[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CJCTKX9EMSS8WJ3SF8VYG0":3,"skill-same:01M7CJCTKX9EMSS8WJ3SF8VYG0":231},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":16,"sourceHtmlUrl":17,"categories":18,"keywords":19,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":14,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M7CJCTKX9EMSS8WJ3SF8VYG0","vox-director",null,"Turn ONE topic into a finished Vox-style paper-collage explainer \u002F ad video, end to end on the Atlas Cloud API + local ffmpeg — script, collage keyframes, motion, voice-over, music, captions, all automated. Use this whenever the user wants a \"Vox style\" video, a paper\u002Ftorn-paper collage animation, a \"motion collage\", a narrated explainer or short ad built from AI-generated collage posters, a scrapbook-style tribute, or wants to turn a topic \u002F product \u002F person into a punchy narrated collage video — even if they don't say the word \"Vox\". Also use when reproducing Stav Zilber \u002F rom1trs \u002F Higgsfield-style collage ad workflows. Three input modalities: a topic (B-roll), a talking-head video (A-roll mode), or a single photo of a person\u002Fproduct anchored into the collage (C-roll mode). Triggers: \"vox video\", \"collage video\", \"motion collage\", \"paper collage explainer\", \"make a collage ad\", \"turn this topic into a collage video\", \"turn my photo\u002Fthis product shot into a collage video\".","josenaves\u002Fcity-video-generator","josenaves","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F137637?v=4","01M7CJ9EAPD0EHBWVAY5RBAYC8","José Naves Moura Neto","01M7CJ9EBD6R8H20KHP3NDNQ7T",0,"MIT","https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: vox-director\ndescription: >\n  Turn ONE topic into a finished Vox-style paper-collage explainer \u002F ad video, end to end\n  on the Atlas Cloud API + local ffmpeg — script, collage keyframes, motion, voice-over,\n  music, captions, all automated. Use this whenever the user wants a \"Vox style\" video,\n  a paper\u002Ftorn-paper collage animation, a \"motion collage\", a narrated explainer or short\n  ad built from AI-generated collage posters, a scrapbook-style tribute, or wants to turn\n  a topic \u002F product \u002F person into a punchy narrated collage video — even if they don't say\n  the word \"Vox\". Also use when reproducing Stav Zilber \u002F rom1trs \u002F Higgsfield-style collage\n  ad workflows.\n  Three input modalities: a topic (B-roll), a talking-head video (A-roll mode), or a single\n  photo of a person\u002Fproduct anchored into the collage (C-roll mode).\n  Triggers: \"vox video\", \"collage video\", \"motion collage\", \"paper collage\n  explainer\", \"make a collage ad\", \"turn this topic into a collage video\", \"turn my\n  photo\u002Fthis product shot into a collage video\".\n---\n\n# Vox Director\n\nTurn a one-line topic into a finished **Vox-style paper-collage video**: a bold, punchy,\nnarrated explainer\u002Fad where each beat is a torn-paper collage poster that comes alive, with\nvoice-over, music and captions. Runs on **one Atlas Cloud API key** + local **ffmpeg**.\n\nThe look is the modern editorial paper-collage popularized by Vox explainers and creators\nlike Stav Zilber \u002F rom1trs: hand-cut paper cut-outs, torn edges, tape, halftone dots,\nnewspaper clippings, bold flat color per beat, big cut-out headlines.\n\n## The core idea (read this first)\n\nThe Vox collage look and the collage motion are **two different steps**:\n\n1. **The look is born in the IMAGE step.** Each beat is a finished collage *poster* made by a\n   text-to-image model. All the collage DNA (torn paper, cut-outs, halftone, bold color,\n   headline text) lives in that image. If the image isn't a rich collage, nothing downstream\n   will save it.\n2. **The motion is added after.** By default an AI video model animates the whole poster (the\n   \"living poster\" path — simple, automated). For dramatic *piece-by-piece* assembly you cut\n   the poster into parts and drive them with the local keyframe engine (advanced path).\n\nEverything hinges on the prompts. **Before writing any image or video prompt, read\n`references\u002Fprompt-guide.md`** — it has the exact prompt structures that make the difference\nbetween \"a real Vox collage\" and \"a moving PowerPoint\".\n\n## Prerequisites (check, don't skip)\n\n- `echo \"${ATLASCLOUD_API_KEY:+set}\"` — if empty, tell the user to set it (get one at\n  https:\u002F\u002Fwww.atlascloud.ai\u002Fconsole\u002Fapi-keys) and stop.\n- `command -v ffmpeg ffprobe` — required for assembly (`brew install ffmpeg` on macOS).\n- `python3 -c \"import PIL\"` — Pillow, for captions\u002Fwatermark overlays.\n\n## Standard workflow (topic → film)\n\nThis is the default, most-automated path. Every stage is one script, all driven by a single\n`beats.json` per project under `out\u002F\u003Cproject>\u002F`.\n\n1. **Topic → beat map.** First **read `references\u002Fbeat-layer.md`** (the story layer) and pick a\n   narrative `arc` that fits the topic (`timeline` for history, `pas`\u002F`bab` for ads,\n   `how_it_works` for explainers, `man_in_hole` for transformations, …). Then write\n   `out\u002F\u003Cproject>\u002Fbeats.json` following that arc: **beat-1 headline must be a ≤3s hook**; beat\n   count per duration (30s→6–8, 60s→10–12); split each beat into **2 shots** (wide+detail) with\n   **per-shot `camera_move` VARIED across adjacent beats** (never repeat; `static` on the payoff)\n   and **rich `element_motion`** (see step 4). Each beat: `narration`, `title_cn`\u002F`title_en`,\n   `scene`, `bg`, `feel`, `hook`. This draft is the **first mandatory approval gate** — show the\n   user the beat map before generating (the aspect-routing approximation in step 4 is the other\n   one). Examples in `examples\u002F`.\n\n2. **Pick the visual style (hybrid — do this BEFORE keyframes).** Do not reuse one house style\n   for every topic. Read `references\u002Fprompt-guide.md` (§5 theme presets); pick 3–4 **theme presets**\n   (`styles.THEME_PRESETS`: `american-retro`, `swiss-modern`, `punk-zine`,\n   `soviet-constructivist`, `wpa-propaganda`, `70s-groovy`, `chinese-ink`, `atomic-age`,\n   `newsprint-editorial`) that fit\n   the topic's era\u002Fculture\u002Ftone — **or compose a custom theme** by mixing the prompt-guide dimensions\n   (medium\u002Fera\u002Fpalette\u002Ftype\u002Ffinish) when none fit. Match the topic, **not** the language (an\n   English film on Chinese history should look Chinese). A theme bundles the whole LOOK layer\n   (idiom+palette+type+finish+mood+motion). Run a bake-off and let the user pick by eye — AI\n   proposes, the library is the quality floor, the human decides. Set the pick as `\"theme\"`:\n   `python3 scripts\u002Fstyle_bakeoff.py out\u002F\u003Cproject> american-retro,swiss-modern,punk-zine,atomic-age`\n   Set the chosen name as `\"collage_style\"` in beats.json (keyframes.py reads it).\n\n3. **Keyframes (the collage look).** `python3 scripts\u002Fkeyframes.py out\u002F\u003Cproject>`\n   Generates one collage poster per beat\u002Fshot with **google\u002Fnano-banana-2\u002Ftext-to-image**,\n   headline text baked in. Compose prompts with the 5-part structure in\n   `references\u002Fprompt-guide.md`. Verify each poster looks like a *real layered collage*\n   before animating — re-roll cheap ($0.08) here rather than paying to animate a weak image.\n\n4. **Motion.** `python3 scripts\u002Fclips.py out\u002F\u003Cproject>`\n   Animates each poster with **google\u002Fgemini-omni-flash\u002Fimage-to-video**. Two independent axes\n   (see `references\u002Fbeat-layer.md` §3, tested on our stack):\n   • **`camera_move`** — ONE move per shot. Safe\u002Fdefault: `{static, push_in, pull_out, pan, tilt,\n     parallax}`. **Bold\u002Fexperimental** `{orbit, dolly_zoom, roll, whip}` are **available, not\n     banned** — they can warp the flat art, so pair with `constraints: loose` and **re-roll**.\n     Any custom phrase also passes through.\n   • **`element_motion`** — where the energy lives; **AI writes it per beat to fit that scene** (not a\n     template). Make it RICH (several elements moving) — be bold. A **hero element flying across\n     the frame** (paper bird\u002Fplane\u002Fcoins) is a great **occasional** punch on a key beat, **not\n     every shot** (a flyer in every frame reads as a formula).\n   `motion_style` = amplitude `calm | punchy | max` (the theme sets a default). **`constraints`**\n   = `strict` (default: defect guards on — flat-2D, one-way, no-morph; best for clean text-heavy\n   explainers) or `loose` (let the model explore 3D\u002Fbold moves; re-roll the misses). **Headline\n   text is hard-protected only on shots that have a title** (detail shots without a headline are\n   free to go wild). For **real people \u002F brand logos**, Omni & Seedance refuse — set\n   `\"video_model\": \"kwaivgi\u002Fkling-video-o3-pro\u002Fimage-to-video\"`.\n   **Aspect routing** (`styles.resolve_video_aspect`, second approval gate): `clips.py` resolves\n   `doc[\"aspect\"]` against the chosen `video_model`'s own supported ratios — exact match wins;\n   Omni is 16:9\u002F9:16 only, Kling reference-to-video adds 1:1, Kling image-to-video\u002Fvideo-edit and\n   Seedance just follow the input\u002Fratio param. When there's no exact match it picks the nearest\n   ratio but **stops and asks you to confirm** (set `\"aspect_approx_confirmed\": true` once you\n   have) rather than silently reframing the film — every clip in one run shares the same resolved\n   aspect so the finished film is never mixed.\n\n5. **Voice + music.** `python3 scripts\u002Faudio.py out\u002F\u003Cproject>`\n   One consistent narrator via **xai\u002Ftts-v1** + instrumental BGM via **minimax\u002Fmusic-2.6**.\n   **Pick `voice_id` to fit the topic + language** (don't just keep the default) — see\n   `references\u002Fvoices.md` for the full roster (5 multilingual + ~66 native voices by language,\n   with gender). Default `leo` (male, documentary). To narrate in a REAL person's own voice\n   (the presenter of a C-roll photo, a brand voice), set `voice.clone_ref` to a local audio\n   sample — narration switches to seed-audio voice cloning with a pinned-speaker,\n   studio-clean template that keeps timing beat-stable (see gotchas: never hand seed-audio\n   bare narration without that pin).\n\n6. **Assemble.** `python3 scripts\u002Fassemble.py out\u002F\u003Cproject>`\n   ffmpeg: normalize + concat all shots, lay the single narration ducked under the music,\n   burn captions timed per beat, add the watermark. Output `out\u002F\u003Cproject>\u002Ffinal.mp4`.\n\n7. **Verify.** You can't read an mp4 directly — extract frames to jpg and look:\n   `ffmpeg -ss \u003Ct> -i final.mp4 -vf \"scale=640:-1,format=yuvj420p\" -frames:v 1 f.jpg`\n\n### Cadence — how long shots should be\n\nA common mistake is one long shot per beat. On a 9:16 \u002F social piece especially, a static\n10s shot reads as dead air. Aim for a **cut every ~4–6 seconds**:\n\n- **Shots run 3–6s; never let a single shot exceed ~7s** — beyond that the AI motion has\n  nowhere to go and it feels static.\n- **A beat's narration is ~8–10s, so give each beat 2 shots** (a *wide* establishing shot with\n  the headline + a *detail* cut-in without it). The narration plays continuously across both;\n  the visual cuts mid-sentence. This is the single biggest rhythm win.\n- So a ~60s film is typically **~6 beats × 2 shots × ~5s = 12 shots**, not 6 × 10s.\n- Reuse the wide keyframe as shot `a`; generate a tighter detail scene for shot `b`.\n  `keyframes.py` skips any shot that already has a `keyframe_url`, so adding `b` shots and\n  re-running only generates the new ones.\n\nAdd a `shots` array to each beat (see schema). Give each shot its own short `scene` and\n`motion`; set `\"title\": true` only on the wide shot so the headline shows once per beat.\n\n## A-roll mode (talking-head → collage)\n\nThe standard workflow above is **B-roll**: a topic becomes AI-generated collage posters\nthat get animated. **A-roll is the reverse case** — the user already has a real recorded\ntalking-head video (a presenter speaking to camera) and wants it *itself* turned into the\ncollage look, keeping their actual performance (face, lip movement, gestures) intact. There\nis no poster to generate; the \"keyframe\" is the presenter's own footage. Use A-roll when the\nuser gives you a video file of themselves\u002Fa presenter talking, not a topic to write from\nscratch.\n\n1. **Transcribe + auto-segment.** `python3 scripts\u002Fasr_beats.py \u003Cproject_dir> \u003Csource.mp4>`\n   Runs xai\u002Fstt-v1 on the source's own audio and cuts it into beats at sentence-ending\n   punctuation or natural pause gaps (never exceeding ~9.5s, under Omni\u002FKling video-edit's\n   10s per-call cap). Writes `beats.json` with each beat's `start`\u002F`end`\u002F`text` — **this is\n   the same mandatory approval gate as the B-roll beat map**: review it, set `\"theme\"` (run\n   `style_bakeoff.py` the same way — the presenter's segment works fine as the bake-off\n   source), and optionally fill in a `content_beats` string per beat (a sticker\u002Fstamp idea\n   to layer in) before generating anything.\n\n2. **Generate.** `python3 scripts\u002Faroll_clips.py \u003Cproject_dir> [only_ids]`\n   Cuts each beat's time range out of the source, uploads it, and re-styles it with a\n   **photographic paper-cutout sticker** treatment on the presenter — her real likeness,\n   lip movement, eye-line and gestures follow the source frame-for-frame; only the\n   silhouette edge and the world around her are paper-collage. Default model is\n   `google\u002Fgemini-omni-flash\u002Fvideo-edit`; any beat it rejects automatically retries on\n   `bytedance\u002Fseedance-2.0\u002Freference-to-video` (set via `video_model`\u002F`video_model_fallback`\n   in beats.json). **Never ask the model to redraw or halftone-texture the face itself** —\n   that gets rejected regardless of how the prompt is worded (tried both a strong and a\n   softened phrasing; both failed). Uses the same aspect-routing confirm gate as `clips.py`.\n\n3. **Assemble.** `python3 scripts\u002Faroll_assemble.py \u003Cproject_dir>`\n   Muxes each generated clip with the *original* beat segment's own audio (never whatever\n   audio the video model produced) so lip-sync is guaranteed regardless of which model\n   handled that beat, normalizes every beat to one canvas, and concats into `final.mp4`.\n\n## C-roll mode (one photo → collage)\n\nThe third input modality — \"cutout roll\". A-roll re-styles a talking-head VIDEO; B-roll\ngenerates everything from a topic; **C-roll takes a single still PHOTO** (a selfie, an\navatar card, a product shot) and anchors it inside the collage world: the subject is cut\nout as a PHOTOGRAPHIC sticker — never redrawn — and per-beat posters are generated around\nit with an image-EDIT model, then animated through the normal clip stage. Use C-roll when\nthe user gives you one photo and a topic: a personal explainer fronted by their own face,\nor a collage ad built around a real product shot (validated on both, 2026-07-17).\n\n1. **Beat map.** Same as B-roll (`references\u002Fbeat-layer.md`, same approval gate), plus the\n   C-roll fields in beats.json: `\"mode\": \"croll\"`, `\"anchor_photo\"`, `\"croll_subject\"`\n   (`portrait` | `product`), and `subject_wardrobe` (portrait — lock the outfit or the\n   paper-doll body drifts) or `subject_desc` (product). Set `\"title\": false` on shots —\n   C-roll posters carry no headline; text belongs to captions. If there is no separate\n   script, transcribe\u002Fderive narration first and let the audio's ASR timestamps define the\n   beats (audio-first, like A-roll — not text-first like B-roll).\n\n2. **Anchored keyframes.** `python3 scripts\u002Fcroll_keyframes.py \u003Cproject_dir>`\n   Uploads the photo once and generates one anchored poster per shot via\n   `google\u002Fnano-banana-2\u002Fedit` (fallback `openai\u002Fgpt-image-2\u002Fedit`). Portraits get a\n   photographic face + illustrated paper-doll body; products get a pixel-faithful sticker\n   with label typography intact. Prompt rules that are baked in (all three cost a re-run to\n   learn): poses\u002Fexpressions go to the BODY only — asking for a wink redraws the face;\n   halftone must be scoped to the background or it bleeds onto skin; portrait clothing must\n   be locked explicitly. The script also writes `anchor_freeze` into beats.json.\n\n3. **Animate + audio + assemble.** Standard `clips.py` → `audio.py` → `assemble.py`.\n   `clips.py` injects the `anchor_freeze` guard into every motion prompt — without it the\n   video stage can re-letter a product label (observed: \"PARFUM\" → \"PAREUM\") or re-time a\n   face. For narration in the subject's own voice, set `voice.clone_ref` (see Voice + music\n   above); derive stamp\u002Fsnap-zoom timing from the narration's ASR word timestamps\n   (`asr_beats.py` works on any audio, not just A-roll footage).\n\n## beats.json schema\n\n```json\n{\n  \"project\": \"my-film\", \"topic\": \"...\", \"language\": \"en\",\n  \"aspect\": \"9:16\",                       \u002F\u002F 16:9 | 9:16 | 1:1 | 3:4\n  \"style\": \"collage\",\n  \"provider\": \"atlas_cloud\",              \u002F\u002F media backend — default; pluggable (scripts\u002Fprovider.py)\n  \"theme\": \"american-retro\",              \u002F\u002F THEME_PRESET (styles.THEME_PRESETS) — the LOOK layer\n  \"arc\": \"timeline\",                      \u002F\u002F narrative arc (beat-layer.md) — the STORY skeleton\n  \"video_model\": \"google\u002Fgemini-omni-flash\u002Fimage-to-video\",  \u002F\u002F Kling for real people\n  \"image_model\": \"google\u002Fnano-banana-2\u002Ftext-to-image\",       \u002F\u002F keyframes; or openai\u002Fgpt-image-2\u002Ftext-to-image\n  \"image_resolution\": \"1k\",               \u002F\u002F 1k (default) | 2k | 4k\n  \"video_resolution\": \"720p\",             \u002F\u002F 720p (default); Seedance also 480p\u002F1080p (Omni is 720p-only)\n  \"motion_style\": \"punchy\",               \u002F\u002F amplitude: calm | punchy | max (theme sets a default)\n  \"constraints\": \"strict\",                \u002F\u002F strict = defect guards on | loose = let AI explore + re-roll\n  \"voice\": {\"voice_id\": \"leo\", \"language\": \"en\", \"speed\": 1.0},  \u002F\u002F pick per topic\u002Flanguage — see references\u002Fvoices.md\n                                          \u002F\u002F + optional \"clone_ref\": \"path\u002Fto\u002Fsample.mp3\" (clone that voice via seed-audio)\n                                          \u002F\u002F   and \"persona\": \"YouTube tutorial creator\" (delivery style for cloned VO)\n  \"music\": \"epic cinematic orchestral, instrumental, no vocals\",\n  \"mix\": {\"music\": 0.6, \"voice\": 1.25},   \u002F\u002F audio balance — optional; these are the defaults (BGM ducks under the VO)\n  \"caption_style\": \"white\",               \u002F\u002F white (default: clean white subtitle) | paper (cream cut-out collage look)\n  \"captions\": true,                       \u002F\u002F false = no burned-in captions (deliver clean, subtitle in post)\n  \"watermark\": \"Made with Atlas Cloud\",\n  \"mode\": \"croll\",                        \u002F\u002F C-roll only — plus the four fields below\n  \"anchor_photo\": \"path\u002Fto\u002Fphoto.png\",    \u002F\u002F C-roll: the still to anchor (person or product)\n  \"croll_subject\": \"portrait\",            \u002F\u002F C-roll: portrait | product\n  \"subject_wardrobe\": \"a cream knitted sweater and charcoal trousers\",  \u002F\u002F C-roll portrait: outfit lock\n  \"subject_desc\": \"the perfume bottle\",   \u002F\u002F C-roll product: short noun phrase for the sticker\n  \"beats\": [\n    {\n      \"id\": 1, \"title_cn\": \"\", \"title_en\": \"BEFORE MONEY\",\n      \"bg\": \"earthy clay tan\", \"feel\": \"ancient, humble\", \"hook\": \"surprising_stat\",\n      \"narration\": \"For most of history, there was no money...\",\n      \"shots\": [\n        \u002F\u002F shot_size: EST_WIDE|WIDE|MEDIUM|CLOSE|DETAIL ; camera_move: static|push_in|\n        \u002F\u002F pull_out|pan|tilt|parallax (flat-safe only) — VARY per adjacent beat, static for payoff\n        {\"id\": \"a\", \"dur\": 5, \"title\": true,  \"shot_size\": \"WIDE\", \"camera_move\": \"push_in\",\n         \"scene\": \"...wide establishing collage...\",\n         \"element_motion\": \"traders gesture, goat bobs, a paper bird flaps across the frame, coins scatter\"},\n        {\"id\": \"b\", \"dur\": 5, \"title\": false, \"shot_size\": \"CLOSE\", \"camera_move\": \"parallax\",\n         \"scene\": \"...close cut-in detail...\",\n         \"element_motion\": \"the exchanged goods slide together, halftone pulses\"}\n      ]\n    }\n  ]\n}\n```\n`theme`+`arc` set the two big layers; `element_motion` per shot is the energy (make it rich — see\nbelow). `motion`\u002F`collage_style`\u002F`era` are still read for back-compat.\n\n## Model selection (always verify IDs live)\n\nModel IDs change — fetch the live list first: `GET https:\u002F\u002Fapi.atlascloud.ai\u002Fapi\u002Fv1\u002Fmodels`\n(no auth; keep only `display_console: true`). Defaults that work today:\n\n| Job | Model | Note |\n|---|---|---|\n| Keyframe \u002F collage poster | `google\u002Fnano-banana-2\u002Ftext-to-image` | default; renders CN+EN text well; `image_resolution` 1k\u002F2k\u002F4k |\n| Keyframe (alternative) | `openai\u002Fgpt-image-2\u002Ftext-to-image` | set via `image_model`; size+quality auto-mapped from aspect+resolution |\n| Cut out an element | `youchuan\u002Fv8.1\u002Fremove-background` | advanced path only |\n| Animate (non-real content) | `google\u002Fgemini-omni-flash\u002Fimage-to-video` | keeps text stable, layered motion |\n| Animate (**real people \u002F brands**) | `kwaivgi\u002Fkling-video-o3-pro\u002Fimage-to-video` | Omni & Seedance BLOCK celebrities |\n| Narration | `xai\u002Ftts-v1` | clean, multilingual, `voice_id` |\n| Music | `minimax\u002Fmusic-2.6` | `is_instrumental: true` |\n\nSee `references\u002Fmodels-and-gotchas.md` for the full model-choice reasoning and every\nAPI \u002F ffmpeg gotcha (auth header, curl downloads, no-libass captions, content blocks, etc.).\nRead it before debugging any failure — most failures are already documented there.\n\n**Backends are pluggable.** Every API call goes through a **provider** (`scripts\u002Fprovider.py`);\nAtlas Cloud is the default and only backend today. Set `\"provider\"` in beats.json to route to a\ndifferent backend once one is added — the stage scripts don't change. `scripts\u002Fprovider.py`'s\n`run_jobs()` also does the submit\u002Fpoll with **auto-resubmit on a stalled or failed job**.\n\n## Advanced: element-level motion collage\n\nThe standard path animates the *whole* poster (great, automated, \"living poster\"). For the\ndramatic **pieces-fly-in-and-assemble** motion collage (à la cr7v2), or to animate **real\npeople with full control and zero content filters**, cut each poster into independent\nelements and drive them with the local keyframe engine (no video model needed).\n\nRead `references\u002Flocal-engine.md`. In short: `extract_elements.py` (crop + background-removal\n+ residue\u002Ferase cleanup) → `motion.py` (Layer + keyframes, `fly_in`\u002F`slap`\u002F`drop`\u002F`pop_settle`\neasings, procedural confetti\u002Fstarburst, camera zoom+shake+whip, frame render). Pieces fly\nback to their **original positions** on a blurred-placeholder backdrop, so the assembled\nframe reconstructs the original poster.\n\n## Editions\n\n- **Auto edition** (this skill): topic in, film out, all on Atlas.\n- **Manual prompt-pack**: if the user isn't on Atlas, just produce the beat map + the per-beat\n  image prompts + the per-clip motion prompts + the narration script for them to paste into\n  any generator. The creative engine (the prompts) is identical.\n","# Vox Director\n\nTurn a one-line topic into a finished **Vox-style paper-collage video**: a bold, punchy,\nnarrated explainer\u002Fad where each beat is a torn-paper collage poster that comes alive, with\nvoice-over, music and captions. Runs on **one Atlas Cloud API key** + local **ffmpeg**.\n\nThe look is the modern editorial paper-collage popularized by Vox explainers and creators\nlike Stav Zilber \u002F rom1trs: hand-cut paper cut-outs, torn edges, tape, halftone dots,\nnewspaper clippings, bold flat color per beat, big cut-out headlines.\n\n## The core idea (read this first)\n\nThe Vox collage look and the collage motion are **two different steps**:\n\n1. **The look is born in the IMAGE step.** Each beat is a finished collage *poster* made by a\n   text-to-image model. All the collage DNA (torn paper, cut-outs, halftone, bold color,\n   headline text) lives in that image. If the image isn't a rich collage, nothing downstream\n   will save it.\n2. **The motion is added after.** By default an AI video model animates the whole poster (the\n   \"living poster\" path — simple, automated). For dramatic *piece-by-piece* assembly you cut\n   the poster into parts and drive them with the local keyframe engine (advanced path).\n\nEverything hinges on the prompts. **Before writing any image or video prompt, read\n`references\u002Fprompt-guide.md`** — it has the exact prompt structures that make the difference\nbetween \"a real Vox collage\" and \"a moving PowerPoint\".\n\n## Prerequisites (check, don't skip)\n\n- `echo \"${ATLASCLOUD_API_KEY:+set}\"` — if empty, tell the user to set it (get one at\n  https:\u002F\u002Fwww.atlascloud.ai\u002Fconsole\u002Fapi-keys) and stop.\n- `command -v ffmpeg ffprobe` — required for assembly (`brew install ffmpeg` on macOS).\n- `python3 -c \"import PIL\"` — Pillow, for captions\u002Fwatermark overlays.\n\n## Standard workflow (topic → film)\n\nThis is the default, most-automated path. Every stage is one script, all driven by a single\n`beats.json` per project under `out\u002F\u003Cproject>\u002F`.\n\n1. **Topic → beat map.** First **read `references\u002Fbeat-layer.md`** (the story layer) and pick a\n   narrative `arc` that fits the topic (`timeline` for history, `pas`\u002F`bab` for ads,\n   `how_it_works` for explainers, `man_in_hole` for transformations, …). Then write\n   `out\u002F\u003Cproject>\u002Fbeats.json` following that arc: **beat-1 headline must be a ≤3s hook**; beat\n   count per duration (30s→6–8, 60s→10–12); split each beat into **2 shots** (wide+detail) with\n   **per-shot `camera_move` VARIED across adjacent beats** (never repeat; `static` on the payoff)\n   and **rich `element_motion`** (see step 4). Each beat: `narration`, `title_cn`\u002F`title_en`,\n   `scene`, `bg`, `feel`, `hook`. This draft is the **first mandatory approval gate** — show the\n   user the beat map before generating (the aspect-routing approximation in step 4 is the other\n   one). Examples in `examples\u002F`.\n\n2. **Pick the visual style (hybrid — do this BEFORE keyframes).** Do not reuse one house style\n   for every topic. Read `references\u002Fprompt-guide.md` (§5 theme presets); pick 3–4 **theme presets**\n   (`styles.THEME_PRESETS`: `american-retro`, `swiss-modern`, `punk-zine`,\n   `soviet-constructivist`, `wpa-propaganda`, `70s-groovy`, `chinese-ink`, `atomic-age`,\n   `newsprint-editorial`) that fit\n   the topic's era\u002Fculture\u002Ftone — **or compose a custom theme** by mixing the prompt-guide dimensions\n   (medium\u002Fera\u002Fpalette\u002Ftype\u002Ffinish) when none fit. Match the topic, **not** the language (an\n   English film on Chinese history should look Chinese). A theme bundles the whole LOOK layer\n   (idiom+palette+type+finish+mood+motion). Run a bake-off and let the user pick by eye — AI\n   proposes, the library is the quality floor, the human decides. Set the pick as `\"theme\"`:\n   `python3 scripts\u002Fstyle_bakeoff.py out\u002F\u003Cproject> american-retro,swiss-modern,punk-zine,atomic-age`\n   Set the chosen name as `\"collage_style\"` in beats.json (keyframes.py reads it).\n\n3. **Keyframes (the collage look).** `python3 scripts\u002Fkeyframes.py out\u002F\u003Cproject>`\n   Generates one collage poster per beat\u002Fshot with **google\u002Fnano-banana-2\u002Ftext-to-image**,\n   headline text baked in. Compose prompts with the 5-part structure in\n   `references\u002Fprompt-guide.md`. Verify each poster looks like a *real layered collage*\n   before animating — re-roll cheap ($0.08) here rather than paying to animate a weak image.\n\n4. **Motion.** `python3 scripts\u002Fclips.py out\u002F\u003Cproject>`\n   Animates each poster with **google\u002Fgemini-omni-flash\u002Fimage-to-video**. Two independent axes\n   (see `references\u002Fbeat-layer.md` §3, tested on our stack):\n   • **`camera_move`** — ONE move per shot. Safe\u002Fdefault: `{static, push_in, pull_out, pan, tilt,\n     parallax}`. **Bold\u002Fexperimental** `{orbit, dolly_zoom, roll, whip}` are **available, not\n     banned** — they can warp the flat art, so pair with `constraints: loose` and **re-roll**.\n     Any custom phrase also passes through.\n   • **`element_motion`** — where the energy lives; **AI writes it per beat to fit that scene** (not a\n     template). Make it RICH (several elements moving) — be bold. A **hero element flying across\n     the frame** (paper bird\u002Fplane\u002Fcoins) is a great **occasional** punch on a key beat, **not\n     every shot** (a flyer in every frame reads as a formula).\n   `motion_style` = amplitude `calm | punchy | max` (the theme sets a default). **`constraints`**\n   = `strict` (default: defect guards on — flat-2D, one-way, no-morph; best for clean text-heavy\n   explainers) or `loose` (let the model explore 3D\u002Fbold moves; re-roll the misses). **Headline\n   text is hard-protected only on shots that have a title** (detail shots without a headline are\n   free to go wild). For **real people \u002F brand logos**, Omni & Seedance refuse — set\n   `\"video_model\": \"kwaivgi\u002Fkling-video-o3-pro\u002Fimage-to-video\"`.\n   **Aspect routing** (`styles.resolve_video_aspect`, second approval gate): `clips.py` resolves\n   `doc[\"aspect\"]` against the chosen `video_model`'s own supported ratios — exact match wins;\n   Omni is 16:9\u002F9:16 only, Kling reference-to-video adds 1:1, Kling image-to-video\u002Fvideo-edit and\n   Seedance just follow the input\u002Fratio param. When there's no exact match it picks the nearest\n   ratio but **stops and asks you to confirm** (set `\"aspect_approx_confirmed\": true` once you\n   have) rather than silently reframing the film — every clip in one run shares the same resolved\n   aspect so the finished film is never mixed.\n\n5. **Voice + music.** `python3 scripts\u002Faudio.py out\u002F\u003Cproject>`\n   One consistent narrator via **xai\u002Ftts-v1** + instrumental BGM via **minimax\u002Fmusic-2.6**.\n   **Pick `voice_id` to fit the topic + language** (don't just keep the default) — see\n   `references\u002Fvoices.md` for the full roster (5 multilingual + ~66 native voices by language,\n   with gender). Default `leo` (male, documentary). To narrate in a REAL person's own voice\n   (the presenter of a C-roll photo, a brand voice), set `voice.clone_ref` to a local audio\n   sample — narration switches to seed-audio voice cloning with a pinned-speaker,\n   studio-clean template that keeps timing beat-stable (see gotchas: never hand seed-audio\n   bare narration without that pin).\n\n6. **Assemble.** `python3 scripts\u002Fassemble.py out\u002F\u003Cproject>`\n   ffmpeg: normalize + concat all shots, lay the single narration ducked under the music,\n   burn captions timed per beat, add the watermark. Output `out\u002F\u003Cproject>\u002Ffinal.mp4`.\n\n7. **Verify.** You can't read an mp4 directly — extract frames to jpg and look:\n   `ffmpeg -ss \u003Ct> -i final.mp4 -vf \"scale=640:-1,format=yuvj420p\" -frames:v 1 f.jpg`\n\n### Cadence — how long shots should be\n\nA common mistake is one long shot per beat. On a 9:16 \u002F social piece especially, a static\n10s shot reads as dead air. Aim for a **cut every ~4–6 seconds**:\n\n- **Shots run 3–6s; never let a single shot exceed ~7s** — beyond that the AI motion has\n  nowhere to go and it feels static.\n- **A beat's narration is ~8–10s, so give each beat 2 shots** (a *wide* establishing shot with\n  the headline + a *detail* cut-in without it). The narration plays continuously across both;\n  the visual cuts mid-sentence. This is the single biggest rhythm win.\n- So a ~60s film is typically **~6 beats × 2 shots × ~5s = 12 shots**, not 6 × 10s.\n- Reuse the wide keyframe as shot `a`; generate a tighter detail scene for shot `b`.\n  `keyframes.py` skips any shot that already has a `keyframe_url`, so adding `b` shots and\n  re-running only generates the new ones.\n\nAdd a `shots` array to each beat (see schema). Give each shot its own short `scene` and\n`motion`; set `\"title\": true` only on the wide shot so the headline shows once per beat.\n\n## A-roll mode (talking-head → collage)\n\nThe standard workflow above is **B-roll**: a topic becomes AI-generated collage posters\nthat get animated. **A-roll is the reverse case** — the user already has a real recorded\ntalking-head video (a presenter speaking to camera) and wants it *itself* turned into the\ncollage look, keeping their actual performance (face, lip movement, gestures) intact. There\nis no poster to generate; the \"keyframe\" is the presenter's own footage. Use A-roll when the\nuser gives you a video file of themselves\u002Fa presenter talking, not a topic to write from\nscratch.\n\n1. **Transcribe + auto-segment.** `python3 scripts\u002Fasr_beats.py \u003Cproject_dir> \u003Csource.mp4>`\n   Runs xai\u002Fstt-v1 on the source's own audio and cuts it into beats at sentence-ending\n   punctuation or natural pause gaps (never exceeding ~9.5s, under Omni\u002FKling video-edit's\n   10s per-call cap). Writes `beats.json` with each beat's `start`\u002F`end`\u002F`text` — **this is\n   the same mandatory approval gate as the B-roll beat map**: review it, set `\"theme\"` (run\n   `style_bakeoff.py` the same way — the presenter's segment works fine as the bake-off\n   source), and optionally fill in a `content_beats` string per beat (a sticker\u002Fstamp idea\n   to layer in) before generating anything.\n\n2. **Generate.** `python3 scripts\u002Faroll_clips.py \u003Cproject_dir> [only_ids]`\n   Cuts each beat's time range out of the source, uploads it, and re-styles it with a\n   **photographic paper-cutout sticker** treatment on the presenter — her real likeness,\n   lip movement, eye-line and gestures follow the source frame-for-frame; only the\n   silhouette edge and the world around her are paper-collage. Default model is\n   `google\u002Fgemini-omni-flash\u002Fvideo-edit`; any beat it rejects automatically retries on\n   `bytedance\u002Fseedance-2.0\u002Freference-to-video` (set via `video_model`\u002F`video_model_fallback`\n   in beats.json). **Never ask the model to redraw or halftone-texture the face itself** —\n   that gets rejected regardless of how the prompt is worded (tried both a strong and a\n   softened phrasing; both failed). Uses the same aspect-routing confirm gate as `clips.py`.\n\n3. **Assemble.** `python3 scripts\u002Faroll_assemble.py \u003Cproject_dir>`\n   Muxes each generated clip with the *original* beat segment's own audio (never whatever\n   audio the video model produced) so lip-sync is guaranteed regardless of which model\n   handled that beat, normalizes every beat to one canvas, and concats into `final.mp4`.\n\n## C-roll mode (one photo → collage)\n\nThe third input modality — \"cutout roll\". A-roll re-styles a talking-head VIDEO; B-roll\ngenerates everything from a topic; **C-roll takes a single still PHOTO** (a selfie, an\navatar card, a product shot) and anchors it inside the collage world: the subject is cut\nout as a PHOTOGRAPHIC sticker — never redrawn — and per-beat posters are generated around\nit with an image-EDIT model, then animated through the normal clip stage. Use C-roll when\nthe user gives you one photo and a topic: a personal explainer fronted by their own face,\nor a collage ad built around a real product shot (validated on both, 2026-07-17).\n\n1. **Beat map.** Same as B-roll (`references\u002Fbeat-layer.md`, same approval gate), plus the\n   C-roll fields in beats.json: `\"mode\": \"croll\"`, `\"anchor_photo\"`, `\"croll_subject\"`\n   (`portrait` | `product`), and `subject_wardrobe` (portrait — lock the outfit or the\n   paper-doll body drifts) or `subject_desc` (product). Set `\"title\": false` on shots —\n   C-roll posters carry no headline; text belongs to captions. If there is no separate\n   script, transcribe\u002Fderive narration first and let the audio's ASR timestamps define the\n   beats (audio-first, like A-roll — not text-first like B-roll).\n\n2. **Anchored keyframes.** `python3 scripts\u002Fcroll_keyframes.py \u003Cproject_dir>`\n   Uploads the photo once and generates one anchored poster per shot via\n   `google\u002Fnano-banana-2\u002Fedit` (fallback `openai\u002Fgpt-image-2\u002Fedit`). Portraits get a\n   photographic face + illustrated paper-doll body; products get a pixel-faithful sticker\n   with label typography intact. Prompt rules that are baked in (all three cost a re-run to\n   learn): poses\u002Fexpressions go to the BODY only — asking for a wink redraws the face;\n   halftone must be scoped to the background or it bleeds onto skin; portrait clothing must\n   be locked explicitly. The script also writes `anchor_freeze` into beats.json.\n\n3. **Animate + audio + assemble.** Standard `clips.py` → `audio.py` → `assemble.py`.\n   `clips.py` injects the `anchor_freeze` guard into every motion prompt — without it the\n   video stage can re-letter a product label (observed: \"PARFUM\" → \"PAREUM\") or re-time a\n   face. For narration in the subject's own voice, set `voice.clone_ref` (see Voice + music\n   above); derive stamp\u002Fsnap-zoom timing from the narration's ASR word timestamps\n   (`asr_beats.py` works on any audio, not just A-roll footage).\n\n## beats.json schema\n\n```json\n{\n  \"project\": \"my-film\", \"topic\": \"...\", \"language\": \"en\",\n  \"aspect\": \"9:16\",                       \u002F\u002F 16:9 | 9:16 | 1:1 | 3:4\n  \"style\": \"collage\",\n  \"provider\": \"atlas_cloud\",              \u002F\u002F media backend — default; pluggable (scripts\u002Fprovider.py)\n  \"theme\": \"american-retro\",              \u002F\u002F THEME_PRESET (styles.THEME_PRESETS) — the LOOK layer\n  \"arc\": \"timeline\",                      \u002F\u002F narrative arc (beat-layer.md) — the STORY skeleton\n  \"video_model\": \"google\u002Fgemini-omni-flash\u002Fimage-to-video\",  \u002F\u002F Kling for real people\n  \"image_model\": \"google\u002Fnano-banana-2\u002Ftext-to-image\",       \u002F\u002F keyframes; or openai\u002Fgpt-image-2\u002Ftext-to-image\n  \"image_resolution\": \"1k\",               \u002F\u002F 1k (default) | 2k | 4k\n  \"video_resolution\": \"720p\",             \u002F\u002F 720p (default); Seedance also 480p\u002F1080p (Omni is 720p-only)\n  \"motion_style\": \"punchy\",               \u002F\u002F amplitude: calm | punchy | max (theme sets a default)\n  \"constraints\": \"strict\",                \u002F\u002F strict = defect guards on | loose = let AI explore + re-roll\n  \"voice\": {\"voice_id\": \"leo\", \"language\": \"en\", \"speed\": 1.0},  \u002F\u002F pick per topic\u002Flanguage — see references\u002Fvoices.md\n                                          \u002F\u002F + optional \"clone_ref\": \"path\u002Fto\u002Fsample.mp3\" (clone that voice via seed-audio)\n                                          \u002F\u002F   and \"persona\": \"YouTube tutorial creator\" (delivery style for cloned VO)\n  \"music\": \"epic cinematic orchestral, instrumental, no vocals\",\n  \"mix\": {\"music\": 0.6, \"voice\": 1.25},   \u002F\u002F audio balance — optional; these are the defaults (BGM ducks under the VO)\n  \"caption_style\": \"white\",               \u002F\u002F white (default: clean white subtitle) | paper (cream cut-out collage look)\n  \"captions\": true,                       \u002F\u002F false = no burned-in captions (deliver clean, subtitle in post)\n  \"watermark\": \"Made with Atlas Cloud\",\n  \"mode\": \"croll\",                        \u002F\u002F C-roll only — plus the four fields below\n  \"anchor_photo\": \"path\u002Fto\u002Fphoto.png\",    \u002F\u002F C-roll: the still to anchor (person or product)\n  \"croll_subject\": \"portrait\",            \u002F\u002F C-roll: portrait | product\n  \"subject_wardrobe\": \"a cream knitted sweater and charcoal trousers\",  \u002F\u002F C-roll portrait: outfit lock\n  \"subject_desc\": \"the perfume bottle\",   \u002F\u002F C-roll product: short noun phrase for the sticker\n  \"beats\": [\n    {\n      \"id\": 1, \"title_cn\": \"\", \"title_en\": \"BEFORE MONEY\",\n      \"bg\": \"earthy clay tan\", \"feel\": \"ancient, humble\", \"hook\": \"surprising_stat\",\n      \"narration\": \"For most of history, there was no money...\",\n      \"shots\": [\n        \u002F\u002F shot_size: EST_WIDE|WIDE|MEDIUM|CLOSE|DETAIL ; camera_move: static|push_in|\n        \u002F\u002F pull_out|pan|tilt|parallax (flat-safe only) — VARY per adjacent beat, static for payoff\n        {\"id\": \"a\", \"dur\": 5, \"title\": true,  \"shot_size\": \"WIDE\", \"camera_move\": \"push_in\",\n         \"scene\": \"...wide establishing collage...\",\n         \"element_motion\": \"traders gesture, goat bobs, a paper bird flaps across the frame, coins scatter\"},\n        {\"id\": \"b\", \"dur\": 5, \"title\": false, \"shot_size\": \"CLOSE\", \"camera_move\": \"parallax\",\n         \"scene\": \"...close cut-in detail...\",\n         \"element_motion\": \"the exchanged goods slide together, halftone pulses\"}\n      ]\n    }\n  ]\n}\n```\n`theme`+`arc` set the two big layers; `element_motion` per shot is the energy (make it rich — see\nbelow). `motion`\u002F`collage_style`\u002F`era` are still read for back-compat.\n\n## Model selection (always verify IDs live)\n\nModel IDs change — fetch the live list first: `GET https:\u002F\u002Fapi.atlascloud.ai\u002Fapi\u002Fv1\u002Fmodels`\n(no auth; keep only `display_console: true`). Defaults that work today:\n\n| Job | Model | Note |\n|---|---|---|\n| Keyframe \u002F collage poster | `google\u002Fnano-banana-2\u002Ftext-to-image` | default; renders CN+EN text well; `image_resolution` 1k\u002F2k\u002F4k |\n| Keyframe (alternative) | `openai\u002Fgpt-image-2\u002Ftext-to-image` | set via `image_model`; size+quality auto-mapped from aspect+resolution |\n| Cut out an element | `youchuan\u002Fv8.1\u002Fremove-background` | advanced path only |\n| Animate (non-real content) | `google\u002Fgemini-omni-flash\u002Fimage-to-video` | keeps text stable, layered motion |\n| Animate (**real people \u002F brands**) | `kwaivgi\u002Fkling-video-o3-pro\u002Fimage-to-video` | Omni & Seedance BLOCK celebrities |\n| Narration | `xai\u002Ftts-v1` | clean, multilingual, `voice_id` |\n| Music | `minimax\u002Fmusic-2.6` | `is_instrumental: true` |\n\nSee `references\u002Fmodels-and-gotchas.md` for the full model-choice reasoning and every\nAPI \u002F ffmpeg gotcha (auth header, curl downloads, no-libass captions, content blocks, etc.).\nRead it before debugging any failure — most failures are already documented there.\n\n**Backends are pluggable.** Every API call goes through a **provider** (`scripts\u002Fprovider.py`);\nAtlas Cloud is the default and only backend today. Set `\"provider\"` in beats.json to route to a\ndifferent backend once one is added — the stage scripts don't change. `scripts\u002Fprovider.py`'s\n`run_jobs()` also does the submit\u002Fpoll with **auto-resubmit on a stalled or failed job**.\n\n## Advanced: element-level motion collage\n\nThe standard path animates the *whole* poster (great, automated, \"living poster\"). For the\ndramatic **pieces-fly-in-and-assemble** motion collage (à la cr7v2), or to animate **real\npeople with full control and zero content filters**, cut each poster into independent\nelements and drive them with the local keyframe engine (no video model needed).\n\nRead `references\u002Flocal-engine.md`. In short: `extract_elements.py` (crop + background-removal\n+ residue\u002Ferase cleanup) → `motion.py` (Layer + keyframes, `fly_in`\u002F`slap`\u002F`drop`\u002F`pop_settle`\neasings, procedural confetti\u002Fstarburst, camera zoom+shake+whip, frame render). Pieces fly\nback to their **original positions** on a blurred-placeholder backdrop, so the assembled\nframe reconstructs the original poster.\n\n## Editions\n\n- **Auto edition** (this skill): topic in, film out, all on Atlas.\n- **Manual prompt-pack**: if the user isn't on Atlas, just produce the beat map + the per-beat\n  image prompts + the per-clip motion prompts + the narration script for them to paste into\n  any generator. The creative engine (the prompts) is identical.\n",{"name":5,"description":7},"main","4f0cf017cbae61ed3cafe039d6adeaa93144686f",".opencode\u002Fskills\u002Fvox_director","https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator","https:\u002F\u002Fgithub.com\u002Fjosenaves","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fjosenaves\u002F","Software Engineer ::: Java, Kotlin, SpringBoot, GraphQL, Rust, blockchain, AI, and other cool stuff  :::","2026-09-20 22:57:52",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add josenaves\u002Fcity-video-generator --skill vox-director",{"tool":39,"command":40},"bunx","bunx skills add josenaves\u002Fcity-video-generator --skill vox-director",{"tool":42,"command":43},"pnpm","pnpm dlx skills add josenaves\u002Fcity-video-generator --skill vox-director","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002FSKILL.md",true,[47,53,59,63,67,71,75,79,83,87,91,95,99,103,107,111,115,119,123,127,131,135,139,143,147,153,157,161,165,169,173,177,181,185,189,193,197,201,205,209,213,217,221,225],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"AGENTS.md","REFERENCE","参考文档",1259,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002FAGENTS.md",{"path":54,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":57,"sourceHtmlUrl":58},"assets\u002Fshowcase-football.mp4","ASSET","资产文件",8257315,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fshowcase-football.mp4",{"path":60,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":61,"sourceHtmlUrl":62},"assets\u002Fshowcase-money.mp4",8493950,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fshowcase-money.mp4",{"path":64,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":65,"sourceHtmlUrl":66},"assets\u002Fshowcase-silicon-valley.mp4",6144834,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fshowcase-silicon-valley.mp4",{"path":68,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":69,"sourceHtmlUrl":70},"assets\u002Fshowcase-tang.mp4",7928635,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fshowcase-tang.mp4",{"path":72,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":73,"sourceHtmlUrl":74},"assets\u002Fthumbs\u002Ffootball.jpg",49052,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fthumbs\u002Ffootball.jpg",{"path":76,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":77,"sourceHtmlUrl":78},"assets\u002Fthumbs\u002Fmexican.jpg",42912,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fthumbs\u002Fmexican.jpg",{"path":80,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":81,"sourceHtmlUrl":82},"assets\u002Fthumbs\u002Fmoney.jpg",25761,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fthumbs\u002Fmoney.jpg",{"path":84,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":85,"sourceHtmlUrl":86},"assets\u002Fthumbs\u002Fsilicon-valley.jpg",35292,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fthumbs\u002Fsilicon-valley.jpg",{"path":88,"resourceType":55,"resourceTypeDesc":56,"sizeBytes":89,"sourceHtmlUrl":90},"assets\u002Fthumbs\u002Ftang.jpg",47537,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fassets\u002Fthumbs\u002Ftang.jpg",{"path":92,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":93,"sourceHtmlUrl":94},"examples\u002Fcr7-act.elements_spec.json",1795,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fexamples\u002Fcr7-act.elements_spec.json",{"path":96,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":97,"sourceHtmlUrl":98},"examples\u002Fmoney-15s.beats.json",2637,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fexamples\u002Fmoney-15s.beats.json",{"path":100,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":101,"sourceHtmlUrl":102},"examples\u002Fmoney-60s-9x16-english.beats.json",26503,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fexamples\u002Fmoney-60s-9x16-english.beats.json",{"path":104,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":105,"sourceHtmlUrl":106},"examples\u002Fronaldo-9x16-kling.beats.json",4983,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fexamples\u002Fronaldo-9x16-kling.beats.json",{"path":108,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":109,"sourceHtmlUrl":110},"examples\u002Ftang-30s.beats.json",13462,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fexamples\u002Ftang-30s.beats.json",{"path":112,"resourceType":112,"resourceTypeDesc":113,"sizeBytes":114,"sourceHtmlUrl":16},"LICENSE","许可证",1067,{"path":116,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":117,"sourceHtmlUrl":118},"llms.txt",1442,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fllms.txt",{"path":120,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":121,"sourceHtmlUrl":122},"README.md",8006,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002FREADME.md",{"path":124,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":125,"sourceHtmlUrl":126},"README.zh.md",7704,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002FREADME.zh.md",{"path":128,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":129,"sourceHtmlUrl":130},"references\u002Fbeat-layer.md",10451,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Freferences\u002Fbeat-layer.md",{"path":132,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":133,"sourceHtmlUrl":134},"references\u002Flocal-engine.md",4556,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Freferences\u002Flocal-engine.md",{"path":136,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":137,"sourceHtmlUrl":138},"references\u002Fmodels-and-gotchas.md",5389,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Freferences\u002Fmodels-and-gotchas.md",{"path":140,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":141,"sourceHtmlUrl":142},"references\u002Fprompt-guide.md",20588,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Freferences\u002Fprompt-guide.md",{"path":144,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":145,"sourceHtmlUrl":146},"references\u002Fvoices.md",4181,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Freferences\u002Fvoices.md",{"path":148,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":151,"sourceHtmlUrl":152},"scripts\u002Faroll_assemble.py","SCRIPT","脚本",3413,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Faroll_assemble.py",{"path":154,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":155,"sourceHtmlUrl":156},"scripts\u002Faroll_clips.py",9224,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Faroll_clips.py",{"path":158,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":159,"sourceHtmlUrl":160},"scripts\u002Fasr_beats.py",6885,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fasr_beats.py",{"path":162,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":163,"sourceHtmlUrl":164},"scripts\u002Fassemble.py",7571,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fassemble.py",{"path":166,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":167,"sourceHtmlUrl":168},"scripts\u002Fatlas_cloud.py",8555,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fatlas_cloud.py",{"path":170,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":171,"sourceHtmlUrl":172},"scripts\u002Faudio.py",5311,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Faudio.py",{"path":174,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":175,"sourceHtmlUrl":176},"scripts\u002Fclips.py",10317,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fclips.py",{"path":178,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":179,"sourceHtmlUrl":180},"scripts\u002Fconfetti.py",2008,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fconfetti.py",{"path":182,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":183,"sourceHtmlUrl":184},"scripts\u002Fcroll_keyframes.py",6834,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fcroll_keyframes.py",{"path":186,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":187,"sourceHtmlUrl":188},"scripts\u002Fextract_elements.py",2881,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fextract_elements.py",{"path":190,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":191,"sourceHtmlUrl":192},"scripts\u002Fkenburns.py",3217,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fkenburns.py",{"path":194,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":195,"sourceHtmlUrl":196},"scripts\u002Fkeyframes.py",3857,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fkeyframes.py",{"path":198,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":199,"sourceHtmlUrl":200},"scripts\u002Fmg_scrapbook.py",7299,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fmg_scrapbook.py",{"path":202,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":203,"sourceHtmlUrl":204},"scripts\u002Fmotion.py",9874,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fmotion.py",{"path":206,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":207,"sourceHtmlUrl":208},"scripts\u002Fprovider.py",5132,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fprovider.py",{"path":210,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":211,"sourceHtmlUrl":212},"scripts\u002Fstyle_bakeoff.py",3143,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fstyle_bakeoff.py",{"path":214,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":215,"sourceHtmlUrl":216},"scripts\u002Fstyles.py",14950,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Fstyles.py",{"path":218,"resourceType":149,"resourceTypeDesc":150,"sizeBytes":219,"sourceHtmlUrl":220},"scripts\u002Ftext_overlay.py",6487,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fscripts\u002Ftext_overlay.py",{"path":222,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":223,"sourceHtmlUrl":224},"SKILL.zh.md",18613,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002FSKILL.zh.md",{"path":226,"resourceType":227,"resourceTypeDesc":228,"sizeBytes":229,"sourceHtmlUrl":230},"vox-director.skill","OTHER","其他",92695,"https:\u002F\u002Fgithub.com\u002Fjosenaves\u002Fcity-video-generator\u002Fblob\u002Fmain\u002F.opencode\u002Fskills\u002Fvox_director\u002Fvox-director.skill",[]]