Video generation with AI: the models and how to use them well

What it is

A video model creates a short clip from a written description, a still image or a reference clip; some also create sound, including speech. Avatar tools turn a script into a video of a digital presenter speaking, often in many languages.

Think of a fast film crew that forgets each scene unless you remind it. You are the director.

Main tools, as their makers describe them (as of November 2026; check the maker's page):

Limits shared by all of them: short clips, uneven text and hands, credit-based plans; check each maker's terms for commercial use and likeness.

OpenAI's Sora app was discontinued in April 2026, according to OpenAI's help centre.

Why it matters

• Training clips, explainers and announcements without a studio.
• Each clip can take minutes and uses plan credits, so careless iteration wastes time and budget.
• Realistic video of real people can mislead, damage trust or break the law.

How to do it

1. Check which video tools your organisation approves; never upload confidential material or anyone's face or voice unless allowed (see guide 10).
2. Write a short script and a shot list: one line per shot with subject, action, setting, camera (close-up, wide, slow pan), lighting and sound.
3. Generate one shot at a time; clips last seconds, not minutes, so plan to join them.
4. Keep consistency: reuse the same reference images and repeat key descriptions word for word in every shot.
5. Add sound deliberately: generated sound, a recorded voice-over or music you have rights to (see guide 38 for AI voice).
6. Review: a named person watches every frame for odd hands or faces, wrong text or logos and anything misleading, then approves.
7. Label the video as AI-generated and keep a record of the tool, prompts and consents.

Example

A team needs a one-minute clip explaining how to book a meeting room.

Before: "make a video about booking rooms" gives 8 seconds of random office footage. After: a shot list.

Style for every shot: bright modern office, flat soft lighting,
calm pace, no logos, no readable screens.
Shot 1 (wide): an employee walks to a meeting room door with a
tablet beside it. Sound: quiet office background.
Shot 2 (close-up): a finger taps a green "Book" button on the
tablet. Sound: soft click.
Shot 3 (medium): two colleagues sit down in the room and open a
laptop. Sound: quiet office background.

The team adds a voice-over and on-screen steps in their editor, checks every frame, and opens with "This video contains AI-generated footage".

Common mistakes

• Using a real person's face or voice without consent. Fix: get written consent; avatar makers such as Synthesia and HeyGen say they require on-camera consent or identity checks for custom avatars.
• Trusting text and screens in the video. Fix: add text afterwards in an editor.
• Publishing without a label. Fix: say clearly that the video is AI-generated.

Try it this week

1. Pick an approved tool and a non-confidential topic with no real people (2 minutes).
2. Write a three-shot list using the format above (5 minutes).
3. Generate shot 1, then change one detail and generate again (5 minutes).
4. Watch both versions frame by frame and note every error (3 minutes).

Going further

• Never make a deepfake of a real person without their consent and a clear purpose.
• EU AI Act, Article 50 (European Union; general information, not legal advice): from 2 August 2026, organisations that use AI to create deepfake content in their work must disclose it clearly, at the latest when people first see it; makers of the tools must mark output in a machine-readable way (tools already on sale have until 2 December 2026). Evidently artistic or satirical work needs only a disclosure that does not spoil it. Ask your legal team.
• Veo videos carry Google's SynthID watermark; it does not replace your own visible label.

Key words

• Video model: an AI that creates video clips from text, images or reference clips.
• Avatar: a digital presenter that speaks a script on screen.
• Shot list: a plan of each scene, with camera, action and sound.
• Deepfake: realistic synthetic media of real people or events that could be mistaken for genuine.

Sources

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (281 KB), or the complete AI Field Guide (2.5 MB).