AI AI
AI tools are used in video for everything from transcribing interviews to generating whole shots. We group them by what they do to the recording — analyse it, alter it, fill it in or create it — because that is what determines both the risk and the obligations.
- video generation
- enhancement
- restoration
- upscaling
- AI-assisted editing
- AI workflow
What this section answers
- When does AI content have to be labelled?
- Can upscaling be trusted?
- How can AI be brought into a workflow without losing control?
Articles and tools
Quick reference
| category | effect on the recording | watch out for |
|---|---|---|
| analysis and editing (transcription, search) | none | transcription errors, data confidentiality |
| enhancement (noise, sharpness, audio) | changes pixels and samples | artificial detail, ‘waxy’ faces |
| restoration and colourisation | fills in by estimation | it is not historical fact |
| upscaling and frame interpolation | invents detail and in-between frames | hallucinated textures, distortion |
| generation | creates new content | deepfakes, consent, labelling (Art. 50 AI Act) |
Terms
- Upscaling super-resolution
- Increasing resolution. Classic methods interpolate pixels; machine-learning models estimate detail, which means filling in what was never in the recording.
- Deepfake
- Under the EU AI regulation, AI-generated or manipulated image, audio or video content that resembles existing persons, places or events and would falsely appear to be authentic.
- Frame interpolation
- Calculating in-between frames to raise the frame rate or create slow motion. The result is an estimate of the motion, not a recorded reality.
In preparation
Topics we are working on. They are not published yet — we would rather have fewer pages, but verified ones.
- AI transcription and subtitles: quality control
- Upscaling archive footage: comparing methods on the same material
- Generated shots in previsualisation
- Labelling AI content in practice
Sources for the tables
Checked .