Hey, it's Cindy 🤍 A faceless channel is a YouTube or Instagram account where you never show your face: it's a topic, a voice, and visuals. This is the exact stack from the reel, and it's three moving parts: a niche that actually pays, a content system that never runs dry, and a voice that sounds human. The voice layer is
Fish Audio, and it's free to build with right now.
Pick a niche that pays 💰
The biggest mistake is picking a niche you like instead of one that pays. What decides your income is your RPM (revenue per mille): what you actually keep per 1,000 monetised views after YouTube takes its share. That is not the same as CPM, which is what advertisers pay per 1,000 impressions before the platform's cut. Same view count, wildly different payout, and the difference is your topic.
🏦 Finance, top of the list Roughly $15 to $30 RPM. Advertisers (brokers, banks, fintech) pay the most to reach money-minded viewers, so a million views is worth the most here.
🤖 AI and tech, strong second Roughly $10 to $25 RPM. Not quite finance, but close, and the audience is exploding right now. This is the sweet spot if finance feels dry to you.
🎬 Entertainment and story, volume play Roughly $2 to $5 RPM. Way lower per view, so this only works if you can pull huge view counts to make up the gap.
🎯 How to choose yours Pick the highest-paying niche you can make 30 videos about without getting bored. RPM sets the ceiling, consistency decides whether you reach it.
📊 These are ranges, not promises. RPM shifts with your audience country, watch time, and the season (it climbs in Q4 when ad budgets peak). The numbers above are 2026 averages pulled from vidIQ, Fluxnote, and OutlierKit, so treat them as a realistic map, not a payslip.
part two
A content system that never runs dry 🔁
Faceless channels die when the person behind them runs out of ideas by video ten. The fix is to stop making videos one at a time and let Claude plan a whole month at once: titles, hooks, outlines, and a search phrase for each one, all clustered so the channel reads as one clear promise instead of random uploads. Paste this into Claude and you have your next 30 videos before lunch.
📋 the content-plan prompt
You are my faceless channel strategist. My niche is [AI tools for small business owners]. Build me a 30-day content plan as a table. For each video give me: a scroll-stopping title (front-load the payoff or a number, under 60 characters), the single hook line for the first 3 seconds, a 5-beat outline, and the exact search phrase it targets. Cluster the 30 videos into 4 or 5 content pillars so the channel reads as one clear promise, not random uploads. Then flag the 5 titles with the highest search demand and the lowest competition so I film those first. Never write a title I cannot fully pay off inside the video.
💡 Keep the plan alive. Once a few videos land, feed the winners back to Claude ("this title got 4x my average, give me 10 more angles like it") so the system learns what your audience actually clicks. That feedback loop is what separates a channel that grows from one that stalls.
part two, the look
Visuals that stop the scroll 🎬
Faceless does not mean boring stock footage. You want visuals that feel made for the video, and you can generate them from a text prompt. Real tools that do this well right now:
🎥 Higgsfield Strong for stylised, cinematic AI clips and consistent scenes you can string into a whole video. Good default for a faceless channel that wants a signature look.
🌊 Runway or Kling Two other solid text-to-video tools. Runway is great for clean, controllable motion; Kling is known for realistic movement and longer shots. Try both and keep whichever matches your niche.
🎨 Pick one visual style and stick to it. A faceless channel builds recognition through its look and its voice, not a face. Same colour grade, same pacing, same voice on every video, and viewers start to recognise you without ever seeing you.
part three
The voice (this is where it gets real) 🎙️
The voice is what makes or breaks a faceless channel, and it's the part most people get wrong with flat, robotic narration. Fish Audio is the voice layer here, and it does the two things that matter: it sounds human, and it's free to build with right now.
🎨 Design or clone a voice Describe the voice you want in plain text with Voice Design, or clone your own in about 10 seconds from a short clean sample. Either way you get a reusable voice you type any script into.
🎭 Emotion tags Tag the feeling right in the script, in square brackets, so it reads with real emotion instead of a flat robot tone. This is the thing old AI voices could never do.
🌍 83 languages Narrate the same channel in 83 languages, so one idea can travel to an international audience in the same voice.
🆓 Free via the API right now The s2.1-pro-free model is free for developers in beta right now. Great window to build a whole channel on it (see the note below before you count on it long term).
🤍 Honest quality framing. Fish Audio is some of the best voice quality for the price: a top open-weight model at a fraction of what ElevenLabs costs. It is not the single highest-quality voice on the market, but for a faceless channel on a budget the value is hard to beat.
the walkthrough
Wire the voice into your workflow 🔌
- 🔑 Get your API key. Sign in at Fish Audio, open the developer or API settings, and create an API key. Copy it.
- 🔒 Paste it into Claude Code as an env var. Store the key as an environment variable named
FISH_AUDIO_API_KEY so it stays out of your code and can't leak. Never hard-code it into a file. Then ask Claude Code to read the key from that env var when it calls the API.
- ✍️ Write the script in Claude with emotion tags. Have Claude write your narration, then tag the feeling right in the text so the read lands. On the S2.1 model the tags go in square brackets before the line:
🎭 emotion tags in the script
[excited] Okay, this next tool is going to change how you work.
[calm] Here is exactly how it works, step by step.
[whispering] And almost nobody is using it yet.
- 🔊 Generate the speech. Ask Claude Code to send the tagged script to Fish Audio using the model string
s2.1-pro-free and save the audio file. Drop that file under your generated visuals and you have a finished faceless video, all from one script.
proof it works
This is a real business, not a dream 📈
Fortune (Dec 30, 2025) verified a 22-year-old, Adavia Davis, running a
network of faceless YouTube channels that brings in around
$40,000 to $60,000 a month, on roughly
2 hours of work a day. Note the shape of it: it's a network of channels, not one lucky video, and five figures a month, not overnight riches. Source:
Fortune, Dec 30, 2025.
read this first
A few honest notes 📌
⏳ The "free" is time-bound. The s2.1-pro-free model is free for developers right now, in beta. Fish Audio has extended that window before, but treat it as free right now, not free forever, and check the current status before you build something you can't afford to move off.
🎙️ Retention dies in the first few seconds if the voice sounds robotic. A flat, obviously-AI narration is the fastest way to lose a viewer. Spend your effort here: use the emotion tags, pick a voice that fits the niche, and read your intro out loud to check it lands before you publish.
🌱 Be genuine, no get-rich-quick promises. A faceless channel is a real content business: it rewards a clear niche, consistent uploads, and videos that actually help someone. The tools remove the bottlenecks, but the audience still shows up for something worth watching. Build it like you mean it.
The links 🔗
🎙️ Fish Audio (the voice, free API right now): fish.audio
🎥 Higgsfield (AI visuals): higgsfield.ai
📊 RPM ranges: vidIQ, Fluxnote, OutlierKit
Follow
@cindiezhu for more AI you can actually use 🤍