
FacelessKit | Faceless YouTube Script and B-roll
FacelessKit | Faceless YouTube Script and B-roll
Send a topic or your research notes. FacelessKit writes the plan for one faceless YouTube video of 8 to 15 minutes, then Python checks the length, the chapter timestamps and every number in the script before any file is written. You get the files you need to record a voiceover, gather or generate footage, and fill in the upload page.
These are plans and prompts. FacelessKit does not generate video, voice or images, and it does not upload anything.
What you get
script.mdandscript.pdf: the narration by chapter and paragraph, each with its start and end time, on-screen text, B-roll prompt and stock search phrase.voiceover.txt: the narration alone, ready to read or paste into a text-to-speech tool.chapters.txt: YouTube chapter timestamps.broll_prompts.csv: one row per paragraph with timing, number of clips needed at your clip length, the generator prompt, a stock footage search phrase and the on-screen text.metadata.txt: 5 title options with character counts, a description with the chapter list added, tags, and 3 thumbnail texts.thumbnail_mockups.png: the 3 thumbnail texts drawn as layout mockups, plus the same three at search-result size.fact_check.csv: every factual claim and the numbers in it, markedverifyunless you supplied the source.check.md: the results of every check below.
How the checks work
Length. Python counts the words a narrator actually says, with numbers expanded the way they are read ("1872" is three words, "$4.5 million" is five), and divides by your pace in words per minute. The script must land within 5% of your target or the build stops and says how many words to add or cut. The default pace is 150 WPM. We measured six LibriVox narrators reading public-domain texts (8.2 hours of audio): their pace ran from 127 to 160 WPM with a median of 150, and at 150 WPM our estimate was off by 5.9% on average (largest 15.2%, the slowest reader). Record one chapter and set your own pace if yours differs.
Chapters. Timestamps start at 00:00, rise in order, and each chapter is at least 10 seconds, which are the rules on YouTube's chapter help page. Chapter lengths add up exactly to the total length.
Numbers. Every number in the narration, on-screen text, titles, thumbnail texts, description and chapter titles must appear in the fact list, or the build stops. In our tests, 120 out of 120 planted unlisted numbers were caught and 400 out of 400 test sentences were counted and read correctly. Facts the assistant wrote from its own knowledge are marked verify: check them against a primary source before you publish.
B-roll. Each paragraph gets a prompt you can paste into a video generator and a short search phrase for stock libraries. Choose a generator and the clip count follows your clip length. Clip lengths are enforced only where we confirmed them in official documentation: Runway Gen-4 (5 or 10 s, prompts up to 1,000 characters), Luma Ray2 (5 or 10 s), Veo 3 (8 s). Kling, Seedance 2.0 and Hailuo are marked "verify" because we could not confirm their current limits. Sora is refused because OpenAI has ended the Sora API. Prompts never name or depict real people, brands or film and game characters.
Thumbnails. Each text is at most 5 words and 30 characters, drawn at 72 px or larger on a 1280 x 720 frame, and the text-to-background contrast is checked at 4.5:1 or higher. Attach a still image and it is used as the background; otherwise a hatched box marks where your image goes.
Before you publish
The kit adds these notes to metadata.txt, summarised from YouTube Help pages read on 2026-11-18:
- YouTube's monetization policies exclude reused content (content from YouTube or elsewhere without significant original commentary or changes) and inauthentic content, the July 2025 name for repetitive or mass-produced content. Narrate in your own voice, edit the footage, and do not publish a series that only swaps the topic in one template.
- Realistic AI-generated footage, such as a scene that never happened or altered footage of a real place, must be disclosed with the "AI use" setting in YouTube Studio.
- FacelessKit does not promise views, income or monetization approval.
Plans
| Plan | Price | Requests |
|---|---|---|
| Daily | $2.99 | 5 per day |
| Weekly | $7.99 | 10 per week |
| Monthly | $19.99 | 30 per month |
The weekly plan has a 1-day trial with 2 free requests. One message is one build; a build may include up to 2 automatic fix rounds.
Supported requests
- Length: 8 to 15 minutes. Pace: 110 to 200 WPM (default 150).
- Genres: history, mystery, finance, documentary, science, true story, general. Finance scripts are general education and the description gets a "not financial advice" line.
- Generators: any, Runway Gen-4, Luma Ray2, Veo 3, Kling, Seedance 2.0, Hailuo. Aspect 16:9, 9:16 or 1:1.
- Input: a topic in your message, optionally a notes file and a still image for the thumbnail background.
Limits
- Narration is written in English; the timing model and fonts are built for Latin text. You can write to the agent in another language.
- Timing is estimated from the text and your WPM, not measured from a recording.
- The assistant can make factual mistakes. Facts marked
verifyhave not been checked. - Results from video generators and stock libraries depend on those services.
API externes
PyPI (pypi.org) · Python package files (files.pythonhosted.org)

