Short-form video works for a photo studio, but only if it shows up consistently. Selecting clips, writing captions, and preparing posts every single day doesn't scale with a two-person company. So I built a pipeline that does it every morning, with no human input.
10
videos generated every morning
1.14M
views on the resulting account in 30 days
Solo
designed and built alone
What it does
- Selects source footage
- Generates captions
- Renders 10 short-form videos
- Prepares them for distribution
How it's built
Python for the pipeline, LLM APIs for the captions, video processing for rendering, running on AWS.
# The shape of the daily run.def morning_run(): clips = select_footage() for clip in clips[:10]: caption = write_caption(clip) render_video(clip, caption) prepare_for_distribution()None of the individual steps is unusual. What makes it useful is that nothing in the chain waits for a person, so it doesn't slip on a busy day.
Designing for unattended
A pipeline that needs someone to start it is a chore with extra steps. The design goal was a daily run that completes on its own: pick, caption, render, prepare.
Where the LLM fits
The language model handles the part that used to need someone to write something every day: the captions.
Result
Ten videos come out every morning, unattended, and the studio account they feed drew 1.14M views in the last 30 days (Sep 2026).
For a client, the transferable part isn't video. It's taking a task someone does by hand every day and making the daily run happen without them.