Projects / Internal automation
I shoot.
The process
takes shape.
A practical exploration of applied AI: turning raw video and directions spoken during filming into an edit to review and publish.
The problem
Filming ends.
The work continues.
Cuts, framing, subtitles, effects and audio levels: a short video contains many small steps. Repeating them by hand makes editing a more fragmented task than it might seem.
I built a process where directions spoken while recording can also become working instructions: “let’s do that again”, “zoom in here”, “add the effect here”.
The workflow
From spoken words
to a sequence.
Analyse the raw footage
The transcript and reference frames help reconstruct the content and distinguish the footage from the editing directions.
Prepare the edit
AI translates the directions into a configuration: cuts, framing, text and effects. The transcript is corrected when speech recognition makes a mistake.
Generate and review
The pipeline produces the video and previews. Reviewing the result remains an explicit stage: publishing is a separate step.
Behind the process
Several tools.
One coordinated workflow.
The project combines Python, Whisper transcription, FFmpeg processing and image analysis tools. AI interprets the directions and prepares the work; code carries out the transformations.
The interesting part is how the stages connect: material moves from analysis to editing through a procedure that can be repeated and checked.
What it shows
Automation starts
with a specific process.
This pipeline was built for my own video workflow. It is not a universal service that edits any content without supervision. It shows how instructions, AI and specialist tools can work together on a real task.
The next step
What work do you
keep repeating?
We can study the steps and work out which ones are worth connecting, automating or simplifying.