A HeyGen custom avatar, currently called a Digital Twin, is trained from a real presenter's footage. It carries a recognizable face, movements, and possibly voice, so consent, approved use, and removal are production requirements.

By the end, you will have an avatar test sheet, a consent record, and one short approved clip. You will record suitable footage, create the Twin, pair it with a script and voice, render a test, and define what happens if permission is withdrawn.

// concept

Decide Between a Stock Avatar and a Digital Twin

Use a stock avatar to test scripts and layouts without processing a real likeness. Create a Twin only when the presenter understands every intended use.

Decision	Stock avatar	Custom Digital Twin
Setup	Select from Public Avatars	Upload footage and pass consent verification
Best fit	Drafts and generic explainers	Approved spokesperson or recurring training
Main risk	Generic brand fit	Impersonation or use beyond permission
Stop-use action	Replace it in the project	Stop renders, remove agreed uses, delete the Twin/looks

Avoid a custom avatar for medical advice, bank instructions, political messages, emergency announcements, or endorsements unless the presenter reviews that exact use and relevant rules are checked.

Create this record before uploading footage. HeyGen's verification video is separate; it does not replace your agreement with the presenter.

// prompt — copy me11 lines

CUSTOM AVATAR CONSENT RECORD
Presenter: [name/internal ID — no CNIC copy]
Workspace owner: [person/company]
Purpose and channels: [exact uses]
Languages and voice: [public voice or consented clone]
Approval: [who reviews each script and final render]
Permission period: [start plus review/expiry date]
Prohibited uses: [politics, testimonials, ads, etc.]
Withdrawal: [contact, stop-render and removal steps]
Evidence location: [restricted folder]
Presenter confirmation/date: [documented approval]

If permission is withdrawn, pause renders, list published clips, follow the removal process, and log each action. In Avatars, use the Twin's current manage/overflow control to delete it and its looks. Confirm the control in HeyGen Help; labels can move.

// concept

Capture Clean, Authorized Training Footage

HeyGen's detailed filming guidance currently recommends an uninterrupted speaking take. Immediately before filming, confirm the current length, format, and plan requirements on the official creation page.

Record one continuous take with no edits. The current detailed guide says at least two minutes and recommends up to five; follow the requirement displayed in your account if it differs.
Use at least 1080p and 30 fps when available. Match orientation to the destination.
Light the face evenly. Avoid backlighting, harsh shadows, filters, green screens, and moving backgrounds.
Keep the whole face visible, look into the lens, and keep natural gestures inside the frame.
Record clear speech in a quiet room, even if another authorized voice will be used later.
Inspect a short test for focus, echo, lip-to-audio sync, clothing noise, and cropped hands before recording the final take.

Keep consent evidence away from public edit assets:

// prompt — copy me6 lines

avatar_p01/
  restricted/p01_consent-record.pdf
  restricted/p01_platform-consent.mp4
  source/p01_training-take_v01.mp4
  tests/p01_avatar-test_v01.mp4
  approvals/p01_test-approval.txt

As of the current interface, go to Avatars → Create New Avatar → Digital Twin, then record or upload. The depicted person must personally read HeyGen's displayed consent text. A remote presenter can use its QR-phone flow. Never use an old clip, screen replay, or AI-generated consent.

// concept

Pair the Twin With a Script and Render a Test

Create the smallest clip that reveals problems before spending more generation quota:

Open Create → Create a Video → Create in AI Studio. Confirm the current route in HeyGen Help if the UI has changed.
In the right panel, choose Avatar → My Avatars, then select the new Twin.
Open Script, paste 20–30 seconds of approved text, and choose an authorized public or consented voice. Use Play beside the voice to check pronunciation.
Choose the real destination's orientation. Keep scene one plain so reviewers can judge lips, eyes, hands, and delivery.
Submit for generation. HeyGen previews audio during editing, but the animated presenter requires a generated render for visual review.
Review on a phone. Where a viewer might assume a live recording, add “AI-generated presenter used with permission.”

Quotas, access, resolution, credits, and watermark rules can change by region and plan. Verify official pricing and in-account usage before promising an export. HeyGen currently says free outputs carry a watermark and eligible paid plans can remove it. Use a watermarked result only as a test.

Record the decision in an avatar test sheet:

Field	Entry
File/date	Exact export filename and generation date
Inputs	Script version, Twin, and voice
Destination	For example, 9:16 social draft
Checks	Likeness, lips, eyes, hands, pronunciation, disclosure
Decision	Reject / revise / approved for one named use
Reviewers	Presenter and content owner, with dates

// worked_example

Worked Example

The first AI Studio input is:

// prompt — copy me3 lines

Assalam-o-alaikum. Welcome to the sample orientation.
In this short lesson, we will set up your practice folder and open the first exercise.
This presenter is AI-generated with my permission.

The representative first render is rejected: “practice folder” sounds rushed, the eyes drift from off-axis notes, and a hand leaves the frame. A longer script cannot repair the source.

The producer re-records with notes beside the lens and quieter gestures, then uses:

// prompt — copy me5 lines

Assalam-o-alaikum. [add a HeyGen pause here]
Welcome to the sample orientation.
First, we will create your practice folder.
Then we will open exercise one.
This presenter is AI-generated, with my permission.

They replace the bracketed instruction with HeyGen's Pause control, preview, render, and log the file. The owner phone-checks it and approves only this use—not later scripts.

// failure_cases

Failure Cases to Diagnose

7 cases to diagnose

Identity mismatch
HeyGen rejects consent because the people differ. Re-record the same person in similar lighting using the live text.
Wandering eyes
The source speaker read off-axis. Put notes beside the lens and re-record.
Unstable mouth timing
Echo, background speech, or edits polluted training. Use one quiet, uninterrupted take.
Warped hands
Gestures left a tight crop or covered the face. Widen the frame and slow the gestures.
Missing voice rights
The face is authorized but a clone is not. Use a public voice until voice consent is explicit.
False live endorsement
An old Twin states a new offer the presenter never reviewed. Require script approval and disclose synthesis.
Revoked likeness still renders
Freeze the Twin, remove scheduled uses, follow the takedown log, then delete the Twin/looks.

// pakistan_angle

Pakistan Angle

Convert subscriptions into PKR using the card's current rate plus bank charges; verify both before client approval. International card access may block checkout. Start with a stock avatar or watermarked free test when permitted, and pay after format approval. Do not stack video, voice, and editing subscriptions before finding the bottleneck.

Record locally during a load-shedding-safe window and keep the original on the phone until cloud upload is verified. Upload large training footage over stable off-peak Wi-Fi; use HeyGen's QR flow for remote consent. Explain your private agreement in Urdu or Roman Urdu if useful, but the presenter must read HeyGen's displayed consent text exactly. Do not collect a CNIC image, and do not treat “theek hai” in a group chat as unrestricted commercial permission.

// hands_on

Hands-On Exercise

7 steps

Write the allowed purpose, channels, voice, review period, prohibited uses, and withdrawal process in the consent record.
Obtain and store the presenter's documented approval in a restricted folder.
Check HeyGen's current requirements; inspect a test take, then record the final continuous footage.
Create the Twin and have the presenter complete HeyGen's consent step personally.
Add a 20–30 second factual script and authorized voice in AI Studio; preview pronunciation and generate one draft.
Fill the test sheet. Have the presenter and content owner review likeness, delivery, disclosure, and scope.
Save the first approved clip with its script and approval note. “Done” means all three artifacts exist and the approval names this use.

// completion_rubric

Completion Rubric

6 checks — tick as you verify

0/6

// sources

Sources

4 official sources — check every claim yourself

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4

1 / 4 · diagnose
Your work shows this failure mode: “Wandering eyes.” The lesson describes it like this: “The source speaker read off-axis.” What does the lesson tell you to do about it?