The faceless YouTube system: niche prompt + Fish Audio voice

Faceless channels are having a moment, but most of them quietly make nothing. The ones that actually pay do two boring things right: they pick a high-RPM niche on purpose, and they do not sound like a robot. Here is the exact setup for both, plus the honest version of what it really takes.

The two-tool system 🧠

Everything runs on two tools, and Claude does most of the heavy lifting.

1. 🎯 Claude is the brain. It scores the niche, writes the scripts, and structures the videos. On its own it writes pretty generic scripts, so the system below is what actually makes them good.
2. 🎙️ Fish Audio is the voice. It reads your script with real emotion, or in a brand new voice you design from a text prompt, so your watch time does not collapse in the first ten seconds.

Step 1, pick the niche with Claude 🎯

Niche is where most people lose, because you get paid on RPM, which is revenue per 1000 views, and advertisers pay wildly different amounts depending on the audience. A finance channel can run $15 to $30 RPM, while entertainment is more like $3 to $8 for the exact same effort. Paste this into Claude to score your options instead of guessing:

You are a faceless YouTube strategist. I want to start a faceless channel I can run mostly with AI. Score 10 niches I could realistically enter on four things: average RPM (give the dollar range), how saturated it is right now, how easy it is to produce faceless with an AI voice plus stock or generated visuals, and realistic months to monetization. Rank them best to worst for a beginner who wants the most earnings per hour of work, and explain why the top three win. Then for my number one pick, give me 10 video title ideas based on what already goes viral in that niche.

Step 2, the content system, not just Claude 🎬

This is the part nobody tells you. If you just ask Claude to write a YouTube script, you get something flat and generic. The system is what makes it work:

1. 🔎 Model the winners first. Find 5 videos already winning in your niche, paste their titles, hooks, and view counts into Claude, and ask it to reverse engineer the pattern before it writes anything.
2. ✍️ Write to a structure, not a blank page. Tell Claude the exact shape: a hook in the first 5 seconds, a reason to stay every 30 seconds, the payoff, and a loop back at the end.
3. 🔁 Batch and iterate. Generate 5 scripts at once, keep the best, and feed the winners back so each round gets sharper.

Here are 5 winning videos in my niche (titles, hooks, and view counts pasted below). Reverse engineer what they have in common: the hook style, the structure, the pacing, the emotional angle. Then write me a 6 to 8 minute faceless script in that exact pattern on [my topic], with a hook in the first 5 seconds and a reason to keep watching every 30 seconds.

Step 3, the voice with Fish Audio 🎙️

A robotic voiceover kills your retention instantly. Fish Audio fixes that two ways:

1. 🎭 Emotion tags. You type the feeling right into the script, like [excited] on the hook or [whisper] on the reveal, and the voice actually performs it instead of reading flat.
2. ✨ Voice Design. If you do not want to use your own voice, you describe the voice you want in a text prompt and Fish Audio builds a brand new original one in about 15 seconds.

Setup, start to finish:

1. 🔑 Get your key. Make a free account at fish.audio and grab your API key. Keep it in a file (a .env file), never paste it into a chat.
2. 🗣️ Pick your voice. clone your own from a short recording, or use Voice Design to generate an original one.
3. 🏷️ Tag and generate. Write your script with the emotion tags inline, then generate the audio.

Read this script in a warm, confident documentary voice. Use the emotion tags exactly as written: [calm] for the intro, [excited] on the big number, [whisper] on the twist. Keep the pacing slow enough to feel premium.

Step 4, the visuals without a camera 🎞

Faceless means no face, not no video. Three ways to fill the screen, cheapest first:

1. 🆓 Stock footage: Pexels, Pixabay and Coverr are free with no attribution. Ask Claude to turn your script into a shot list first ("give me a b-roll search term for every 15 seconds of this script") so you're searching with intent instead of scrolling.
2. 🎬 AI-generated b-roll: for shots stock can't give you, generate clips in Higgsfield (cinematic camera controls) or the free open-source Open Generative AI (pay cents per clip through your own key).
3. ✂️ Assemble in CapCut: voiceover on the timeline first, cut the visuals to it, auto-captions on (a large share of viewers watch muted), and change the visual at least every 5 seconds to hold retention.

3 bonus prompts to run next 🎁

🧲 Title and thumbnail tester.

Give me 10 title and thumbnail-text combos for this video, ranked by likely click-through. For each, tell me the curiosity gap it opens and who it is aimed at.

📅 30-day content calendar.

Build me a 30-day upload calendar for my faceless [niche] channel: 12 video ideas ordered so the easy wins come first, each with a title, a one-line hook, and the search intent it targets.

🔬 Retention autopsy.

Here is the script for a video that underperformed. Find the exact lines where a viewer would click away, and rewrite those moments to keep them watching.

How to get the most out of it 🎓

The honest part 🫶

The links 🔗

🎙 Fish Audio: fish.audio (voice cloning, Voice Design, emotion tags)
🆓 Free stock footage: Pexels · Pixabay · Coverr
🎬 AI b-roll: Higgsfield · Open Generative AI
✂️ Assembly: CapCut