web-ready, copied): $0.005 + 1 × $0.002 = $0.007.
• A 30-second 1080p clip re-encoded to mp4 or webm (e.g. a MOV/HEVC phone video, a 4K source scaled to 1080p, or any trimmed video): $0.005 + 1 × $0.04 = $0.045. A 10-minute 1080p re-encode is $0.005 + 10 × $0.04 = $0.405.
• A 30-second 720p clip re-encoded to mp4: $0.005 + 1 × $0.002 = $0.007.
• Cut costs with trim: converting only the first 10 minutes of a 1-hour recording bills 10 minutes, not 60.

To cap spend on a run, set a maximum total charge in the run options.

Input reference

Field	Type	Default	Description
url	string	–	Public URL of one audio/video file.
target	string	mp3	Output format: mp3, wav, wav16k, flac, ogg, m4a, mp4, webm, gif, thumbnail. Default for all files in a batch.
startSeconds	integer	–	Trim start in seconds (for thumbnail: the frame time). Applies to url and is the default for every files item.
durationSeconds	integer	–	Seconds to keep from startSeconds. Applies to url and is the default for every files item.
files	array	–	Optional batch. Each item: url, and optionally target, startSeconds, durationSeconds. Plain URL strings also work.
maxDurationSeconds	integer	10800 (3 h)	Safety cap. Longer inputs are rejected (free) unless durationSeconds is set (top-level or on the item).

Per-file options inside files (each falls back to the top-level value when omitted):

Field	Description
url	Public URL of the media file.
target	Overrides the top-level target for this file.
startSeconds	Where to start (trim start, or the thumbnail frame time). Can be fractional, e.g. 12.5.
durationSeconds	How many seconds to convert from startSeconds.

Video output (mp4, webm) is capped at 1080p: larger inputs are scaled down to fit 1920×1080 (or 1080×1920 for portrait), keeping the aspect ratio. Inputs that are already in the target codec (H.264 for mp4; VP9, VP8 or AV1 for webm), 8-bit 4:2:0 and at most 1080p are remuxed without re-encoding when no trim is set, so quality is unchanged.

Limits: up to 2 GB per file, 3 hours maximum duration by default. Only process media you have the right to use.

Output reference

One dataset row per input file. The converted files are stored in the run's default key-value store, and outputUrl points to the file.

{
  "sourceUrl": "https://example.com/recordings/team-call.mp4",
  "target": "wav16k",
  "status": "ok",
  "outputKey": "57f39f65123e.wav",
  "outputUrl": "https://api.apify.com/v2/key-value-stores/<storeId>/records/57f39f65123e.wav",
  "mime": "audio/wav",
  "outputBytes": 1920078,
  "outputDurationSeconds": 60.0,
  "input": {
    "format": "mov,mp4,m4a,3gp,3g2,mj2",
    "duration": 60.0,
    "sizeBytes": 4821337,
    "video": { "codec": "h264", "width": 1280, "height": 720 },
    "audio": { "codec": "aac", "sampleRate": 44100, "channels": 2 }
  },
  "billedMinutes": 1,
  "minuteEvent": "media-minute",
  "videoReencoded": null
}

A failed file has only sourceUrl, target, status: "error" and error:

{ "sourceUrl": "https://example.com/missing.mp4", "target": "mp3", "status": "error", "error": "<why it failed, e.g. the HTTP 404 from the download>" }

Field	Description
sourceUrl	The URL you sent.
target	The output format used.
status	"ok" for a converted file, "error" for a failed one (not charged).
outputKey	Key of the file in the run's default key-value store.
outputUrl	Download URL of the converted file.
mime	MIME type of the output.
outputBytes	Output file size in bytes.
outputDurationSeconds	Output duration in seconds.
input	Probed source metadata: format, duration (s), sizeBytes, video (codec, width, height, or null if none) and audio (codec, sampleRate, channels, or null if none).
billedMinutes	Started minutes of output that were charged (0 for thumbnails).
minuteEvent	The per-minute event charged for this file: media-minute or video-reencode-minute (null for thumbnails).
videoReencoded	For mp4/webm: true if the video was re-encoded, false if it was copied without re-encoding; null for other targets.
error	Error message (failed files only).

How to use it

Use from MCP (Claude, Cursor)

Add Apify's MCP server to your client and load this Actor as a tool:

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com?tools=ent-e/media-converter-ffmpeg"
    }
  }
}

The first connection signs you in to Apify with OAuth. Then ask, for example: "Convert https://example.com/call.mp4 to a 16 kHz WAV and give me the link." Because this Actor is pay-per-event, agents can also pay per run through Apify's agentic payments.

Use from n8n

Add the Apify node (@apify/n8n-nodes-apify), choose the Run an Actor and get dataset operation, select ent-e/media-converter-ffmpeg and paste your input JSON. Each returned item's outputUrl can go straight into your transcription or upload step. For long jobs, use Run an Actor, then Get Dataset Items with the run's defaultDatasetId. No community nodes? Use an HTTP Request node with the API call below.

Use from Make

Add the Apify Run an Actor module with Run synchronously set to Yes, then Get Dataset Items with the run's defaultDatasetId, and map outputUrl into the next module. Make waits at most 120 seconds for a synchronous run, so for long media use the Watch Actor Runs trigger, or an HTTP "Make a request" module calling the API below.

Use from Zapier

Use the Apify Run Actor and Fetch Results action (best for runs under 5 minutes), or Run Actor followed by Fetch Dataset Items. For long runs, run asynchronously and start the next Zap from the Finished Actor Run trigger. Map outputUrl into your next step.

Use from the API

Run it and get results in one request:

curl -X POST "https://api.apify.com/v2/acts/ent-e~media-converter-ffmpeg/run-sync-get-dataset-items?token=<APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/talk.mp4", "target": "wav16k"}'

Synchronous calls wait up to 5 minutes. For long files or big batches, start an asynchronous run (POST /v2/acts/ent-e~media-converter-ffmpeg/runs) and read the dataset when it finishes.

FAQ

Is this an FFmpeg API I can call without installing ffmpeg?

Yes. FFmpeg runs on Apify's infrastructure. You send JSON with a URL and a target and get a download link back. There is nothing to install, containerize or host.

How do I prepare audio for Whisper?

Use "target": "wav16k". The output is a mono 16 kHz WAV, the format speech-to-text models such as Whisper expect. It works with video input too, and the audio track is extracted automatically.

How do I convert a video to MP3 or extract audio from a video?

Send the video URL with "target": "mp3" (or m4a, ogg, wav, flac). The audio track becomes the output file.

Can I make a GIF or a thumbnail from a video?

Yes. gif gives an 8 fps, 360 px animated preview, and thumbnail gives a JPEG frame. Use startSeconds (with durationSeconds for GIFs), at the top level or per file, to choose the moment. GIFs are billed per output minute like audio and video, and thumbnails only per file.

Can I use it as an FFmpeg MCP tool in Claude or Cursor?

Yes, through Apify's MCP server at https://mcp.apify.com. Add ?tools=ent-e/media-converter-ffmpeg to load just this tool.

How much does video conversion cost?

It depends on whether the video has to be re-encoded. Copying already web-ready video (H.264 to mp4, VP9/VP8/AV1 to webm, at most 1080p, no trim), audio, GIFs and re-encodes at 720p or lower cost $0.005 per file plus $0.002 per output minute. Re-encoding to mp4/webm above 720p costs $0.005 per file plus $0.04 per output minute instead, because 1080p encoding needs much more compute. A 30-second 1080p re-encode is $0.045. The minuteEvent and videoReencoded fields in each result show what was charged.

What happens if a file fails?

It's returned as a row with status: "error" and an error message. Failed files are not charged, and the rest of the batch continues.

Which URLs work?

Public, directly downloadable audio or video file URLs, up to 2 GB and 3 hours by default (raise maxDurationSeconds or trim with durationSeconds for longer media).

ENT-E media toolkit

Every tool takes a public file URL, needs nothing installed, charges per file, doesn't charge for failures, and works from MCP, n8n, Make, Zapier and the API.

• Media Converter for AI Agents (FFmpeg): audio/video to MP3, Whisper-ready 16 kHz WAV, MP4, WebM, GIF or thumbnails, with trim and batch.
• Image Converter & Compressor: convert to WebP or AVIF, compress to a target file size, resize, strip EXIF/GPS.
• Podcast Loudness Normalizer: normalize to -16, -14, -23 or -24 LUFS with a true-peak ceiling, plus a before/after report.
• Video to Captions: SRT, VTT, TXT and word-level JSON from any video or audio URL, with Whisper.
• Video to Shorts: vertical 9:16 clips with burned-in animated captions for TikTok, Reels and Shorts.