# How I Built a Watermark Tool Using FFmpeg

> Building a dynamic audio and video watermarking SaaS from FFmpeg and AWS Fargate to WebAssembly in the browser.

- Author: [Berkay Bindebir](https://berkaybindebir.dev/about/)
- Published: 2024-03-28
- Topics: FFMPEG, NODEJS, AWS, WASM
- Canonical URL: https://berkaybindebir.dev/posts/how-i-built-a-watermark-tool-using-ffmpeg/

---

I needed to add audio watermarks for a client project. After looking around, I noticed plenty of tools existed for images, videos, and PDFs, but almost nothing dedicated to audio files.

So I decided to build it.

## The Power of FFmpeg

[FFmpeg](https://ffmpeg.org/) is an indispensable superpower in systems engineering. Whether extracting audio streams, mixing multitrack signals, generating GIFs, or transcode pipelines — it handles virtually anything multimedia.

The first question was: *Can FFmpeg mix two audio streams with dynamic intervals and offsets?*

```bash
ffmpeg -i input.mp3 -i watermark.mp3 -filter_complex "[0:a][1:a]amix=inputs=2:duration=first" output.mp3
```

Within two seconds, the watermark was cleanly mixed into the audio.

*Image: FFmpeg audio watermark proof of concept*

## Dynamic Watermarking Logic

Dynamic watermarking means:
- Add **T seconds** initial offset after playback starts.
- Repeat the watermark every **N seconds** across the entire duration.

To achieve this cleanly, the audio is split into distinct time zones and filter chains:

*Image: Audio timeline split into watermark zones*

*Image: FFmpeg filter graph and zone splitting process*

## Serverless Architecture on AWS

Initially, I tested running the worker inside AWS Lambda. However, long-running audio files and stream memory limits proved tricky.

I switched the background processing to **AWS Fargate**:
1. Client uploads audio and watermark configuration to an **Amazon S3** bucket.
2. S3 trigger invokes a lightweight Lambda dispatcher.
3. Lambda launches an on-demand **ECS Fargate task** running our custom Docker image with compiled FFmpeg binaries.
4. Processed file is stored in S3 with a presigned download URL.

*Image: Payload configuration for the AWS Fargate task*

*Image: AWS Fargate watermark processing architecture*

## Moving to WebAssembly (Client-Side)

While the Fargate architecture scaled effortlessly, server processing requires handling file transfers, privacy concerns, and compute costs.

Thanks to **WebAssembly** via [ffmpeg.wasm](https://github.com/ffmpegwasm/ffmpeg.wasm), the entire watermarking pipeline was ported to run directly in the user's browser.

- **Zero server upload latency**: Audio and video files never leave the user's machine.
- **$0 compute bills**: The user's CPU handles transcoding via WebAssembly worker threads.
- **Privacy by default**: Complete end-to-end client-side media manipulation.

This evolved into [Watermark.run](https://watermark.run) — supporting both audio and video watermarking in the browser and cloud.

*Image: Watermark.run browser application*