interview-doc
Documentary / interview editor: transcribes the interviews, builds a paper edit and a radio edit (the story in sound first), then covers every interview cut with B-roll, a second angle or a punch-in, with J/L cuts, room tone, light filler removal and a constructed-quote (Frankenb
Install
npx skills add https://github.com/Viprasol-Tech/claude-video-editing-skills/tree/main/plugins/video-studio/skills/interview-doc
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
Interview Documentary
Real people, their real words, shaped into a story that is honest and moving.
- Best for: mini-docs, brand documentaries, oral histories, profiles, NGO stories, YouTube documentaries
- Inputs: interview recordings (A-cam, optional B-cam), B-roll and archive in a folder, names/titles of everyone, music (licensed)
- Default skin:
doc-real· alternativesessay-mode,minimal-mono,news-desk - Formats: 16:9 1920x1080 master, textless version, captions (SRT/VTT), music cue sheet, 9:16 cutdowns
- Duration target: 3-30 min (shorts 3-8, festival 15-30+); genre preset
interview-doc
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 interview-doc. - The
--genrepreset sets autocut's pause cut (0.9 s) and run.py respects it; pass--max-gaponly to override. - Ethics (non-negotiable): no Frankenbites - never splice words from different sentences into a statement the person did not make. Reordering whole answers is allowed only when the meaning stays true; flag every such move to the user. Reaction shots must not change what someone meant. Release forms for everyone on camera.
How a pro cuts this
| beat | % | what happens |
|---|---|---|
| cold open | 0-5 | the most human moment or the central question |
| world / character | 5-20 | who, where, what life looked like |
| inciting question | 20-30 | what changed / what is at stake |
| development | 30-70 | the journey, obstacles, other voices |
| climax | 70-85 | the hardest or truest moment, given room to breathe |
| resolution | 85-100 | what it means now; end on an image or a line, not a summary |
- Workflow: transcribe -> paper edit (pick soundbites in
paper_edit.md) -> radio edit (sound only, no pictures; it must work with eyes closed) -> cover -> fine cut -> lock -> grade/mix. - Pacing: interview shots 5-15 s, B-roll 2-5 s; breathe 1-3 s after an emotional line (silence, a wide, B-roll with ambience).
- Cut rules: cut at the end of a thought; keep breaths before emotional lines; remove only um/uh (
keep-natural). Keep the person sounding like themselves. - Every interview cut is covered: B-roll overlapping 0.5-1 s each side, a B-cam angle, or a >= 1.2x punch-in. Otherwise it is a jump cut and
editor.py reviewflags it. - J/L cuts: next speaker's voice under the end of B-roll (J); a line continuing over a location (L); 0.3-1 s.
- Transitions: cuts; dissolves for time passing; dip to black between chapters.
- Music/SFX: score bed 26-30 LU under voices, entering on emotional turns and leaving before the important line; ambience under all B-roll; no SFX.
- Colour:
documentary-natural0.5 orfuji-eterna0.5; match A- and B-cam; skin true. - Captions: sentence-case subtitles (preset
cleanor sidecar), <= 17 cps; name supers (lower thirds) on first appearance. - Client judges: story, truthfulness, emotional arc, everyone looking dignified.
- Common mistakes: Frankenbites, uncovered jump cuts, wall-to-wall music, B-roll that illustrates literally every noun, cutting the pause before the tears, wrong name spellings.
Make the edit
Selects and paper edit per interview:
python CORE/scripts/selects.py --work int_maria --input maria.mp4 --model small
# edit int_maria/paper_edit.md: [x] keep / [ ] drop, reorder lines into the story order
python CORE/scripts/selects.py --work int_maria --apply --input maria.mp4 --format 16:9
# -> int_maria/selects_spec.json (FRANKENBITE warnings: read them, confirm meaning with the user)
Combine the interviews' selects into one radio-edit spec (doc.json: concatenate the clips of each selects_spec.json in story order; add jcut/lcut 0.3-0.8 s between speakers), then:
python CORE/scripts/run.py - --timeline doc.json --style SKILLS/doc-real --genre interview-doc \
--smart-cuts --format 16:9 --work doc --stop-after plan
python CORE/scripts/story.py outline --work doc --genre interview-doc # check the arc against the structure
python CORE/scripts/broll.py --work doc --library broll/ --frequency med --merge doc/plan.json # covers cuts first
python CORE/scripts/editor.py review doc/plan.json --genre interview-doc --fix # uncovered jump cuts, pacing
python CORE/scripts/render.py doc/plan.json --frames 5,60,180,300
python CORE/scripts/render.py doc/plan.json
python CORE/scripts/finish.py doc/plan.json --platform youtube
broll.json lists uncovered_joins: cover each by hand (a cutaways entry spanning the cut) or a speaker.zooms punch-in (scale >= 1.2) starting at the cut.
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
- Names:
{"type": "lowerthird", "name": "Maria Okafor", "role": "Nurse, Lagos", "start": 12.0, "end": 16.5}at each first appearance. - Cover cuts:
"cutaways": [{"src": "../broll/hands-sewing.mp4", "start": 41.4, "end": 45.0, "srcStart": 3}](start >= 0.5 s before the join, end >= 0.5 s after). - Punch-in:
"speaker": {"zooms": [{"start": 63.2, "end": 75.0, "scale": 1.2}]}from the cut to the next cut. - Breathing room: after the climax line, a 2-3 s wide/B-roll with ambience and no music.
- Archive/quotes:
quoteonly with the exact words;chipwith the archive credit. - Audio:
"audio": {"voice": {"denoise": "light", "eq": "warm", "deess": true}}, "roomTone": true,"music": {"src": "score.wav", "under": 28, "duck": "speech-aware"}. - Look:
"color": {"look": "documentary-natural", "intensity": 0.5, "match": "hero_frame.png"}. - Captions:
{"type": "captions", "preset": "clean", "cpsMax": 17}or none (sidecar only). - Footage clean-up: dim interviews:
--denoise light(video noise; the voice has its ownaudio.voice.denoise), e.g.python CORE/scripts/prep.py int01.mp4 --work fix/int01 --denoise lightand usefix/int01/source.mp4indoc.json; observational handheld B-roll:--stabilize; fluorescent offices:--deflicker. - Privacy: blur people who did not sign a release, screens, documents and plates with
"blurs": [{"x", "y", "w", "h", "start", "end"}](output pixels, top-left box;"pixelate": truefor a mosaic).
Deliver
python CORE/scripts/deliver.py doc/plan.json --to youtube --out-dir doc/deliver
python CORE/scripts/qc.py doc/deliver/final_youtube.mp4 --plan doc/plan.json --platform youtube
python CORE/scripts/export_edl.py doc --out-dir doc/handoff --title "DOC_TITLE" # if a colourist/mixer takes over
- Textless: copy the plan, remove lower thirds/quotes/captions,
"outputFile": "doc_textless.mp4", render + finish, deliverprores-422hqif requested. - Music cue sheet:
cue_sheet.md- for each music use: track, composer/library, licence, in/out timecode, duration (read themusicblock / premix list). - Clearances log: every B-roll/archive source and its licence; release forms on file.
Check before delivering
- No constructed quotes; every reordered answer listed and approved; reactions honest.
- No uncovered interview jump cut (
review.mdclean); room tone under every audio edit. - Names/titles/places triple-checked; captions match the audio.
- Music never covers an important line;
qc.pypasses. - Tell the user: structure with times, reordering decisions, deliverables, file paths.
Example requests
- "Make a 10-minute documentary from these three interviews and the B-roll folder."
- "Paper edit first - show me the soundbites you'd use and the story order."
- "Cover all the jump cuts in my interview with B-roll and add name supers."
Files (claude-video-editing-skills)
-
SKILL.md 8.3 KB
--- name: interview-doc description: "Documentary / interview editor: transcribes the interviews, builds a paper edit and a radio edit (the story in sound first), then covers every interview cut with B-roll, a second angle or a punch-in, with J/L cuts, room tone, light filler removal and a constructed-quote (Frankenbite) guard; delivers the film, a textless version, captions and a music cue sheet. Use when the user says: documentary edit, interview film, edit my interviews into a story, mini-doc, brand documentary, paper edit, radio edit, cover the jump cuts with B-roll, oral history. Default skin doc-real; any Video Studio skin works." license: MIT --- # Interview Documentary Real people, their real words, shaped into a story that is honest and moving. - **Best for:** mini-docs, brand documentaries, oral histories, profiles, NGO stories, YouTube documentaries - **Inputs:** interview recordings (A-cam, optional B-cam), B-roll and archive in a folder, names/titles of everyone, music (licensed) - **Default skin:** `doc-real` · alternatives `essay-mode`, `minimal-mono`, `news-desk` - **Formats:** 16:9 1920x1080 master, textless version, captions (SRT/VTT), music cue sheet, 9:16 cutdowns - **Duration target:** 3-30 min (shorts 3-8, festival 15-30+); genre preset `interview-doc` ## 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 interview-doc`. - The `--genre` preset sets autocut's pause cut (0.9 s) and run.py respects it; pass `--max-gap` only to override. - **Ethics (non-negotiable):** no Frankenbites - never splice words from different sentences into a statement the person did not make. Reordering whole answers is allowed only when the meaning stays true; flag every such move to the user. Reaction shots must not change what someone meant. Release forms for everyone on camera. ## How a pro cuts this | beat | % | what happens | | --- | --- | --- | | cold open | 0-5 | the most human moment or the central question | | world / character | 5-20 | who, where, what life looked like | | inciting question | 20-30 | what changed / what is at stake | | development | 30-70 | the journey, obstacles, other voices | | climax | 70-85 | the hardest or truest moment, given room to breathe | | resolution | 85-100 | what it means now; end on an image or a line, not a summary | - **Workflow:** transcribe -> paper edit (pick soundbites in `paper_edit.md`) -> radio edit (sound only, no pictures; it must work with eyes closed) -> cover -> fine cut -> lock -> grade/mix. - **Pacing:** interview shots 5-15 s, B-roll 2-5 s; breathe 1-3 s after an emotional line (silence, a wide, B-roll with ambience). - **Cut rules:** cut at the end of a thought; keep breaths before emotional lines; remove only um/uh (`keep-natural`). Keep the person sounding like themselves. - **Every interview cut is covered:** B-roll overlapping 0.5-1 s each side, a B-cam angle, or a >= 1.2x punch-in. Otherwise it is a jump cut and `editor.py review` flags it. - **J/L cuts:** next speaker's voice under the end of B-roll (J); a line continuing over a location (L); 0.3-1 s. - **Transitions:** cuts; dissolves for time passing; dip to black between chapters. - **Music/SFX:** score bed 26-30 LU under voices, entering on emotional turns and leaving before the important line; ambience under all B-roll; no SFX. - **Colour:** `documentary-natural` 0.5 or `fuji-eterna` 0.5; match A- and B-cam; skin true. - **Captions:** sentence-case subtitles (preset `clean` or sidecar), <= 17 cps; name supers (lower thirds) on first appearance. - **Client judges:** story, truthfulness, emotional arc, everyone looking dignified. - **Common mistakes:** Frankenbites, uncovered jump cuts, wall-to-wall music, B-roll that illustrates literally every noun, cutting the pause before the tears, wrong name spellings. ## Make the edit Selects and paper edit per interview: ```bash python CORE/scripts/selects.py --work int_maria --input maria.mp4 --model small # edit int_maria/paper_edit.md: [x] keep / [ ] drop, reorder lines into the story order python CORE/scripts/selects.py --work int_maria --apply --input maria.mp4 --format 16:9 # -> int_maria/selects_spec.json (FRANKENBITE warnings: read them, confirm meaning with the user) ``` Combine the interviews' selects into one radio-edit spec (`doc.json`: concatenate the `clips` of each `selects_spec.json` in story order; add `jcut`/`lcut` 0.3-0.8 s between speakers), then: ```bash python CORE/scripts/run.py - --timeline doc.json --style SKILLS/doc-real --genre interview-doc \ --smart-cuts --format 16:9 --work doc --stop-after plan python CORE/scripts/story.py outline --work doc --genre interview-doc # check the arc against the structure python CORE/scripts/broll.py --work doc --library broll/ --frequency med --merge doc/plan.json # covers cuts first python CORE/scripts/editor.py review doc/plan.json --genre interview-doc --fix # uncovered jump cuts, pacing python CORE/scripts/render.py doc/plan.json --frames 5,60,180,300 python CORE/scripts/render.py doc/plan.json python CORE/scripts/finish.py doc/plan.json --platform youtube ``` `broll.json` lists `uncovered_joins`: cover each by hand (a `cutaways` entry spanning the cut) or a `speaker.zooms` punch-in (scale >= 1.2) starting at the cut. 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 - **Names:** `{"type": "lowerthird", "name": "Maria Okafor", "role": "Nurse, Lagos", "start": 12.0, "end": 16.5}` at each first appearance. - **Cover cuts:** `"cutaways": [{"src": "../broll/hands-sewing.mp4", "start": 41.4, "end": 45.0, "srcStart": 3}]` (start >= 0.5 s before the join, end >= 0.5 s after). - **Punch-in:** `"speaker": {"zooms": [{"start": 63.2, "end": 75.0, "scale": 1.2}]}` from the cut to the next cut. - **Breathing room:** after the climax line, a 2-3 s wide/B-roll with ambience and no music. - **Archive/quotes:** `quote` only with the exact words; `chip` with the archive credit. - **Audio:** `"audio": {"voice": {"denoise": "light", "eq": "warm", "deess": true}}, "roomTone": true`, `"music": {"src": "score.wav", "under": 28, "duck": "speech-aware"}`. - **Look:** `"color": {"look": "documentary-natural", "intensity": 0.5, "match": "hero_frame.png"}`. - **Captions:** `{"type": "captions", "preset": "clean", "cpsMax": 17}` or none (sidecar only). - **Footage clean-up:** dim interviews: `--denoise light` (video noise; the voice has its own `audio.voice.denoise`), e.g. `python CORE/scripts/prep.py int01.mp4 --work fix/int01 --denoise light` and use `fix/int01/source.mp4` in `doc.json`; observational handheld B-roll: `--stabilize`; fluorescent offices: `--deflicker`. - **Privacy:** blur people who did not sign a release, screens, documents and plates with `"blurs": [{"x", "y", "w", "h", "start", "end"}]` (output pixels, top-left box; `"pixelate": true` for a mosaic). ## Deliver ```bash python CORE/scripts/deliver.py doc/plan.json --to youtube --out-dir doc/deliver python CORE/scripts/qc.py doc/deliver/final_youtube.mp4 --plan doc/plan.json --platform youtube python CORE/scripts/export_edl.py doc --out-dir doc/handoff --title "DOC_TITLE" # if a colourist/mixer takes over ``` - **Textless:** copy the plan, remove lower thirds/quotes/captions, `"outputFile": "doc_textless.mp4"`, render + finish, deliver `prores-422hq` if requested. - **Music cue sheet:** `cue_sheet.md` - for each music use: track, composer/library, licence, in/out timecode, duration (read the `music` block / premix list). - **Clearances log:** every B-roll/archive source and its licence; release forms on file. ## Check before delivering - No constructed quotes; every reordered answer listed and approved; reactions honest. - No uncovered interview jump cut (`review.md` clean); room tone under every audio edit. - Names/titles/places triple-checked; captions match the audio. - Music never covers an important line; `qc.py` passes. - Tell the user: structure with times, reordering decisions, deliverables, file paths. ## Example requests - "Make a 10-minute documentary from these three interviews and the B-roll folder." - "Paper edit first - show me the soundbites you'd use and the story order." - "Cover all the jump cuts in my interview with B-roll and add name supers."
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.