Aller au contenu

persona.yaml options

Cette page n’est pas encore traduite : voici la version anglaise.

Everything that is taste rather than technique lives in one file. Copy the example and edit it:

Terminal window
cp persona.example.yaml persona.local.yaml # in the skill folder; git-ignored

The Mac app keeps the same settings under Settings; the skill reads persona.local.yaml. Any key you leave out keeps the default shown here.

Creator persona: every taste/brand/platform default the workflows read. Copy to persona.local.yaml (gitignored) and edit; it overrides these keys. Where persona.local.yaml is read from (vstudio.config.persona_sources; all found files are deep-merged, the higher one wins per key), highest precedence first: 1. $VSTUDIO_PERSONA an explicit file (client personas, tests, one-off overrides) 2. $VSTUDIO_HOME/persona.local.yaml default ~/.config/vstudio - shared by the skill, CLI and desktop app 3. /persona.local.yaml next to this file 4. ~/.claude/skills/video-studio/persona.local.yaml the Claude Code skill checkout 5. /persona.yaml, then this file (the defaults) The desktop app runs a bundled engine with no persona next to it: it finds yours via 2 or 4.

Section What it sets Keys
creator Your name and handle (covers, end cards) and the language you speak in your videos. name, handle, language
platforms Default platform and per-platform overrides deep-merged over the built-in profiles. default, xiaohongshu, youtube
speed Playback rates used when you just say “speed it up” (hook, body, B-roll). hook, body, fast_hook, fast_body, b_roll, cjk_max_intelligible
audio Delivery loudness, narration and music levels, pause squeezing. loudness_lufs, voice_lufs, music_lufs, pause_squeeze, pause_threshold
export Frame rate, quality, audio bitrate, colour tags and the H.264 encoder. fps, crf, audio_bitrate, colour
style The design theme every overlay, card and caption uses. footage_theme, look, theme
brand Your ink and ground colours. ink, ground
watermark your mark on every export (python -m vstudio.watermark; the desk: Settings › Watermark, kind, text, image, style, color, position, size, opacity, margin, default, platforms
fonts Your own font files per role (cjk, cjk-bold, cjk-serif, serif, mono…), including one face of a .ttc.
subtitles Caption line length, keyword highlight markup and ASR term fixes. max_cjk_chars, max_en_chars, highlight_markup, term_fixes, terms, glossary
voice Voice rules for scripts and post copy. persona, rules
publish Default hashtags, named tag sets and the chapter line format. tags, tags_always, tag_sets, chapter_line, copy_lang
intake How the planner treats materials (for example burned-in captions). burned_captions
cleanup Default cleanup profile, crossfade, extra fillers and words never to cut. profile, crossfade, fillers_extra, never_cut
formats recurring formats (python -m vstudio.formats list): first-pass defaults per kind of post.
creator:
name: "" # used on covers/end cards when set
handle: ""
language: zh # subtitle + caption language for the creator's own speech
platforms:
# Full per-platform profiles (canvas, UI safe zones, caption band/size, length, LUFS, encode, cover,
# title/description limits) live in lib/vstudio/platform.py (sources: references/PLATFORMS.md).
# Anything under platforms.<name> is deep-merged over those defaults; per-orientation overrides go
# under platforms.<name>.orientations.<vertical|full|horizontal>.
default: xiaohongshu # xiaohongshu | douyin | tiktok | youtube | youtube-shorts | bilibili
xiaohongshu:
post_shape: vertical # shape when only "小红书" is said (intake / plans): vertical = 3:4 (1080x1440), full = 9:16
title_max: 20 # 小红书 counts CJK = 1, latin/digit/space = 0.5 (vstudio.config.xhs_len)
cover_aspect: "3:4" # vertical posts; horizontal posts show a 4:3 centre crop
safe_zone: {top: 240, bottom: 1660, right_lower_keepout: 160} # legacy form, 1080x1920 ("full") canvas
# orientations:
# vertical: {caption: {band: [1060, 1260], size: [56, 76]}} # e.g. bigger captions on 3:4
youtube:
title_max: 100
cover_aspect: "16:9"
# douyin:
# loudness: {lufs: -12, tp: -1.0} # a hotter short-form master, if you prefer
# length: {sweet: [20, 45]}
speed: # generic rates when the creator just says "speed it up" outside a known format; a format's
# own defaults (python -m vstudio.formats show talking-head) win: override them under formats:
hook: 1.3
body: 1.1
fast_hook: 1.6
fast_body: 1.3
b_roll: 1.1 # inserted footage (e.g. a highlights montage)
cjk_max_intelligible: 1.4
audio:
loudness_lufs: -14 # delivery loudness (two-pass loudnorm)
voice_lufs: -16 # narration stem before mixing
music_lufs: -30 # bed level before carving under the voice
pause_squeeze: 0.06 # 气口 target gap in seconds
pause_threshold: 0.35 # gaps longer than this get squeezed
export:
fps: 30
crf: 18
# h264_encoder: libx264 # libx264 | h264_videotoolbox | h264_mf; env VSTUDIO_H264_ENCODER wins (lib/vstudio/h264.py)
audio_bitrate: 192k
colour: bt709
style:
footage_theme: cinema # vlog / photo-story (drone, travel, history over aerials): no cards, serif + sans, soft
# shadows (references/AESTHETICS.md P1-P9); any theme id here, e.g. editorial
look: cinema # photo-story LOOK: cinema (full-bleed, no header) | classic (header + chapter band)
# cover: designed # designed (default: cover engine 2.0, per platform) | cinema (centred serif title) |
# bold (heavy stroked title) | scrapbook (photo-story's paper + polaroid collage)
theme: editorial # editorial | mono | soft | night | xhs-pop (classic = the old red look); see
# references/STYLE_RULES.md. Tokens can be tinted here, e.g. accent: "#3E5C76"
# Per client: client.yaml theme; per recipe: the project's theme param; per output:
# the output-edit `theme` op ("换成更高级的配色").
brand:
# accent / highlight follow style.theme. Setting brand.accent / highlight WITHOUT a style.theme keeps the
# old look (theme classic with these colours), e.g. accent: "#FF2442", highlight: "#FFD60A"
ink: "#ECEEF2"
ground: "#0B1020" # premium dark plate
# panel_theme: notes-red # legacy notes panel (notes-red | notes-yellow | teal | navy); default = the theme's card
watermark: # your mark on every export (python -m vstudio.watermark; the desk: Settings › Watermark,
# which writes $VSTUDIO_HOME/watermark.json over these keys). Nothing is drawn until you set
# a text (or an image) AND default: true; a job / clip can still say watermark: on | off.
kind: text # text (your handle as clean type) | image (your PNG logo) | generate (a badge drawn from text)
text: "" # e.g. "@yourhandle"
image: "" # kind image: path to a PNG with transparency
style: badge # generate: badge | monogram | plain
color: "#FFFFFF"
position: bottom-right # top-left | top-right | bottom-left | bottom-right (inside each platform's safe area,
# moved above the captions on vertical canvases)
size: 0.24 # fits a square of size x the frame's short side
opacity: 0.8
margin: 0.02 # extra inset inside the safe area (x short side)
default: false # true = add it to every export by default
platforms: {} # e.g. {douyin: false} = never on 抖音
fonts: {} # override any role with your own file, e.g. cjk-bold: ~/Fonts/MyFont.ttf
subtitles:
max_cjk_chars: 18
max_en_chars: 90
highlight_markup: "【】" # 【term】 is emphasised the theme's way (marker band / accent), one per line
term_fixes: {} # ASR corrections, e.g. {"cloud code": "Claude Code"}
terms: [] # her glossary: words she says that speech models get wrong (brand, product, people),
# e.g. [Reelfold, HyperFrames]: whisper is primed with them, near-miss spellings are fixed
glossary: {} # bilingual captions: fixed translations {source term: target term}, e.g. {"break the ice": "打破僵局"}
voice: # style rules for scripts and post copy (see workflows/preproduction)
persona: "practitioner sharing what they figured out"
rules:
- no em-dashes
- no parentheses in spoken lines
- one 干货 reframe per piece
- no generic CTA
publish:
tags: [] # standing hashtags, added to a post when they fit it (its words mention them)
tags_always: false # true: every post gets all of them, whatever it is about
tag_sets: {} # named alternatives picked per post with tag_set:, e.g. {art: [看展, 艺术]}
# a post/spec can also set use_persona_tags: false to use only its own tags
chapter_line: "时间线在下面,按需跳👇"
copy_lang: {intl: en} # drafted post copy (vlog / travel): intl = TikTok / X / Instagram; cn = 小红书 /
# 抖音 / B站 (default Chinese); or one platform, e.g. {tiktok: zh}. YouTube / Shorts is
# always English (no setting changes it)
intake: # one sentence + materials -> plan (python -m vstudio.intake)
burned_captions: keep # a finished edit with burned-in captions, cut again: band = crop the old captions off,
# picture in a band, new captions below; keep = leave the old captions, add none
cleanup: # shared 气口 / filler / repeat tool (python -m vstudio.cleanup, references/CLEANUP.md)
profile: standard # default profile: gentle | standard | tight (vlog / photo-story speech clips use gentle)
crossfade: 0.02 # seconds (20 ms) equal-power crossfade at every join; keep within 0.01-0.03
fillers_extra: [] # your own 口头禅, treated like 那个 / 就是, e.g. [其实, 对不对]
never_cut: [] # tokens never flagged as filler / repeat, e.g. [然后]
# policy: true # answer low-risk confirm edits automatically (batch stages do by default); learned
# # answers live in ~/.config/vstudio/cleanup_policy.json (or policy_file: <path>)
formats: {} # recurring formats (python -m vstudio.formats list): first-pass defaults per kind of post.