How to create a faceless YouTube video with AI

Faceless is the most popular format for starting a channel without exposing yourself. The trick is in the pipeline — not in any single tool. This recipe ties the 4 pieces into a flow that produces an 8-10 minute video in 90 minutes of work.

Practical Updated 2026-04-26

What this recipe solves

You want to start a channel but: you don't want to be on camera, you don't have a team, you don't know how to edit. The pipeline below removes all 3 blockers.

5-step pipeline

STEP 1

Pick a niche — "evergreen + searched"

Faceless works in niches where authority comes from content, not the creator. Good fits: history docs, basic finance, mysteries, pop science, top-10s, biographies, oddly satisfying.

Bad fits: niches that depend on charisma (comedy, lifestyle, vlog).

Test: search 5 videos with >500k views in your idea. If all show a face, skip. If half don't, go.

STEP 2

Write the script (HVC structure)

HVC = Hook 5s + Value (body) + CTA. The first 5 seconds drive 80% of retention.

Paste this into ChatGPT:

You are a senior scriptwriter for faceless YouTube channels in the [NICHE] niche. Topic: [SPECIFIC TOPIC] Target length: 8 minutes (about 1200 spoken words) Structure: 1. 5s hook — question or shocking fact (no "today we'll talk about") 2. 15s promise — what the viewer will learn 3. Body — 5 sections of ~200 words, each ending with a micro-cliffhanger 4. Closing reflection — 60 words 5. CTA — "subscribe" + point to next video Rules: - Short sentences (max 14 words) - 1 concrete number every 200 words - Tone: informative, slightly intriguing, no "did you know" - Output: spoken script only, no scene directions

STEP 3

Generate the voice (ElevenLabs)

For faceless, pick a Narration voice — authoritative, clear. Stability 50-60%, Style 20%. Adam, Bill, Charlotte work well.

Split the script into 500-character chunks before generating. ElevenLabs breathes better between chunks.

STEP 4

Visuals: 3 strategies

Stock footage (Pictory, Storyblocks) — fastest, less unique
AI images + Ken Burns (Midjourney + CapCut zoom) — balanced
AI short clips (Veo 5s clips chained) — most cinematic, slowest

Starter channel: option 1 or 2. Option 3 only after 5 published videos.

STEP 5

Edit in CapCut

Track 1: voice
Track 2: visuals (1 every 4-6s, with Ken Burns micro-zoom)
Track 3: ambient music at -22dB (Epidemic Sound, Artlist)
Track 4: punctual SFX (whoosh on transitions)
Auto-captions, then proofread

Export 1080p, 30fps. Build the thumbnail separately (Photopea or Canva).

Hook example (HVC structure)

"In 1986, three fishermen found a sealed box on the coast of Maine. Inside: 14 letters. All written by the same person. All dated different years. All with the same signature. Here's where it gets strange — the person had died in 1962."

[promise]: "Today we follow the trail of these letters. What they uncovered shouldn't exist."

Minimum stack (tools and cost)

ChatGPT Plus — $20/mo (script)
ElevenLabs Starter — $5/mo (voice)
Pictory or Storyblocks — $25/mo (stock)
CapCut — free (edit)
Music — Epidemic Sound $12/mo

Total ~$62/mo. Enough for 4-8 videos a month.

Common mistakes

Robotic voice. Stability above 70% kills expression. Keep it 50-60%.
A visual every 2 seconds. Viewers get nauseous. 1 visual every 4-6s is the sweet spot.
Hook with no promise. The hook intrigues — but the second line must promise concrete value, not "come find out."
Skipping the thumbnail. 50% of CTR is the thumbnail. Spend 30 minutes on it.
Publishing before having 5. The YouTube algorithm needs signal — line up 5 in a series before evaluating.

Related recipes

Next step

Learn to scale — 1 video per week

Once you have 5 published, jump to Module 4 of the course: Scale — production at volume with AI.

Open Module 4 →