An AI clone video is made in seven steps: consent and source recording, a written script, a person's approval, voice and avatar generation, editing, a final review and publishing. The clone speaks the words; it does not decide what is said. The steps that decide quality are mostly human ones, before and after the generation.
- Step 1Consent and sourceWritten consent. Clean video and voice recordings of the person being cloned.Client
- Step 2ScriptA draft written from the person's own notes, ideas and approved facts.AI drafts
- Step 3ApprovalThe person reads and approves every script before anything is generated.Person approves
- Step 4Voice and avatarThe approved script is turned into speech and video on the platform the clone lives in.Automation
- Step 5EditPacing, captions, graphics and on-screen text are added and checked against the script.QS
- Step 6ReviewA person watches the finished video for errors, tone and claims.Person approves
- Step 7PublishPosted with any disclosure the channel or the client wants. Results logged.QS
In words: First the person gives written consent and records source video and voice. A script is drafted from their notes and the person approves it. The platform turns the approved script into speech and video. An editor adds pacing, captions and graphics. The person reviews the finished video, and only then is it published and logged.
1. Consent and source recordings
A clone is built from the real person's footage and voice, and only with their written consent. Good source material is simple: steady lighting, a quiet room, the camera at eye level, and a few minutes of natural speech. Some platforms require the person to verify their own likeness or voice before a clone can be created. The consent should say what the clone may be used for. Safety and consent are covered in more detail here.
2. The script
Scripts are where the person's actual ideas go in. AI can draft from notes, past emails, talks or a recorded conversation, but the facts and opinions belong to the person. We keep a short list of approved claims and avoid inventing statistics, testimonials or results. A script written for speaking is shorter and plainer than one written for reading.
3. Approval before generation
The person approves each script before it is turned into video. This is the main control. Once approved, the clone delivers exactly those words. Anything involving price, medical, legal or financial advice, or promises about results should be written or reviewed by the person with the expertise.
4. Voice and avatar generation
The approved script is rendered by the platform that holds the avatar and the voice model. The avatar and the voice live inside that platform; they are not files you can usually export, so the portability plan is to keep the source recordings, scripts and prompts, which can rebuild a clone elsewhere. Check pronunciation of names and terms, and expect some takes to need regenerating.
5. Editing
Editing is what separates a usable video from a demo. It includes cutting dead air, adding captions and on-screen text, matching brand colors and fonts, and checking that every caption matches the approved script. Edits that change the meaning of the words send the script back for approval.
6. Final review
A person watches the finished video once, start to finish, for errors the script check cannot catch: odd emphasis, lip-sync glitches, a wrong word in the captions, or a claim that no longer fits. If the channel expects disclosure of AI-generated content, the review step is where it is confirmed.
7. Publish and learn
Publishing includes the title, description, thumbnail and where else the video will be reused. Record what was published, where and when, so you can see which topics earn replies and which do not. This turns one recording session into a repeatable content habit.
Who is responsible for what
| Step | The person being cloned | QS Digital | AI tools |
|---|---|---|---|
| Consent and source | Gives consent, records, completes any platform verification | Briefs the recording and checks quality | None |
| Script | Supplies ideas and approves facts | Structures and edits | Draft from the person's notes |
| Approval | Approves every script | Holds generation until approved | None |
| Voice and avatar | Reviews samples | Runs the generation | Platform renders speech and video |
| Edit | Gives feedback | Edits and checks against the script | Assists with captions and cutting |
| Review and publish | Final approval | Publishes and logs | None |
A hypothetical week
Imagine a consultant who wants one short video a week. Once, she records source footage and voice and gives written consent. Each week she sends a few bullet points, a script is drafted and she approves it, the video is generated and edited, she watches the finished version and it is posted. Her time each week goes to the bullets and two reviews, not to filming. This is an invented scenario to show the pattern, not a result from a client.
What this workflow does not do
- It does not decide what you should say. The words and claims are yours.
- It does not remove the need for review. Every video is checked by a person before publishing.
- It does not make a clone portable. Avatar and voice models stay with the platform that built them.
- It does not suit work that depends on live conversation or unscripted judgement.
Next step
The AI clones page explains what QS Digital builds and what we need from you. For platform choices see HeyGen vs Synthesia vs D-ID, and for costs, how much an AI clone costs.