Every software rollout runs into the same problem: the guide is written, the screenshots are old, and new colleagues still ask. A short video explains a flow faster than three pages of text. It only has to stay current.
How I work: with source · with approval · in your system.
Starting point
Training videos are recorded when a version is finished. Then the interface changes, and the video shows a button that no longer exists. Re-recording costs time, so the old video stays and explains the wrong system.
What I do
I build videos from a pipeline: script, shot script, capture, voice, pauses, music, mix, subtitles, validation, render. Each stage has a defined input and output. When the software changes, the affected capture is replaced and re-rendered instead of repeating the whole production.
The voice is cloned, but only with written consent: my own or a speaker’s who has agreed. The video carries a note that the voice is AI-generated. Subtitles come out as VTT. The pipeline provides for watermarking (AudioSeal) and a C2PA signature.
One limitation belongs here: watermark and signature are off by default and have to be switched on when rendering. The pipeline can do it, it does not enforce it. The duty to label under Art. 50 of the AI Act sits with whoever publishes, not with the tool.
Result
A video series with subtitles and the pipeline configuration that reproduces it. Plus a validation run that finds missing files and broken timing before the video goes live.
Example
The pipeline is built as a package with 15 modules and has its own validation command that runs the chain before rendering. One product video has come out of it: myPlace Parking, 97 seconds, explaining the app in its flow.
The video sits as a file on this server, not on a platform. The voice is a clone and labelled as AI-generated; the note sits directly with the video.
Limit
This is explainer material for systems and processes, not advertising or social media production. A voice without written consent is out. Labelling duties under Art. 50 of the AI Act apply from 2 August 2026; the pipeline holds watermark and signature ready, switching them on and checking them happens at publication. That responsibility stays with the client.
How it runs
- Define topic and audience, write the script
- Create shot script and capture
- Generate voice and subtitles, switch the AI note and watermark on
- Mix, validate, render
- Publish, re-render on changes
Common questions
Can the voice be my own?
Yes, with written consent. The same applies to a client’s voice; without the document there is no clone.
How long does one video take?
A short explainer of two to three minutes takes a few days depending on capture and script. The series is the real effort, not the single video.
Where are the videos hosted?
On your own or the client’s server, as a file with subtitles. A YouTube embed is not the default for data protection reasons.
What happens on a software update?
The affected scene is replaced, the rest stays. That is exactly what the pipeline is built for.