Motion Production: Media Pipeline
The tooling that prepares motion work for the web: H.264 and VP9 encodes with faststart, silent hover previews, poster extraction and storyboard contact sheets. 28 tests invoke ffmpeg for real. No finished film is included.
This is a working demonstration, not a client engagement. The brand and data are fictional.

Background
Motion work needs a repeatable path from a master file to the files a website actually serves. This is that path, written as two command-line tools.
The challenge
The details that matter in web video are easy to get wrong and invisible until someone complains: a file without faststart will not begin playing until it has largely downloaded, a hover preview with audio is a bug, and an odd pixel dimension makes VP9 fail outright.
Objectives
- Take a master file to the files a website actually serves, repeatably.
- Get the details right that are invisible until someone complains: faststart, silent previews, even dimensions.
- Verify the output by probing it rather than trusting the flags.
My role
Sole author
- Wrote the media preparation tool: probes the source, encodes H.264 with faststart and VP9, extracts a poster a quarter of the way in, and builds a silent 6-second hover preview
- Made it refuse to upscale past the source, keep dimensions even for both codecs, and drop the audio track when the source is silent rather than failing
- Wrote the contact-sheet tool: evenly spaced frames inset from both ends, so slates and fade-ups are skipped
- Wrote 28 tests that run ffmpeg and ffprobe against a generated clip and assert the real output properties
- Wrote the shot-list and case-study templates
Technical approach
Encode decisions are made from a real probe of the source rather than from assumptions, and the tests invoke ffmpeg rather than mocking it, so if an encoder flag is wrong, the test sees the wrong output.
What is implemented
- H.264 encode with +faststart and a VP9 alternative
- Silent six-second hover preview, with the audio track dropped rather than failing on silent sources
- Poster extraction a quarter of the way in, with dimensions reported back
- Contact sheets from evenly spaced frames, inset from both ends to skip slates and fade-ups
- A JSON snippet generated for pasting into the project data file
Process
- Wrote the probe and the argument builders for each output.
- Wrote tests that generate a clip with ffmpeg, run the pipeline over it and probe the results: faststart present, preview silent, poster dimensions correct, no upscaling.
- Built the contact-sheet tool and tested the frame timestamps and grid geometry.
Engineering and creative decisions
Encode decisions come from a real probe
The tool probes the source and refuses to upscale past it, keeps dimensions even for VP9, and drops audio when the source is silent. None of that is assumed from the filename.
The tests run ffmpeg
They generate a clip, run the pipeline over it and probe the results. A mocked encoder would have passed with the wrong flags; probing the output cannot.
Frames are inset from both ends
A contact sheet sampled from 0% to 100% is mostly slate and fade-up. Insetting gives frames that represent the content.
What was delivered
Two tools that take a master file and produce everything the portfolio needs, including the JSON snippet to paste into this project data file with the poster's real dimensions filled in.
Results
- 28 tests pass by running ffmpeg and ffprobe for real.
- **No finished video is in this repository.** Footage and edits have not been supplied, and nothing has been generated to stand in for them. The cover image here is a placeholder, labelled as sample media.
- What is verified is the pipeline, on a generated test clip. What is not shown is any motion work, because there is none to show yet.
These are tested capabilities of the demonstration, not commercial outcomes.

