faceless-explainer
Faceless-channel editor: turns a script (or a recorded voice-over) into a finished explainer with no presenter on camera - offline TTS voice-over (Kokoro / Windows SAPI / say), one visual per sentence (B-roll, photos, charts, maps, kinetic type), heavy word-synced captions and a
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/faceless-explainer
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install viprasol-tech-claude-video-editing-skills@llmmart
git clone https://github.com/Viprasol-Tech/claude-video-editing-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole viprasol-tech/claude-video-editing-skills collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Faceless Explainer
Script in, explainer out: a natural voice-over, a new visual every sentence, captions that carry muted viewers.
- Best for: explainers, top-N lists, history/facts, finance or tech breakdowns, article-to-video
- Inputs: a script (
script.txt) or a recorded VO (voice.wav), a folder of visuals (clips/photos named after what they show), optional music - Default skin:
essay-mode· alternativesdata-story,swiss-poster(use--no-matte),countdown-list(9:16 lists) - Formats: 16:9 1920x1080 master + 9:16 cutdown;
captions.srt, chapters for long ones - Duration target: 60 s - 15 min; Shorts 30-60 s; genre preset
faceless-explainer
Before you start
- Needs video-studio-core in the same skills folder.
CORE=video-studio-core,SKILLS= the folder holding both. - First time:
python CORE/scripts/check_env.py. Best free voice:pip install kokoro soundfile; without ittts.pyfalls back to Windows SAPI / macOSsay/ espeak-ng. List voices:python CORE/scripts/tts.py --list-voices. - Preset:
python CORE/scripts/editor.py genre faceless-explainer. - Platform note: YouTube demonetises mass-produced, repetitive content. Every video needs a real script, real research and varied visuals; disclose AI voices where the platform asks. Use only footage, photos and music you have rights to.
How a pro cuts this
| beat | % | what happens |
|---|---|---|
| hook | 0-5 | the most surprising fact or question, on screen as text, by 2 s |
| promise | 5-10 | what the viewer will know by the end |
| points / chapters | 10-85 | one visual per sentence; a chapter card between sections |
| payoff | 85-95 | the answer, the twist, the number |
| CTA / next | 95-100 | one line + end screen space |
- Script first: short sentences (8-16 words), one claim each, numbers written as digits. Read it aloud before generating.
- Pacing: a new visual every 2-4 s (one per sentence, two for long ones); text changes every 1.5-3 s in Shorts. Leave 0.3-0.5 s of air between sections (pauses in the script via line breaks).
- Visual rules: the visual matches the words being said (literal for nouns, metaphor for ideas). Numbers become
counter/bars/linechart, places becomemap, lists becomesteps/checklist. Never the same stock clip twice. - Transitions: cuts on sentence ends; one soft transition kind (
fade0.3 s) between chapters only. - Music/SFX: a bed 20-24 LU under the VO, ducked ("speech-aware"); whoosh on chapter cards, ticks on counters; sparse.
- Colour: unify mixed stock with one look:
documentary-natural0.5, orfinance-navy/cool-corporatefor money/tech. - Captions: preset
boxedorclean(16:9),hormozi(9:16); <= 20 cps; key words in accent. - Client judges: audio that doesn't sound robotic, cut density, accuracy, a recognisable template.
- Common mistakes: monotone TTS at default speed, visuals unrelated to the words, a 20 s logo intro, wall-to-wall text, music louder than the voice, the same stock clip reused.
Make the edit
- Voice-over (skip if the user recorded
voice.wav; then transcribe it in step 3):
python CORE/scripts/tts.py --script script.txt --out vo/voice.wav --words --model base --rate 1.05
# -> vo/voice.wav + vo/words.json (word times, script spelling). Keep it OUT of the work folder fx/.
- Read
vo/words.jsonand write a timeline specfx/spec.json: one visual per sentence, itsdur(images) orin/out(clips) equal to that sentence's length, the voice as the specmusicat full level so it becomes the soundtrack (paths are relative to the spec file):
{"format": "16:9", "fps": 30,
"clips": [{"src": "../visuals/ocean-waves.mp4", "in": 2, "out": 6.1, "audio": "mute"},
{"src": "../visuals/old-map.jpg", "dur": 3.4},
{"src": "../visuals/ship-deck.jpg", "dur": 4.0, "transition": {"type": "fade", "dur": 0.3}}],
"music": {"src": "../vo/voice.wav", "volume": 1.0, "duck": false, "fadeOut": 0.2, "voiceover": true}}
Clip durations must add up to the voice length (words.json duration) plus every transition dur (a transition overlaps two clips). Images get a slow centre push-in automatically.
- Assemble and plan.
"voiceover": truein the spec'smusicblock makestimeline.pytranscribe the voice, sowords.jsongets word times for the captions (a plain music bed is never transcribed):
python CORE/scripts/run.py - --timeline fx/spec.json --style SKILLS/essay-mode --genre faceless-explainer \
--model base --work fx --stop-after plan
# fallback (no "voiceover" flag, or to keep the exact TTS spelling): run with --stop-after timeline, then
# cp vo/words.json fx/words.json (TTS words already sit on the timeline clock: the voice starts at 0)
# python CORE/scripts/make_plan.py --work fx --style SKILLS/essay-mode --genre faceless-explainer
python CORE/scripts/editor.py review fx/plan.json --genre faceless-explainer --fix
python CORE/scripts/render.py fx/plan.json --frames 1,5,10,20
python CORE/scripts/render.py fx/plan.json
python CORE/scripts/finish.py fx/plan.json --platform youtube
Quick version (one background, B-roll placed by keyword): mux a still with the voice (ffmpeg -loop 1 -i bg.jpg -i voice.wav -r 30 -pix_fmt yuv420p -c:v libx264 -c:a aac -shortest base.mp4), then run.py base.mp4 --style SKILLS/essay-mode --genre faceless-explainer --no-cut --broll visuals/ --work fx2 --stop-after plan. Name visual files after the words they illustrate or broll.py will not match them.
Refine the plan
- Title card: the skin's auto
titlecardtakes the first words of the VO and can cut mid-phrase; rewrite itstitleto the hook (<= 7 words). - Words: fix any mis-heard word in
plan.wordsto the script spelling (keep times). - Hook: a
headline0-2.5 s with the hook question;"sfx": [{"t": 0, "kind": "impact", "gain": 0.4}]. - Chapter cards:
{"type": "titlecard", "kicker": "Part 2", "title": "The Wreck", "start": 41.0, "end": 43.0, "in": "maskUp"}plus a top-level"chapters": [{"start": 0, "title": "Intro"}, {"start": 41.0, "title": "The Wreck"}]list in the plan for delivery. - Data:
counter,bars,linechart,donut,barracetimed 0.1 s before the number is spoken;mapwithpinsfor places (seeded dot map, not real geography - say so if accuracy matters). - Lists:
stepsorchecklistwithitems[].at= spoken time. - Captions:
{"type": "captions", "preset": "boxed", "keywords": ["1912", "iceberg"]}. - Voice polish:
"audio": {"voice": {"denoise": false, "eq": "broadcast", "deess": true}}(TTS is clean; don't denoise). - Music:
"music": {"src": "bed.mp3", "under": 22, "duck": "speech-aware", "fitTo": "beats", "endButton": true}. - Look:
"color": {"look": "documentary-natural", "intensity": 0.5},"finish": {"grain": 5, "vignette": 0.3}.
Deliver
python CORE/scripts/deliver.py fx/plan.json --to youtube,shorts --out-dir fx/deliver
python CORE/scripts/qc.py fx/deliver/final_youtube.mp4 --plan fx/plan.json --platform youtube
- YouTube: upload
captions.srtas a sidecar; pastechapters.txtinto the description (needs >= 3 chapters of >= 10 s). - The 9:16 from a 16:9 plan is a blurred pillarbox. For a real vertical, re-run step 3 with
"format": "9:16"in the spec,--format 9:16andcountdown-listorbold-bounce. - Thumbnails:
thumb_*.png; check the headline is the actual hook.
Check before delivering
- Every sentence has a matching visual; no repeats; no visual left on screen > 6 s without motion.
- VO natural (rate, pronunciation of names); captions match the script; numbers and dates verified.
- Music never masks the VO;
qc.py-14 LUFS / -1 dBTP; no black frames between clips. - Rights: every clip, photo and track is licensed or the user's own; credits in the description if required (CC BY).
- Tell the user: the voice used, video length, chapter list, file paths.
Example requests
- "Turn this script into a faceless YouTube video with an AI voice and stock footage."
- "Make a 60-second faceless Short from these 6 facts, big captions."
- "Narrate my blog post over these photos, add a chart for the numbers."
Files (claude-video-editing-skills)
-
SKILL.md 8.7 KB
--- name: faceless-explainer description: "Faceless-channel editor: turns a script (or a recorded voice-over) into a finished explainer with no presenter on camera - offline TTS voice-over (Kokoro / Windows SAPI / say), one visual per sentence (B-roll, photos, charts, maps, kinetic type), heavy word-synced captions and a consistent channel template, in 16:9 and 9:16. Use when the user says: faceless video, make a video from this script, explainer with voice-over, AI voice video, top-10 / facts / history channel, narrate these images, turn my article into a video. Default skin essay-mode; any Video Studio skin works." license: MIT --- # Faceless Explainer Script in, explainer out: a natural voice-over, a new visual every sentence, captions that carry muted viewers. - **Best for:** explainers, top-N lists, history/facts, finance or tech breakdowns, article-to-video - **Inputs:** a script (`script.txt`) or a recorded VO (`voice.wav`), a folder of visuals (clips/photos named after what they show), optional music - **Default skin:** `essay-mode` · alternatives `data-story`, `swiss-poster` (use `--no-matte`), `countdown-list` (9:16 lists) - **Formats:** 16:9 1920x1080 master + 9:16 cutdown; `captions.srt`, chapters for long ones - **Duration target:** 60 s - 15 min; Shorts 30-60 s; genre preset `faceless-explainer` ## Before you start - Needs **video-studio-core** in the same skills folder. `CORE` = `video-studio-core`, `SKILLS` = the folder holding both. - First time: `python CORE/scripts/check_env.py`. Best free voice: `pip install kokoro soundfile`; without it `tts.py` falls back to Windows SAPI / macOS `say` / espeak-ng. List voices: `python CORE/scripts/tts.py --list-voices`. - Preset: `python CORE/scripts/editor.py genre faceless-explainer`. - **Platform note:** YouTube demonetises mass-produced, repetitive content. Every video needs a real script, real research and varied visuals; disclose AI voices where the platform asks. Use only footage, photos and music you have rights to. ## How a pro cuts this | beat | % | what happens | | --- | --- | --- | | hook | 0-5 | the most surprising fact or question, on screen as text, by 2 s | | promise | 5-10 | what the viewer will know by the end | | points / chapters | 10-85 | one visual per sentence; a chapter card between sections | | payoff | 85-95 | the answer, the twist, the number | | CTA / next | 95-100 | one line + end screen space | - **Script first:** short sentences (8-16 words), one claim each, numbers written as digits. Read it aloud before generating. - **Pacing:** a new visual every 2-4 s (one per sentence, two for long ones); text changes every 1.5-3 s in Shorts. Leave 0.3-0.5 s of air between sections (pauses in the script via line breaks). - **Visual rules:** the visual matches the *words being said* (literal for nouns, metaphor for ideas). Numbers become `counter`/`bars`/`linechart`, places become `map`, lists become `steps`/`checklist`. Never the same stock clip twice. - **Transitions:** cuts on sentence ends; one soft transition kind (`fade` 0.3 s) between chapters only. - **Music/SFX:** a bed 20-24 LU under the VO, ducked ("speech-aware"); whoosh on chapter cards, ticks on counters; sparse. - **Colour:** unify mixed stock with one look: `documentary-natural` 0.5, or `finance-navy` / `cool-corporate` for money/tech. - **Captions:** preset `boxed` or `clean` (16:9), `hormozi` (9:16); <= 20 cps; key words in accent. - **Client judges:** audio that doesn't sound robotic, cut density, accuracy, a recognisable template. - **Common mistakes:** monotone TTS at default speed, visuals unrelated to the words, a 20 s logo intro, wall-to-wall text, music louder than the voice, the same stock clip reused. ## Make the edit 1. Voice-over (skip if the user recorded `voice.wav`; then transcribe it in step 3): ```bash python CORE/scripts/tts.py --script script.txt --out vo/voice.wav --words --model base --rate 1.05 # -> vo/voice.wav + vo/words.json (word times, script spelling). Keep it OUT of the work folder fx/. ``` 2. Read `vo/words.json` and write a timeline spec `fx/spec.json`: one visual per sentence, its `dur` (images) or `in`/`out` (clips) equal to that sentence's length, the voice as the spec `music` at full level so it becomes the soundtrack (paths are relative to the spec file): ```json {"format": "16:9", "fps": 30, "clips": [{"src": "../visuals/ocean-waves.mp4", "in": 2, "out": 6.1, "audio": "mute"}, {"src": "../visuals/old-map.jpg", "dur": 3.4}, {"src": "../visuals/ship-deck.jpg", "dur": 4.0, "transition": {"type": "fade", "dur": 0.3}}], "music": {"src": "../vo/voice.wav", "volume": 1.0, "duck": false, "fadeOut": 0.2, "voiceover": true}} ``` Clip durations must add up to the voice length (`words.json` `duration`) **plus** every transition `dur` (a transition overlaps two clips). Images get a slow centre push-in automatically. 3. Assemble and plan. `"voiceover": true` in the spec's `music` block makes `timeline.py` transcribe the voice, so `words.json` gets word times for the captions (a plain music bed is never transcribed): ```bash python CORE/scripts/run.py - --timeline fx/spec.json --style SKILLS/essay-mode --genre faceless-explainer \ --model base --work fx --stop-after plan # fallback (no "voiceover" flag, or to keep the exact TTS spelling): run with --stop-after timeline, then # cp vo/words.json fx/words.json (TTS words already sit on the timeline clock: the voice starts at 0) # python CORE/scripts/make_plan.py --work fx --style SKILLS/essay-mode --genre faceless-explainer python CORE/scripts/editor.py review fx/plan.json --genre faceless-explainer --fix python CORE/scripts/render.py fx/plan.json --frames 1,5,10,20 python CORE/scripts/render.py fx/plan.json python CORE/scripts/finish.py fx/plan.json --platform youtube ``` Quick version (one background, B-roll placed by keyword): mux a still with the voice (`ffmpeg -loop 1 -i bg.jpg -i voice.wav -r 30 -pix_fmt yuv420p -c:v libx264 -c:a aac -shortest base.mp4`), then `run.py base.mp4 --style SKILLS/essay-mode --genre faceless-explainer --no-cut --broll visuals/ --work fx2 --stop-after plan`. Name visual files after the words they illustrate or `broll.py` will not match them. ## Refine the plan - **Title card:** the skin's auto `titlecard` takes the first words of the VO and can cut mid-phrase; rewrite its `title` to the hook (<= 7 words). - **Words:** fix any mis-heard word in `plan.words` to the script spelling (keep times). - **Hook:** a `headline` 0-2.5 s with the hook question; `"sfx": [{"t": 0, "kind": "impact", "gain": 0.4}]`. - **Chapter cards:** `{"type": "titlecard", "kicker": "Part 2", "title": "The Wreck", "start": 41.0, "end": 43.0, "in": "maskUp"}` plus a top-level `"chapters": [{"start": 0, "title": "Intro"}, {"start": 41.0, "title": "The Wreck"}]` list in the plan for delivery. - **Data:** `counter`, `bars`, `linechart`, `donut`, `barrace` timed 0.1 s before the number is spoken; `map` with `pins` for places (seeded dot map, not real geography - say so if accuracy matters). - **Lists:** `steps` or `checklist` with `items[].at` = spoken time. - **Captions:** `{"type": "captions", "preset": "boxed", "keywords": ["1912", "iceberg"]}`. - **Voice polish:** `"audio": {"voice": {"denoise": false, "eq": "broadcast", "deess": true}}` (TTS is clean; don't denoise). - **Music:** `"music": {"src": "bed.mp3", "under": 22, "duck": "speech-aware", "fitTo": "beats", "endButton": true}`. - **Look:** `"color": {"look": "documentary-natural", "intensity": 0.5}`, `"finish": {"grain": 5, "vignette": 0.3}`. ## Deliver ```bash python CORE/scripts/deliver.py fx/plan.json --to youtube,shorts --out-dir fx/deliver python CORE/scripts/qc.py fx/deliver/final_youtube.mp4 --plan fx/plan.json --platform youtube ``` - YouTube: upload `captions.srt` as a sidecar; paste `chapters.txt` into the description (needs >= 3 chapters of >= 10 s). - The 9:16 from a 16:9 plan is a blurred pillarbox. For a real vertical, re-run step 3 with `"format": "9:16"` in the spec, `--format 9:16` and `countdown-list` or `bold-bounce`. - Thumbnails: `thumb_*.png`; check the headline is the actual hook. ## Check before delivering - Every sentence has a matching visual; no repeats; no visual left on screen > 6 s without motion. - VO natural (rate, pronunciation of names); captions match the script; numbers and dates verified. - Music never masks the VO; `qc.py` -14 LUFS / -1 dBTP; no black frames between clips. - Rights: every clip, photo and track is licensed or the user's own; credits in the description if required (CC BY). - Tell the user: the voice used, video length, chapter list, file paths. ## Example requests - "Turn this script into a faceless YouTube video with an AI voice and stock footage." - "Make a 60-second faceless Short from these 6 facts, big captions." - "Narrate my blog post over these photos, add a chart for the numbers."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.