news-explainer
News and explainer editor for Claude: turns a presenter clip, voice-over or interview into a sourced, inverted-pyramid explainer (key fact first, why it matters, context, data, what next) with attribution on screen, charts and maps, neutral music, accurate captions, a 16:9 master
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/news-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
News Explainer
The fact first, the source on screen, the context after, and nothing the evidence doesn't support.
- Best for: news updates, market and policy explainers, "what happened and why", research summaries, local news
- Inputs: presenter clip or VO, the sources (links, documents, data), optional B-roll/maps/stills with their licences
- Default skin:
news-desk· alternativesdata-story,essay-mode,swiss-poster(needs--no-matteor a matte) - Formats: 16:9 master, 9:16 cutdowns
- Durations: 60 s-8 min master; 30-60 s cutdowns
- Genre preset:
--genre news-explainer
Before you start
- Needs video-studio-core in the same skills folder (
CORE);SKILLS= the skills folder. - First run:
python CORE/scripts/check_env.py. - Sourcing is the job. Get the source for every number, quote and claim before editing. Each data graphic carries a source chip ("Source: ONS, Oct 2026"). If a claim has no source, raise it with the user; do not illustrate it.
- Quotes of real people: their real words, in context, no frankenbites (
story.py reorderflags them). - Footage and photos must be the user's, licensed, or clearly fair-use commentary with credit. No logos or screenshots styled to look like a real outlet's broadcast, and never a "LIVE" or "BREAKING" badge on a recorded explainer.
- Private individuals, minors and victims: blur or omit unless the user confirms consent or clear public interest.
How a pro cuts this
Structure (inverted pyramid, % of length)
| beat | % | what |
|---|---|---|
| key fact | 10 | the news in one sentence, on screen as a headline, in the first 3 s |
| why it matters | 15 | who is affected and how much |
| context | 35 | how we got here: timeline, map, background |
| data / evidence | 25 | the numbers as charts, each with its source; quotes with attribution |
| what next | 15 | what happens next, what is unknown, where to read more |
Pacing. ASL 3-6 s, visual change at least every 5 s (graphic, B-roll, punch-in). Max gap 0.4 s, fillers removed, retakes removed. Calm, not frantic: a news explainer earns trust by not shouting.
Cut rules.
- No hype hook: the hook is the fact itself. No greeting, no "in this video".
- Every jump cut in a presenter piece is fine at this pace, but cover interview cuts with B-roll or a graphic.
- Hold a chart until it has been read (>= 4 s), and say the number while it is on screen.
- Separate fact from analysis visibly (presenter on camera for analysis, graphics for facts).
- Never change a quote's meaning with the shot around it (Kuleshov) or by trimming its qualifier.
Transition grammar. Straight cuts; J/L cuts 2-4 frames on presenter joins; a short dissolve only between sections. No whips, glitches or flashes.
Music and sound. Neutral bed at -28 dB under speech, ducked; none under quotes from victims or sensitive topics.
SFX very low (a soft ui-tap on a chart reveal at most). Dialogue clear: broadcast EQ, de-ess.
Colour. Natural: documentary-natural at 0.4-0.5 (or none). No stylised looks on real-world footage.
Captions. Two-line sentence-case subtitles ("preset": "clean", <= 20 cps), names and numbers checked against the
sources. Sidecar SRT for YouTube. Lower third for every person who speaks, with role.
What the client judges. Accuracy, attribution on screen, clarity, and speed to publish.
Common mistakes. A number on screen without a source; a misleading thumbnail or headline; a chart axis that exaggerates; old footage presented as new (label "File footage, 2023"); stock music too dramatic; captions with mis-heard names; "BREAKING" on a non-breaking story.
Make the edit
python CORE/scripts/run.py presenter.mp4 --style SKILLS/news-desk --genre news-explainer --format 16:9 --smart-cuts --retakes --prompt "ONS, Reserve Bank, Priya Rao" --model base --work nx --stop-after plan
python CORE/scripts/story.py outline --work nx --genre news-explainer # sentences + the genre beat map
python CORE/scripts/render.py nx/plan.json --frames 1,4,8,13,15
python CORE/scripts/render.py nx/plan.json
python CORE/scripts/finish.py nx/plan.json --platform youtube
The genre preset sets the pause cut (0.4 s). With a B-roll folder
(files named after what they show) add --broll broll/ and stop with --stop-after broll instead of plan
(--stop-after plan exits before the B-roll step).
Reorder so the key fact leads (when the presenter buried it):
python CORE/scripts/story.py reorder --work nx --order "s4,s1-s3,s5-s9" --genre news-explainer --format 16:9,
then python CORE/scripts/run.py - --timeline nx/story_timeline.json --style SKILLS/news-desk --genre news-explainer --format 16:9 --model base --work nx2 --stop-after plan.
Check nx/story_flags.md: whole sentences only.
No presenter on camera: record or synthesise the VO (python CORE/scripts/tts.py --script vo.txt --out vo/voice.wav --words;
label synthetic voices as such), then follow the faceless-explainer skill (VO + B-roll + captions) and keep every sourcing rule above.
Refine the plan
The draft from news-desk adds a "BREAKING" ticker, a "LIVE" chip and a lower third filled with the first sentence. Replace them:
- Delete the
chip"LIVE" and fix theticker:{"type": "ticker", "label": "EXPLAINER", "text": "Small business · three costly mistakes · what to fix first", "y": 1000}(y about 1000 keeps it inside a 1080 frame; 1040 clips). - Lower third:
{"type": "lowerthird", "name": "Priya Rao", "role": "Small-business analyst", "className": "nd-lt", "start": 0.3, "end": 4.5}. - Key-fact headline:
{"type": "headline", "text": "3 mistakes, $10,000 lost", "start": 0, "end": 2.6, "x": 1560, "y": 250, "w": 600, "in": "maskUp", "out": "fade"}— keep it off the face. - Data + source:
{"type": "bars", "title": "Growth after fixing all three", "items": [{"label": "Before", "value": 100}, {"label": "After", "value": 150, "highlight": true}], "suffix": "%", "start": 11.6, "end": 15.5, "x": 1580, "y": 420, "w": 600, "h": 460}and right under it{"type": "chip", "text": "Source: speaker's own figures, 2026", "x": 1580, "y": 700, "start": 11.6, "end": 15.5, "style": {"fontSize": "24px"}}. Trends:linechart; shares:donut; places:map(a stylised dot map, not real geography: label it, or use a real map image with credit as animageelement). - Captions:
{"type": "captions", "preset": "clean", "y": 870}above the ticker. - Fix split numbers in
words: Whisper writes "$10" ",000." and "50" "%" as separate words; the creator captions show "$10 ,000" unless you merge them (join the text, keep the first start and the last end). Check every autocounterfor the same reason. - Colour/sound:
"color": {"look": "documentary-natural", "intensity": 0.5},"finish": {"grain": 0},"audio": {"voice": {"eq": "broadcast", "deess": true}},"music": {"src": "neutral.wav", "under": 24, "duck": "speech-aware"}. python CORE/scripts/editor.py review nx/plan.json --genre news-explainer— its "no hook text in 5 s" note is solved by the headline.- Protect identities: blur private individuals, minors, plates and documents in footage with
"blurs": [{"x": 300, "y": 220, "w": 220, "h": 260, "start": 18.0, "end": 22.5, "pixelate": true}](output pixels, top-left box). It sits under the graphics, so the lower third stays sharp.
Deliver
python CORE/scripts/deliver.py nx/plan.json --to youtube,shorts --out-dir nx/deliver
python CORE/scripts/qc.py nx/deliver/final_youtube.mp4 --plan nx/plan.json --platform youtube --strict
python CORE/scripts/qc.py nx/deliver/final_shorts.mp4 --platform shorts # no --plan: the plan is 16:9, the file is reframed
final_shorts.mp4 from a 16:9 plan is a blurred-fill reframe; for a real vertical cutdown re-run the pipeline with
--format 9:16 on the chosen section. Put every source link in the description (start from post.md, which only
drafts from the transcript: add the sources yourself). Re-check post.md titles: no claim the video doesn't support.
Check before delivering
- Every number, name, date and quote on screen matches its source; every data graphic has a source chip.
- No graphic covers a face; ticker and captions don't overlap; nothing outside the 5% title-safe area.
- Charts honest: bars start at zero, units shown, period labelled.
- Archive or file footage is labelled; nothing implies "live" or "breaking".
qc.py: loudness -14 LUFS / -1 dBTP, captions within 20 cps, no flash frames.- Tell the user which claims you could not verify.
Example requests
- "Make a 3-minute explainer from my recording about the new tax rule, sources are in this doc."
- "Turn this market update into a 16:9 video with charts and a 45-second vertical."
- "News-style breakdown of what happened at the council meeting, lower thirds for each speaker."
Files (claude-video-editing-skills)
-
SKILL.md 9.4 KB
--- name: news-explainer description: "News and explainer editor for Claude: turns a presenter clip, voice-over or interview into a sourced, inverted-pyramid explainer (key fact first, why it matters, context, data, what next) with attribution on screen, charts and maps, neutral music, accurate captions, a 16:9 master and vertical cutdowns. Use when the user says 'make a news explainer', 'explain this story', 'news update video', 'breakdown of what happened', 'market news short', 'fact-based explainer with sources', 'turn this report into a video'. Runs on video-studio-core (transcribe, autocut, charts, colour, deliver, QC). Not for opinion rants dressed as news." license: MIT --- # News Explainer The fact first, the source on screen, the context after, and nothing the evidence doesn't support. - **Best for:** news updates, market and policy explainers, "what happened and why", research summaries, local news - **Inputs:** presenter clip or VO, the sources (links, documents, data), optional B-roll/maps/stills with their licences - **Default skin:** `news-desk` · alternatives `data-story`, `essay-mode`, `swiss-poster` (needs `--no-matte` or a matte) - **Formats:** 16:9 master, 9:16 cutdowns - **Durations:** 60 s-8 min master; 30-60 s cutdowns - **Genre preset:** `--genre news-explainer` ## Before you start - Needs **video-studio-core** in the same skills folder (`CORE`); `SKILLS` = the skills folder. - First run: `python CORE/scripts/check_env.py`. - **Sourcing is the job.** Get the source for every number, quote and claim before editing. Each data graphic carries a source chip ("Source: ONS, Oct 2026"). If a claim has no source, raise it with the user; do not illustrate it. - Quotes of real people: their real words, in context, no frankenbites (`story.py reorder` flags them). - Footage and photos must be the user's, licensed, or clearly fair-use commentary with credit. No logos or screenshots styled to look like a real outlet's broadcast, and never a "LIVE" or "BREAKING" badge on a recorded explainer. - Private individuals, minors and victims: blur or omit unless the user confirms consent or clear public interest. ## How a pro cuts this **Structure (inverted pyramid, % of length)** | beat | % | what | | --- | --- | --- | | key fact | 10 | the news in one sentence, on screen as a headline, in the first 3 s | | why it matters | 15 | who is affected and how much | | context | 35 | how we got here: timeline, map, background | | data / evidence | 25 | the numbers as charts, each with its source; quotes with attribution | | what next | 15 | what happens next, what is unknown, where to read more | **Pacing.** ASL 3-6 s, visual change at least every 5 s (graphic, B-roll, punch-in). Max gap 0.4 s, fillers removed, retakes removed. Calm, not frantic: a news explainer earns trust by not shouting. **Cut rules.** - No hype hook: the hook is the fact itself. No greeting, no "in this video". - Every jump cut in a presenter piece is fine at this pace, but cover interview cuts with B-roll or a graphic. - Hold a chart until it has been read (>= 4 s), and say the number while it is on screen. - Separate fact from analysis visibly (presenter on camera for analysis, graphics for facts). - Never change a quote's meaning with the shot around it (Kuleshov) or by trimming its qualifier. **Transition grammar.** Straight cuts; J/L cuts 2-4 frames on presenter joins; a short dissolve only between sections. No whips, glitches or flashes. **Music and sound.** Neutral bed at -28 dB under speech, ducked; none under quotes from victims or sensitive topics. SFX very low (a soft `ui-tap` on a chart reveal at most). Dialogue clear: `broadcast` EQ, de-ess. **Colour.** Natural: `documentary-natural` at 0.4-0.5 (or `none`). No stylised looks on real-world footage. **Captions.** Two-line sentence-case subtitles (`"preset": "clean"`, <= 20 cps), names and numbers checked against the sources. Sidecar SRT for YouTube. Lower third for every person who speaks, with role. **What the client judges.** Accuracy, attribution on screen, clarity, and speed to publish. **Common mistakes.** A number on screen without a source; a misleading thumbnail or headline; a chart axis that exaggerates; old footage presented as new (label "File footage, 2023"); stock music too dramatic; captions with mis-heard names; "BREAKING" on a non-breaking story. ## Make the edit ```bash python CORE/scripts/run.py presenter.mp4 --style SKILLS/news-desk --genre news-explainer --format 16:9 --smart-cuts --retakes --prompt "ONS, Reserve Bank, Priya Rao" --model base --work nx --stop-after plan python CORE/scripts/story.py outline --work nx --genre news-explainer # sentences + the genre beat map python CORE/scripts/render.py nx/plan.json --frames 1,4,8,13,15 python CORE/scripts/render.py nx/plan.json python CORE/scripts/finish.py nx/plan.json --platform youtube ``` The genre preset sets the pause cut (0.4 s). With a B-roll folder (files named after what they show) add `--broll broll/` and stop with `--stop-after broll` instead of `plan` (`--stop-after plan` exits before the B-roll step). **Reorder so the key fact leads** (when the presenter buried it): `python CORE/scripts/story.py reorder --work nx --order "s4,s1-s3,s5-s9" --genre news-explainer --format 16:9`, then `python CORE/scripts/run.py - --timeline nx/story_timeline.json --style SKILLS/news-desk --genre news-explainer --format 16:9 --model base --work nx2 --stop-after plan`. Check `nx/story_flags.md`: whole sentences only. **No presenter on camera:** record or synthesise the VO (`python CORE/scripts/tts.py --script vo.txt --out vo/voice.wav --words`; label synthetic voices as such), then follow the `faceless-explainer` skill (VO + B-roll + captions) and keep every sourcing rule above. ## Refine the plan The draft from news-desk adds a "BREAKING" ticker, a "LIVE" chip and a lower third filled with the first sentence. Replace them: - **Delete** the `chip` "LIVE" and fix the `ticker`: `{"type": "ticker", "label": "EXPLAINER", "text": "Small business · three costly mistakes · what to fix first", "y": 1000}` (y about 1000 keeps it inside a 1080 frame; 1040 clips). - **Lower third:** `{"type": "lowerthird", "name": "Priya Rao", "role": "Small-business analyst", "className": "nd-lt", "start": 0.3, "end": 4.5}`. - **Key-fact headline:** `{"type": "headline", "text": "3 mistakes, $10,000 lost", "start": 0, "end": 2.6, "x": 1560, "y": 250, "w": 600, "in": "maskUp", "out": "fade"}` — keep it off the face. - **Data + source:** `{"type": "bars", "title": "Growth after fixing all three", "items": [{"label": "Before", "value": 100}, {"label": "After", "value": 150, "highlight": true}], "suffix": "%", "start": 11.6, "end": 15.5, "x": 1580, "y": 420, "w": 600, "h": 460}` and right under it `{"type": "chip", "text": "Source: speaker's own figures, 2026", "x": 1580, "y": 700, "start": 11.6, "end": 15.5, "style": {"fontSize": "24px"}}`. Trends: `linechart`; shares: `donut`; places: `map` (a stylised dot map, **not** real geography: label it, or use a real map image with credit as an `image` element). - **Captions:** `{"type": "captions", "preset": "clean", "y": 870}` above the ticker. - **Fix split numbers in `words`:** Whisper writes "$10" ",000." and "50" "%" as separate words; the creator captions show "$10 ,000" unless you merge them (join the text, keep the first start and the last end). Check every auto `counter` for the same reason. - **Colour/sound:** `"color": {"look": "documentary-natural", "intensity": 0.5}`, `"finish": {"grain": 0}`, `"audio": {"voice": {"eq": "broadcast", "deess": true}}`, `"music": {"src": "neutral.wav", "under": 24, "duck": "speech-aware"}`. - `python CORE/scripts/editor.py review nx/plan.json --genre news-explainer` — its "no hook text in 5 s" note is solved by the headline. - **Protect identities:** blur private individuals, minors, plates and documents in footage with `"blurs": [{"x": 300, "y": 220, "w": 220, "h": 260, "start": 18.0, "end": 22.5, "pixelate": true}]` (output pixels, top-left box). It sits under the graphics, so the lower third stays sharp. ## Deliver ```bash python CORE/scripts/deliver.py nx/plan.json --to youtube,shorts --out-dir nx/deliver python CORE/scripts/qc.py nx/deliver/final_youtube.mp4 --plan nx/plan.json --platform youtube --strict python CORE/scripts/qc.py nx/deliver/final_shorts.mp4 --platform shorts # no --plan: the plan is 16:9, the file is reframed ``` `final_shorts.mp4` from a 16:9 plan is a blurred-fill reframe; for a real vertical cutdown re-run the pipeline with `--format 9:16` on the chosen section. Put every source link in the description (start from `post.md`, which only drafts from the transcript: add the sources yourself). Re-check `post.md` titles: no claim the video doesn't support. ## Check before delivering - Every number, name, date and quote on screen matches its source; every data graphic has a source chip. - No graphic covers a face; ticker and captions don't overlap; nothing outside the 5% title-safe area. - Charts honest: bars start at zero, units shown, period labelled. - Archive or file footage is labelled; nothing implies "live" or "breaking". - `qc.py`: loudness -14 LUFS / -1 dBTP, captions within 20 cps, no flash frames. - Tell the user which claims you could not verify. ## Example requests - "Make a 3-minute explainer from my recording about the new tax rule, sources are in this doc." - "Turn this market update into a 16:9 video with charts and a 45-second vertical." - "News-style breakdown of what happened at the council meeting, lower thirds for each speaker."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.