Content Library (LinkedIn Growth Mastery)·LinkedIn Growth Mastery - AI Video Content Creation·23:29

Record Once, Post Forever: Shooting the 3-5 Minute Raw Video Your AI Clone Trains On, and the Teleprompter Hack

Uttam Gupta Trainer - former director of growth at Growth School, founder of Agent Valley (agentvalley.ai)

The short version

The five-stage AI video workflow (idea -> script -> video -> audio -> edit); today is video, audio and clips. Everything starts with one 3-5 minute talking-head recording that HeyGen learns your face, lips and gestures from (0:00-0:03).
The recording spec: phone is fine, plain static background (no green screen needed), still camera, daylight, neutral expression, natural hands that never cover the mouth, solid clothing, silence. What you say does not matter (0:03-0:12).
Teleprompter hack: a free phone app or web prompter scrolls a script over the front camera - 'that's how all the creators record' (0:13-0:23).

At a glance, three clicks deep

Skim here first: the closed row is the glance, open is the study card with the key points and timestamps, and the ↓ link drops to that concept's full write-up below.

01The raw-video spec the clone trains on3-5 min, still, plain, neutral, quiet;›

3-5 min, still, plain, neutral, quiet; content irrelevant.

Spec (0:03-0:10)

Avoid list and tips (0:10-0:12)

Orientation (0:12)

↓ Full write-up of this concept

02The teleprompter hackApp scrolls the script over the camera;›

App scrolls the script over the camera; phone first.

Phone app demo (0:16-0:20)

Web prompter + two-phone rig (0:20-0:23)

↓ Full write-up of this concept

The concepts in full

The raw-video spec the clone trains on

Smile too much in the training video and every serious script comes out grinning.

3-5 minutes; phone camera; static plain background - a white wall in good sunlight beats a bad green screen; tripod or improvised stand; front-facing; soft natural light; neutral expression; natural, generic hand gestures not tied to words; hands and mic never cover the mouth; solid clothing without patterns; no background noise. Avoid overexposure, moving or blurry backgrounds, fast head or hand movement. Test-record 10-15 seconds first; landscape for long-form, portrait for reels. The content is irrelevant - HeyGen analyses hand, eye and lip movement, so read anything (a ChatGPT script will do).

Why it matters

Every flaw in this one recording is replicated in every video the clone ever makes.

This idea elsewherecontinuesThe footage goes into HeyGen next

The teleprompter hack

Nobody memorises three minutes. The script scrolls behind the lens.

PROCEDURE, phone: install the free 'Teleprompter' app (Play Store / App Store) -> Add Script -> paste (ChatGPT-written is fine) -> set scroll speed and font size -> record from the front camera while reading -> stop and export. Desktop: open a web prompter (heard as 'Q Prompter'), paste the script, start, allow webcam, record. Trade-offs: front camera is lower quality but you can read; a two-phone rig (one prompting, one filming on the back camera) is the upgrade. Start with the phone.

Why it matters

Removes the one blocker that stops most people ever recording.

Tools referenced

Tool	Coverage	Context
Teleprompter	demonstrated	Free mobile app
QuePrompter	demonstrated	Web teleprompter (as heard)
HeyGen	explained	Avatar library shown as proof
ChatGPT	mentioned	Generate a script to read

Action items

Extraction notes

This page was built from an auto-generated transcript, which garbles product and people's names. Those were corrected silently in everything above and logged here for transparency. The warnings flag claims that were true on the recording day but change fast.

Transcript corrections applied

The transcript says	The trainer actually means
Q Prompter / queue prompter	a web teleprompter (likely QuePrompter)
Lignin / lean in	LinkedIn
GrowthCool / GrowSchool	Growth School
agent value	Agent Valley
Sutta / Utham / Autumn	Uttam (trainer)

True on recording day — verify before relying

LinkedIn feature names, limits and algorithm behaviour described here are as of the May 2026 recording