Clone yourself
You probably need captions more than a clone. Seven conversion points is what a voice adds on top of a written offer. How to add your voice or face without filming every week, and when to skip both.
Closed captions with a written offer: about +80% conversion lift. Voiceover plus the same written offer: about +87%. Seven points is what your voice is worth on top of text. For most app demos, screen plus captions is enough. Faceless text-forward styles lead Motion’s 2026 hit-rate table; a face over the UI sits near the bottom. If you want voice without weekly recording, use speech-to-speech that keeps your timing. A face clone of you beats stock AI avatars in practitioner teardowns, but neither beats captions-first for product demos.
Do I need a voice at all?
Barely. On-screen text does nearly all the work. Platform-reported lifts put captions and a written offer around +80% conversion, and voiceover plus that written offer around +87%. The gap is seven points.
That is the permission slip. Silent proof and many demos ship with no voice. Tutorials and denser explainers are where the seven points are worth a recording session. Flat perfect TTS without your timing is what sounds like a bot; cadence is the tell, not timbre.
Does faceless actually beat a face on camera?
Often yes for performance ads. Motion’s Creative Benchmarks 2026 (550,000+ Meta ads, about $1.3B spend) ranked visual styles by hit rate: the ad spent at least 10× the account’s median. Text-forward styles lead. Founder face is still good. Greenscreen face over the UI is not.
Ten of twenty-eight styles shown; greenscreen is near the floor, not merely last in this list. Text-only assets overall: 11.6% (mostly static and text-image, not pure video). High production: 6.97%. Baseline ~5%. Founder face beats hired-creator UGC (7.56%) and high production. Faceless still leads. Never composite a face over the app UI.
What should I do instead of filming every week?
Default to screen plus burned-in captions. Add your real voice in a batch when you want the extra lift. Clone the voice with speech-to-speech only if you dislike hearing yourself. Clone the face only if founder presence is a brand asset you refuse to film weekly.
Screen + captions (default)
Highest leverage per minute. Matches the top of the Motion hit-rate table. Zero weekly performance cost.
Your real voice, batched
One session, ten takes, keep the flubs. Cut dead air in Descript. Best path when you want the ~7 point lift without a tool stack.
Speech-to-speech voice clone
You speak; the model keeps timing, breaths and stumbles and swaps timbre. ElevenLabs and equivalents. Use over screen recordings so there is no lip-sync problem.
Face clone of you, filmed once
Two minutes of source, then talking-head beats without weekly camera time. Use before or after the demo, never as a bubble over the UI. Disclose when the platform requires it.
Not yet
- A face clone before one faceless format has worked as an ad.
- Any face composited over the product UI. The screen needs the gaze.
- Stock AI actors reading "I tried this app" scripts. That first-person claim is what they cannot honestly make.
Is a clone of me better than a stock AI avatar?
Yes for finished ads, when the numbers from practitioner teardowns hold. A personal clone has been reported near 87% of a real human’s conversion at about 31% lower CPA. Generic stock AI UGC fails often in the same class of reporting, including large CTR drops from roughly 0.2-second lip-sync lag.
Treat those figures as practitioner claims, not a disclosed large-N study. The pattern is still useful: likeness beats anonymous synthetic faces, and real first-person proof still wins when the format depends on it. Stock AI faces remain a cheap way to test hooks and angles; that cost and workflow job is separate from cloning yourself.
How do I clone my voice without sounding like TTS?
Record yourself performing the line, then run speech-to-speech. Do not paste a script into text-to-speech and hope. Timing and breath are the human signal.
Generate line by line over the cut demo. One trained voice, used consistently, beats rotating stock voices. If disclosure or quality would embarrass the ad, stay on your real mic for that batch and spend the minutes elsewhere.
When is a face clone worth it?
When you want long-run founder recognition and will not film weekly. Not when you need the viewer to learn the UI. Instructor-presence research finds little learning benefit from an on-screen face and extra cognitive load; gaze sticks to the face.
Keep any face segment separate from the demo. Platforms may auto-label synthetic media (Meta AI info, TikTok and YouTube disclosure toggles). Labelled posts tend to draw more skeptical comments. If the label would kill the ad, ship faceless.
Should a clone look polished?
No. Over-polish reads as commercial in feed. High production sat at 6.97% hit rate in Motion 2026, below founder-simple and well below pure text. Keep rough edges: real room, imperfect pace, bold text rather than designed type.
Good to know
How much does voiceover add over captions alone?
About seven conversion points in the platform-reported pair: roughly +80% with captions and a written offer, roughly +87% with voiceover added. Text does most of the work.
Can I clone just my voice for app demos?
Yes. Screen plus speech-to-speech is usually better than a face clone: no lip-sync risk, attention stays on the product, and faceless styles already lead the hit-rate table.
Is a personal clone better than a stock AI actor?
In practitioner teardowns, yes: about 87% of human conversion at lower CPA for a personal clone, versus frequent failure and lip-sync CTR drops for generic stock AI UGC. Stock faces are still useful as cheap hook tests. Neither replaces captions-first product demos.
Do I have to disclose a clone?
Usually yes when the media is synthetic. Meta, TikTok and YouTube all have labelling or disclosure paths. If disclosure would sink the creative, use screen and captions instead.
This expands the First stretch camp on the climb, which lays out the stages either side of it. Tools by job are on the route index, the free ones are here, and the other answers are in insights. Updated 2 August 2026.