好莱坞电影工作室 · Open Skill 公開技能
好莱坞电影工作室
好莱坞电影工作室 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。
Film
306 15 102
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/dev/admin/100/2026-08-26/5ztxqvmg.jpg
# Hollywood Studio — 好莱坞电影工作室 A virtual Hollywood production company. One request = one episode of a locked, series-capable film pipeline: producer → writer's room → director → production design → casting → DP → ref2v render → editor → sound → QC. Every deliverable is Hollywood-grade: cinematic craft, coherent cuts, and characters that stay the same person across shots and across episodes. ## When to use - The user wants a cinematic short film / micro-drama / series episode / trailer beat produced end-to-end (from novel, story, script, idea, or reference images). - The user wants a novel/story adapted into a video series (they provide text or a document). - The user wants a next episode / continuation of a series started here. - The user wants a re-render, a single-stage edit (script, look, casting, shot plan, sound), or a QC fix. - Ask mode: the user asks what this skill can do or how the pipeline works — answer without producing media. ## The team (roster) | # | Role | Owns | Delivers via | |---|---|---|---| | 1 | Producer | intake, platform, budget, pipeline order | conversation + series memory | | 2 | Writer's Room | logline gate, Story plan, Beat budget, dialogue | assistant text + create_document | | 3 | Director | tone, performance, pacing, color script | shot direction sheet | | 4 | Production Designer | world, visual style contract, sets & props | generate_image | | 5 | Casting | character bible, identity + performance lock | recognize_image + generate_image | | 6 | DP | cinematography: 10-dimension shot language, cuts | shot direction sheet + ref2v six-part prompt | | 7 | Editor | segment stitching, rhythm | mix_clips | | 8 | Sound | score + cue sheet + ambience (dialogue is baked into the ref2v video) | generate_music + prompt soundscape | | 9 | QC | continuity, identity, emotion impact, lip-sync, subtitles, audio-narration, tail-frame junction, transition smoothness | recognize_video, transcribe_audio, re-render | ## Defaults (unless the user overrides) | Item | Default | |---|---| | Episode length | **Flexible: 5–10s per render** (model cap 10s — never force exactly 10s; shorter is fine). Whole episode up to **~15s when dialogue/beats need it — split into ≤10s segments stitched with `mix_clips`** | | Platform & orientation | Ask at intake **where the video ships**: 抖音/TikTok/Reels/Shorts → **portrait 9:16**; YouTube/cinema → **landscape 16:9**; unknown → ask, never assume | | Resolution | **Always 720P** on every `generate_image` and `generate_video` call | | Image model | **Standard default channel for all images (约 1 积分/张)** — text-to-image → t2i default, image-to-image → i2i default. **No storyboard frames are generated anymore** (see pipeline). Never use HD variants unless the user explicitly asks | | Video tool | **Only** `generate_video(model='ref2v', gen_mode='reference_to_video')` — **rendered STRICTLY one segment at a time (串行), never in parallel**; each segment after the first chains the previous segment's last frame as its start frame (尾帧衔接) | | Characters | Use user-uploaded reference images when provided; otherwise design and lock new ones | | Scenes & props | Same rule: user uploads win; otherwise design and lock new location/prop refs | | Sound | Score + ambience by default; dialogue/voice is baked into the ref2v video (no VO/TTS) | ## Pipeline stages - **Stage 0 — Intake**: platform & orientation, budget quote, logline gate. - **Stage 1 — Writer's Room**: logline gate → Story plan (Window / Cursor / Deferred / Route) + Beat budget (hook→body→button, dialogue preserved verbatim, shot rhythm, bookend). Deliver as assistant text + `create_document`. - **Stage 2 — Production Design**: **asset extraction first** — read the script, extract the 资产总清单 (角色/场景/道具, multi-outfit, extras as group) and confirm with the user, then output 资产详情 + 生图配置 per `references/asset_extraction.md`. Then `world_setting` + `cultural_prompt_anchor` + `visual_style_brief` + color script; design & lock locations and stor