Images, Video and Voice with AI
A practical, vendor-neutral course for marketers, creators and communicators on using AI for images, video and voice: how generative media works and where it fails, prompting and editing, voice and consent, copyright and likeness uncertainty, EU AI Act transparency and C2PA labelling, and a personal policy against deepfake harm. Five modules, each with a worked example, and a 10-question final quiz. Based on cited official pages. Not an accredited qualification and not legal advice.
What you will learn
- Explain and apply: Explain in plain words how generative image, video and voice systems work, and name their typical limits.
- Explain and apply: Write and refine prompts for images and video by describing subject, style and framing, and keep a campaign consistent.
- Explain and apply: Use text-to-speech responsibly and apply the consent-first rule to any voice cloning.
- Explain and apply: Check copyright and likeness rights, and apply disclosure and labelling, including EU AI Act Article 50 and C2PA Content Credentials.
- Explain and apply: Recognise deepfake harms, use verification habits, report abuse and write a personal policy for responsible use.
Modules
- How generative media works, and where it breaks
- Prompting and editing images and video
- Voice: text-to-speech and voice cloning
- Rights, consent and labelling
- Deepfakes: harms, verification and a personal policy
Course facts
- Level
- Foundation
- Estimated study time
- about 70 minutes
- Final assessment
- 10 questions · Pass mark 80%
- Available in
- English, Français, العربية
- Price
- Free. Optional verifiable certificate: $7
- Last updated
- Published by
- AMAADOR ACADEMY
Who this course is for
Marketers, content creators, communicators and small-team managers who want to use AI image, video and voice tools responsibly.
Prerequisites
No technical background. Basic familiarity with creating social posts, slides or short videos is enough. The course is vendor-neutral and does not rank any product.
Course content
1. How generative media works, and where it breaks
Generative image and video tools are trained on very large collections of examples, and they learn statistical patterns: which shapes, textures, colours and words tend to go together. They do not store a scene and replay it, and they do not understand it the way a person does. When you type a description, the system turns your words into numbers that steer the output, then produces something that fits those patterns. That is why results can look convincing and still be wrong in ways a human would never choose.
Many image and video tools rely on the diffusion idea, in plain words: during training the system learns to remove noise from pictures that were deliberately made fuzzy. At generation time it starts from random noise and removes it step by step, nudged each time by your description, until a picture emerges. Video adds a time dimension, so the system must also keep objects steady from one frame to the next. Other designs use transformer models, which predict the next piece of an image, sound or clip from the pieces before it, much as text models predict the next word. Vendors differ, so treat these as a mental model, not a specification.
2. Prompting and editing images and video
A good media prompt reads like a brief to a photographer. Name the subject first (who or what, doing what), then the setting, then the style (documentary photo, flat illustration, watercolour), then the framing and light (close-up, wide shot, low angle, soft morning light). Add what you want to avoid, such as text in the image or a busy background. Vague prompts get generic results; over-long prompts get ignored in parts. Four or five precise clauses usually beat a paragraph of adjectives.
Iteration is the real skill. Change one thing at a time, so you can tell which words caused which effect. Keep the versions you like, note the settings, and use editing features for local fixes: replacing an object, extending a background, or retouching a region while the rest stays fixed. For video, start with a short clip and a clear camera instruction (slow push-in, static shot), because motion is where errors multiply. Expect several attempts per usable result and budget time for them.
Consistency across a campaign is hard, because each generation starts afresh. Write down a short style sheet: palette, lens feel, lighting, character description. Reuse the same wording, and where your tool offers reference images or fixed settings, use them. Then check each new image against the previous ones. Never ask a tool to imitate a named living artist or to reproduce a recognisable brand or person without permission; that creates legal and ethical problems covered in later modules.
3. Voice: text-to-speech and voice cloning
Text-to-speech turns written text into spoken audio. Modern systems can control pace, emphasis, accent and emotion, and are useful for narration, accessibility, language versions and quick drafts of scripts. Quality depends on the language and on the text: numbers, names, abbreviations and mixed languages are common trouble spots, so always listen to the whole file with the script in hand and fix the text, not only the audio.
Voice cloning goes further: a system learns the characteristics of a particular person’s voice from recordings and then speaks new sentences in it. A voice is personal data in many legal systems, it can identify a person, and it carries reputation. The rule for this course is consent first: clone only your own voice, or the voice of someone who has agreed in writing, knowing exactly what the clone will be used for, for how long, and how to withdraw. Consent for one project is not consent for every future use.
Cloned voices are also a fraud tool. Scammers have imitated relatives and managers to request urgent payments, and a convincing voice proves nothing about who is speaking. Organisations should agree a second channel for sensitive requests, such as a call-back on a known number or a code word. Check the terms of your tool as well: they set what you may upload, what the provider may keep, and whether recordings are used to improve their service.
4. Rights, consent and labelling
Copyright and AI is an area where the law is still moving and differs by country. Questions include whether training on protected works is lawful, whether AI-assisted output can be protected, and how much human creative control matters. Official bodies such as the US Copyright Office and WIPO publish analysis and updates, but nothing here is legal advice. The safe habit is to check your jurisdiction, read the tool’s terms on commercial use, avoid prompts that copy a living artist or a protected character, and keep records of your human creative contribution.
Likeness and personality rights, which differ by country, protect how a real person’s face, voice and name are used, and data-protection law may apply too. Use of a recognisable person without permission can be unlawful or harmful even if the image is entirely generated, and it matters most for advertising, endorsements and anything suggesting that the person said or did something. Get written releases from anyone identifiable, and treat children, private individuals and the dead with extra caution.
5. Deepfakes: harms, verification and a personal policy
Deepfakes are synthetic or manipulated images, video or audio that depict someone doing or saying what they did not. Harms are real and varied: fraud and impersonation, non-consensual intimate imagery, harassment, political disinformation, damage to reputations, and the quieter harm of making genuine evidence easy to dismiss. Creators and marketers should recognise that their tools and skills are the same ones used for these abuses.
Spotting fakes by eye is unreliable. Old tips such as odd blinking or warped hands age quickly as tools improve, and automatic detectors make mistakes in both directions. Do not rely on a single visual check. Verification habits work better: find the original source, look for the same material from independent outlets, search for the earliest upload, check the account that shared it, inspect available provenance data such as Content Credentials, and ask the person concerned through a channel you already trust. Slow down most when content triggers strong emotion or urgency.
When you meet harmful synthetic content, do not share it, even to condemn it. Report it on the platform, keep a record of the link and date, and tell the person depicted. For intimate imagery or threats, contact the relevant authority or a specialist support service in your country. For your own work, write a personal policy of a few lines: what you will make, what you will refuse, how you obtain consent, how you label, and what you do if you make a mistake. A written policy is easier to follow under deadline pressure.
The full lesson continues in the interactive course on this page.
Sources and references
- C2PA
- European Commission AI Act page
- Regulation (EU) 2024/1689 (EUR-Lex)
- US Copyright Office: AI
- WIPO: frontier technologies and IP
- UK ICO: AI guidance
Education only. It is not legal, financial or professional advice and not an accredited qualification. Laws and tools change and differ by country; official sources are summarised as read on the access date, so check the current text. Figures in worked examples are illustrative.