All-SaaS web tools
Less than 1 video/mo; zero technical interest
$20–60/mo · Low
△ Subscriptions exceed what an episode earns; edits feel like data entry
Drawing set · L0–L3
make faceless YouTube videos
You have things to say but don't want your face on camera. The video-editor route costs $50+/mo in subscriptions and hours of timeline-dragging per episode. The pipeline route treats a video as code: record your voice, let AI transcribe and align subtitles, let code render the visuals — every episode reuses the same machine. This page maps every way to build it, then marks the path we actually run every week.
Design philosophy
A faceless channel sells a voice and a point of view. The pipeline exists so that each new episode costs only the thinking, not the editing.
The voice is the brand
Viewers return for how you think and sound. Automate the timeline, the subtitles and the render — never the part people actually came for.
An episode is code
If an episode can be rebuilt from a script, a recording and a template, a typo fix is a re-render, not an evening of dragging clips.
Retention beats polish
Viewers leave in the first thirty seconds, not at the transitions. Spend effort on the hook and the pacing before any visual effect.
If you're one person with more time than money and want episodes to get cheaper the more you make, take the code pipeline (marked recommended). Only pick the SaaS route if you ship less than one video a month.
SheetL0
First episode
Used byYou and your first viewers
Per month$0
①Why move up
You want to find out whether you have a show at all. Nobody is waiting for episode two yet, so finishing one episode matters more than building any system.
②Architecture
Record your voice, let a speech model produce subtitles, and assemble the rest in whatever simple editor you already have. Nothing is automated; you are learning which steps repeat.
Δ New at this level
- + A decent microphone
- + A script template
Shapes on the map: Semi-local (record + AI subtitles + template editor)
③Tools
Use free, local tools for everything. The only purchase worth making at this level is a microphone: bad audio loses viewers faster than plain visuals.
④Code and tests
- Watch the finished episode on a phone with the sound off — the subtitles alone should carry it.
- Measure loudness before uploading; quiet or clipping audio is the most common first-episode flaw.
- Write down every manual step you took; that list is the specification for the next level.
SheetL1
Weekly show
Used bySubscribers who expect the next episode
Per month$0–10
①Why move up
You are repeating the same twenty clicks every week. A schedule now exists, and every hour spent editing is an hour not spent writing.
②Architecture
The episode becomes a build: script, then your recording, then forced alignment of subtitles, then visuals rendered from code templates, then one final mux. One command turns inputs into a finished video.
Δ New at this level
- + Visuals rendered from code
- + Reusable scene templates
- + One command from inputs to video
Shapes on the map: Full code pipeline (our route)
③Tools
Choose tools you can drive from a script, not only from a mouse. Anything that only works through clicks cannot be handed to an agent later.
④Code and tests
- Pin tool and model versions so last month's episode still rebuilds identically.
- Keep a few reference frames per template and compare new renders against them.
- Check loudness automatically in the pipeline instead of by ear.
SheetL2
Monetized channel
Used byViewers, sponsors and the platform's rights system
Per month$10–50
①Why move up
The channel earns money. Now a copyright claim can take the revenue of an episode, and a lost raw recording can mean an episode you can never fix.
②Architecture
The pipeline gains a record of where every asset came from and under which license, backups of all raw recordings, and a check step before publishing. Paid music or stock footage enters only through that record.
Δ New at this level
- + A license record for every asset
- + Backups of raw recordings
- + A pre-publish rights check
- + Sponsor disclosure in the template
Shapes on the map: Full code pipeline (our route)
③Tools
Spend money on licensed music and assets before anything else; it is the cheapest insurance against losing an episode's revenue.
④Code and tests
- No asset enters a render without a license entry — make the pipeline refuse it.
- Back up raw recordings in two places; the render can be rebuilt, the performance cannot.
- Upload privately first and wait for rights checks before going public.
SheetL3
Many channels, many languages
Used byAudiences in several languages
Per month$50–300+
①Why move up
One recording could serve several languages or channels, and rendering on your own machine has become the bottleneck.
②Architecture
Rendering moves to a queue that runs in the cloud, scripts are localized with human review per language, and synthetic voices are used only where the brand allows and the platform's disclosure rules are met.
Δ New at this level
- + A batch render queue
- + Localized scripts with human review
- + Synthetic-content disclosure labels
Shapes on the map: Full code pipeline (our route) · Fully AI-generated (synthetic voice & visuals)
③Tools
Localization is where quality silently drops. Budget for a native reviewer per language before you budget for more render capacity.
④Code and tests
- Label synthetic voices and visuals according to each platform's current rules.
- Sample-review every batch in every language before it publishes.
- Track retention per language; a translation that loses viewers is worse than no translation.
The architecture map
Every shape this can take. Most exist so you know what not to build yet.
Semi-local (record + AI subtitles + template editor)
Weekly cadence, some comfort with desktop tools
$0–20/mo · Low–mid
△ You repeat the same 20 clicks every episode — that's a script waiting to exist
Full code pipeline (our route)
Weekly+, one person, want marginal cost near zero
$0/mo baseline · Mid up front, near-zero after
△ Channel earns enough to buy back your recording time — then automate voice too
Fully AI-generated (synthetic voice & visuals)
Volume play across many channels; voice isn't part of the brand
$10–100+/mo · 中
△ Platforms increasingly label synthetic content; retention drops tell you first
Concepts you'll actually need
One plain sentence and one metaphor each. Click any tool module for the same treatment.
- Forced alignment
- Matching each word of a known script to its exact timestamp in the audio.
- ≈ Like karaoke lyrics that light up exactly when you sing them.
- Frame rate & timeline
- Video is just still images shown fast; 30fps means every frame owns 1/30 second.
- ≈ A flipbook — your code decides what's drawn on each page.
- Rendering
- Turning your code/timeline description into actual video frames — the slow, CPU-hungry step.
- ≈ The oven stage of baking — recipe's done, now it needs heat and time.
- Codec vs container
- The codec (H.264) is how pixels are compressed; the container (.mp4) is the box it ships in.
- ≈ Vacuum-packing (codec) versus the shipping carton (container).
- Loudness & ducking
- BGM must auto-quiet when you speak; platforms also normalize overall loudness (LUFS).
- ≈ A good DJ fades the music down the moment the MC grabs the mic.
Take the skill.md with you
The Skill-Centric starter file for this scenario — hand it to your agent before the first prompt. It carries the plan, the stack and the known pitfalls.