vox-pop
Vox-pop and street-interview editor: puts the question on screen, then many short answers grouped by intent (yes / no / surprising / funny), each trimmed to its strongest sentence, joined with quick J-cuts, name and location supers, and a closing answer that lands - in 16:9 and 9
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/vox-pop
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
Vox Pop
One question, many voices, grouped so the range of opinion becomes a story.
- Best for: street interviews, event/campus questions, brand "what do you think" pieces, social Q&A series
- Inputs: interview clips (one per person is easiest), the question, first names / locations people agreed to show, release forms
- Default skin:
news-desk· alternativesbold-bounce,doc-real,halo-cut - Formats: 16:9 1920x1080 master + 9:16 cutdown;
captions.srt(street audio: captions matter) - Duration target: 45 s - 4 min; genre preset
vox-pop
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. Preset:python CORE/scripts/editor.py genre vox-pop. - The
--genrepreset sets autocut's pause cut (0.5 s) and run.py respects it; pass--max-gaponly to override. - Ethics: an answer appears as an answer to the question shown (not to a different one); never trim "I don't think it's bad" into "it's bad"; respect people who asked not to be shown; blur bystanders' faces only if needed (plan
blurs, see Refine).
How a pro cuts this
| beat | % | what happens |
|---|---|---|
| question card | 0-6 | the question on screen (and asked aloud once) |
| group A | 6-36 | answers with one intent (e.g. "yes"), fastest and clearest first |
| group B | 36-66 | the contrasting intent |
| surprise / outlier | 66-86 | the unexpected or funniest answer |
| closing answer | 86-100 | the line that sums it up or makes people comment |
- Pacing: each answer 2-5 s (one sentence); ASL 2-5 s; runs of 3-4 rapid one-word answers are great.
- Cut rules: start each answer on its first meaningful word (drop "um, well, I guess"); end on the strong word. Keep the person's real tone.
- J/L cuts: the next person's first word 0.1-0.3 s under the end of the previous face (J) keeps it rolling; group changes get a hard cut or a question card.
- Transitions: cuts; whip only between locations if the series uses it.
- Music/SFX: light bed 26 LU under, ducked; a
popon the question card; no laugh effects. - Colour: street footage varies: auto-correct each clip,
documentary-natural0.4. - Captions: always (noisy street audio) - preset
boxed(9:16) orclean(16:9); names via supers. - Client judges: range of voices, pace, fairness, readable captions.
- Common mistakes: all answers the same intent, long answers, captions missing on wind-blown audio, names wrong, cutting someone into saying the opposite.
Make the edit
Transcribe every answer, then write the order:
python CORE/scripts/run.py - --clips vox/*.mp4 --auto story --style SKILLS/news-desk --genre vox-pop \
--smart-cuts --format 16:9 --model small --work vp --stop-after timeline
# read vp/timeline.json + vp/words.json: which answer is which person, what they said
--auto story keeps the speech regions of each clip in file order. Re-order by intent in a hand spec (vp_spec.json, times = each clip's source seconds, jcut: 0.2 on each answer), then:
python CORE/scripts/run.py - --timeline vp_spec.json --style SKILLS/news-desk --genre vox-pop \
--smart-cuts --format 16:9 --work vp2 --stop-after plan
python CORE/scripts/editor.py review vp2/plan.json --genre vox-pop --fix
python CORE/scripts/render.py vp2/plan.json --frames 1,8,20,40
python CORE/scripts/render.py vp2/plan.json
python CORE/scripts/finish.py vp2/plan.json --platform youtube
Always pass --format matching the project: run.py and make_plan.py otherwise use the skin's own format (many skins default to 9:16).
Refine the plan
- Question card:
{"type": "titlecard", "kicker": "We asked Mumbai", "title": "Would you take a 4-day week for less pay?", "start": 0, "end": 2.8}; repeat as a smallchiptop-left for the rest of the video. - Supers:
{"type": "lowerthird", "name": "Aarav", "role": "Student, Pune", "start": 3.1, "end": 5.4}per person (first names only unless they agreed to more). - Group labels (optional): a
chip"YES" / "NO" at each group's start. - Tally: a
pollordonutat the end with the real count of answers you filmed ("options": [{"label": "Yes", "pct": 62}, {"label": "No", "pct": 38}]), labelled "of the people we asked" - never as a survey result. - Captions:
{"type": "captions", "preset": "boxed", "cpsMax": 17}; fix mis-heard words inplan.words(street audio). - Sound:
"audio": {"voice": {"denoise": true, "eq": "clear"}},"roomTone": true,"music": {"src": "bed.mp3", "under": 26, "duck": "speech-aware"}. - Look:
"color": {"look": "documentary-natural", "intensity": 0.4}. - Bystanders: blur faces of people in the background who did not consent with
"blurs": [{"x": 1420, "y": 260, "w": 200, "h": 240, "start": 31.0, "end": 34.5}](output pixels, top-left box;"pixelate": truefor a mosaic; one box per answer). Check each in the--framesstills.
Deliver
python CORE/scripts/deliver.py vp2/plan.json --to youtube,reels --out-dir vp2/deliver
python CORE/scripts/qc.py vp2/deliver/final_reels.mp4 --plan vp2/plan.json --platform reels
- For a native 9:16, rebuild the spec with
"format": "9:16"and"reframe": "face"on every clip. - Keep
releases.md: person, clip file, consent given (yes/no, how).
Check before delivering
- Every answer answers the question shown and keeps its meaning.
- Names and places correct; nobody shown who declined.
- Captions on every answer, synced; audio intelligible over street noise.
qc.pypasses; tell the user: number of answers used, grouping, files.
Example requests
- "Cut my street interviews into a 90-second vox pop - question card, yes answers, no answers, funniest at the end."
- "Vertical vox pop with big captions and first-name supers."
- "Make our event interviews into a quick 'what did you think' video."
Files (claude-video-editing-skills)
-
SKILL.md 6.5 KB
--- name: vox-pop description: "Vox-pop and street-interview editor: puts the question on screen, then many short answers grouped by intent (yes / no / surprising / funny), each trimmed to its strongest sentence, joined with quick J-cuts, name and location supers, and a closing answer that lands - in 16:9 and 9:16. Use when the user says: vox pop, street interviews, man-on-the-street video, asking strangers, public opinion video, what do people think about X, campus/event interviews, survey video. Answers must not be cut to flip their meaning; everyone on camera needs consent. Default skin news-desk; any Video Studio skin works." license: MIT --- # Vox Pop One question, many voices, grouped so the range of opinion becomes a story. - **Best for:** street interviews, event/campus questions, brand "what do you think" pieces, social Q&A series - **Inputs:** interview clips (one per person is easiest), the question, first names / locations people agreed to show, release forms - **Default skin:** `news-desk` · alternatives `bold-bounce`, `doc-real`, `halo-cut` - **Formats:** 16:9 1920x1080 master + 9:16 cutdown; `captions.srt` (street audio: captions matter) - **Duration target:** 45 s - 4 min; genre preset `vox-pop` ## 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`. Preset: `python CORE/scripts/editor.py genre vox-pop`. - The `--genre` preset sets autocut's pause cut (0.5 s) and run.py respects it; pass `--max-gap` only to override. - **Ethics:** an answer appears as an answer to the question shown (not to a different one); never trim "I don't think it's bad" into "it's bad"; respect people who asked not to be shown; blur bystanders' faces only if needed (plan `blurs`, see Refine). ## How a pro cuts this | beat | % | what happens | | --- | --- | --- | | question card | 0-6 | the question on screen (and asked aloud once) | | group A | 6-36 | answers with one intent (e.g. "yes"), fastest and clearest first | | group B | 36-66 | the contrasting intent | | surprise / outlier | 66-86 | the unexpected or funniest answer | | closing answer | 86-100 | the line that sums it up or makes people comment | - **Pacing:** each answer 2-5 s (one sentence); ASL 2-5 s; runs of 3-4 rapid one-word answers are great. - **Cut rules:** start each answer on its first meaningful word (drop "um, well, I guess"); end on the strong word. Keep the person's real tone. - **J/L cuts:** the next person's first word 0.1-0.3 s under the end of the previous face (J) keeps it rolling; group changes get a hard cut or a question card. - **Transitions:** cuts; whip only between locations if the series uses it. - **Music/SFX:** light bed 26 LU under, ducked; a `pop` on the question card; no laugh effects. - **Colour:** street footage varies: auto-correct each clip, `documentary-natural` 0.4. - **Captions:** always (noisy street audio) - preset `boxed` (9:16) or `clean` (16:9); names via supers. - **Client judges:** range of voices, pace, fairness, readable captions. - **Common mistakes:** all answers the same intent, long answers, captions missing on wind-blown audio, names wrong, cutting someone into saying the opposite. ## Make the edit Transcribe every answer, then write the order: ```bash python CORE/scripts/run.py - --clips vox/*.mp4 --auto story --style SKILLS/news-desk --genre vox-pop \ --smart-cuts --format 16:9 --model small --work vp --stop-after timeline # read vp/timeline.json + vp/words.json: which answer is which person, what they said ``` `--auto story` keeps the speech regions of each clip in file order. Re-order by intent in a hand spec (`vp_spec.json`, times = each clip's source seconds, `jcut: 0.2` on each answer), then: ```bash python CORE/scripts/run.py - --timeline vp_spec.json --style SKILLS/news-desk --genre vox-pop \ --smart-cuts --format 16:9 --work vp2 --stop-after plan python CORE/scripts/editor.py review vp2/plan.json --genre vox-pop --fix python CORE/scripts/render.py vp2/plan.json --frames 1,8,20,40 python CORE/scripts/render.py vp2/plan.json python CORE/scripts/finish.py vp2/plan.json --platform youtube ``` Always pass `--format` matching the project: run.py and make_plan.py otherwise use the skin's own format (many skins default to 9:16). ## Refine the plan - **Question card:** `{"type": "titlecard", "kicker": "We asked Mumbai", "title": "Would you take a 4-day week for less pay?", "start": 0, "end": 2.8}`; repeat as a small `chip` top-left for the rest of the video. - **Supers:** `{"type": "lowerthird", "name": "Aarav", "role": "Student, Pune", "start": 3.1, "end": 5.4}` per person (first names only unless they agreed to more). - **Group labels (optional):** a `chip` "YES" / "NO" at each group's start. - **Tally:** a `poll` or `donut` at the end with the real count of answers you filmed (`"options": [{"label": "Yes", "pct": 62}, {"label": "No", "pct": 38}]`), labelled "of the people we asked" - never as a survey result. - **Captions:** `{"type": "captions", "preset": "boxed", "cpsMax": 17}`; fix mis-heard words in `plan.words` (street audio). - **Sound:** `"audio": {"voice": {"denoise": true, "eq": "clear"}}`, `"roomTone": true`, `"music": {"src": "bed.mp3", "under": 26, "duck": "speech-aware"}`. - **Look:** `"color": {"look": "documentary-natural", "intensity": 0.4}`. - **Bystanders:** blur faces of people in the background who did not consent with `"blurs": [{"x": 1420, "y": 260, "w": 200, "h": 240, "start": 31.0, "end": 34.5}]` (output pixels, top-left box; `"pixelate": true` for a mosaic; one box per answer). Check each in the `--frames` stills. ## Deliver ```bash python CORE/scripts/deliver.py vp2/plan.json --to youtube,reels --out-dir vp2/deliver python CORE/scripts/qc.py vp2/deliver/final_reels.mp4 --plan vp2/plan.json --platform reels ``` - For a native 9:16, rebuild the spec with `"format": "9:16"` and `"reframe": "face"` on every clip. - Keep `releases.md`: person, clip file, consent given (yes/no, how). ## Check before delivering - Every answer answers the question shown and keeps its meaning. - Names and places correct; nobody shown who declined. - Captions on every answer, synced; audio intelligible over street noise. - `qc.py` passes; tell the user: number of answers used, grouping, files. ## Example requests - "Cut my street interviews into a 90-second vox pop - question card, yes answers, no answers, funniest at the end." - "Vertical vox pop with big captions and first-name supers." - "Make our event interviews into a quick 'what did you think' video."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.