AI Models for Video Production: a legal guide
Commercial use, ownership, training on your data and compliance: what the public terms of the main language, image, video and voice models say, card by card, and the rules that apply to whoever publishes AI content.
On this page you can download «AI Models for Video Production: a legal guide», free of charge. Under the form you will find its contents and the key points it explains. Fill in the form and we will send you the guide by email.
Contents
Introduction
How to read a card, the three channels and their contracts, the traffic light, and which models are covered.
Chapter 1 — Language models
Anthropic, OpenAI and Google Gemini: the models used to write, adapt and break down scripts and to write prompts for other models.
Chapter 2 — Image generation
OpenAI's image models and Google's Nano Banana: characters, environments, props and storyboards.
Chapter 3 — Video generation
Google Veo, Runway, ByteDance Seedance and HeyGen: video clips and talking avatars.
Chapter 4 — Voices and audio
ElevenLabs and Google's audio models: speech, music and sound effects.
Chapter 5 — The regulatory framework
The European Union (AI Act, GDPR, copyright), Italy, the United States, the United Kingdom, China and South Korea, with a table at a glance.
Chapter 6 — How draft2cut helps you comply
What the product records and documents, and what stays with the producer.
Key points
Three channels, three contracts
The same model reached directly, through an aggregator or run locally comes with different terms, and the API, the web app and each plan can differ again.
A traffic light per plan, not per model
Green, amber or red for commercial use, given for each row of a provider's access and plans, because the answer changes from one plan to the next.
Every statement has a source
Each card points to the provider's own documents, with the date they were consulted. What a provider does not publish is written as «not disclosed», never guessed.
The EU AI Act's disclosure duty
Deep fakes must be disclosed from 2 August 2026, with a lighter duty for evidently artistic or fictional work.
Real faces and voices
Any image or voice of an identifiable person needs a lawful basis under the GDPR; laws in the United States add consent for biometric data, voice and likeness.
A prompt alone is not authorship
In the United States and in Italy, protection depends on human creative input, not on the prompt that produced the result.
Already using draft2cut?
The guide is general information, not legal advice. It reports what the providers' public terms say on the date shown on each card: terms change, so read the current documents and, for a specific project, ask a lawyer.