从剧本到成片，一步一步来From script to finished clips, step by step

快速开始Quick start

账号与工作区Account and workspace

注册需要邀请码，由对接人发放；登录后你会有一个个人工作区。要和多人共用一个空间，由创建者在「设置 → 团队成员」里邀请成员。 Signing up requires an invite code from your contact; after signing in you get a personal workspace. To share one space, the creator invites people under Settings → Team members.

创建短剧与分集Create a series and episodes

在工作台空状态点「新建短剧」，填写短剧名与风格指南；进入后创建分集，填写标题与集数。分集在页头的下拉里切换。 From the workspace empty state choose New series and fill in the title and style guide; then create an episode with its title and number. Switch episodes from the dropdown in the header.

导入剧本Import a script

打开一集，在「剧本 / 分镜文本提取」里粘贴文本，或上传 .txt / .md / .docx 文件。「优化剧本」只在你确认后才会改动原文。提取完成后会得到片段列表与全局信息。 Open an episode and paste your script under Script / storyboard extraction, or upload a .txt / .md / .docx file. Optimize script only changes the text after you confirm. Extraction produces the clip list and the global info.

一集的顺序The order of an episode

提取片段 → 素材库补齐素材 → 绑定到片段 → 生成提示词 → 生成视频 → 成片库 Extract clips → complete assets → bind to clips → generate prompts → generate video → finished clips

工作台Workspace

全局信息Global info

放跨片段共享的设定：角色表、场景表、道具表、画幅与风格设定等。素材抽取和风格提炼都会读这里，所以尽量在提取素材之前写清楚。 Shared settings that span clips: cast, locations, props, aspect ratio, and style notes. Asset extraction and style inference both read it, so fill it in before extracting assets.

片段列表Clip list

一个片段对应一次生成，可以包含多个镜头。每条片段都能改描述与时长、插入或删除、按序号移动；总长超过模型单次上限时，另开一个片段。 One clip is one generation and may contain several shots. Each row has an editable description and duration, and can be inserted, removed, or moved by number; when the total exceeds the model's per-call limit, start a new clip.

描述与提示词Description and prompt

片段描述是你写的原文；「生成提示词」把它改写成结构化的视频提示词。批量按钮只补缺失或过期的结果，手改过的不会被覆盖；改了描述或素材绑定后，需要重新生成。勾选「提交前自动生成提示词」时，派发前会自动补齐。 The description is your source text; Generate prompt rewrites it into a structured video prompt. The batch action only fills missing or outdated results and leaves hand-edited ones alone; after changing a description or binding, regenerate. With Auto-generate prompts before dispatch on, missing prompts are filled right before submitting.

台词配音Dialogue and voices

「台词配音」按集整理台词，并为角色准备音色参考或单句录音。录音和音色参考会作为音频参考随生成一起提交；没有单独录音的台词，才回落到角色的音色参考。 The Dialogue tab collects an episode's lines and lets you prepare a voice reference or per-line recordings. Recordings and voice references are submitted as audio references; only lines without their own recording fall back to the character's voice reference.

写作建议Writing tips

导入剧本时旁边的提示里有完整建议：切镜写清起止秒、每个镜头留出足够时长、台词字数贴合镜头秒数、画内与画外音分开标注。 The tips next to the script input cover the full guidance: mark cut points in seconds, give each shot enough time, match line length to shot duration, and mark on-screen vs. voice-over dialogue.

素材库Asset library

素材与分类Assets and categories

素材库同时放角色、场景、道具，以及视频与音频参考素材；前三类可以归类，其余留在「未分类」。素材跨集复用；只想给某几集用的素材，可以在素材设置里限定适用集数。 The library holds characters, locations, and props alongside video and audio reference assets. The first three can be categorized; the rest stay under Uncategorized. Assets are reused across episodes; to keep an asset to specific episodes, limit its scope in the asset settings.

形象与参考图Looks and reference images

每个角色、场景可以有多套「形象」。描述写清视觉特征；可以 AI 生成参考图、对现有图片做指令式修改（改图），或直接换成自己的图。被替换的版本保留在历史里，可以恢复。 Each character or location can have several looks. Write the visual traits in the description; generate a reference image, edit the current one with instructions, or upload your own. Replaced versions stay in history and can be restored.

总体风格Overall style

整部剧共享一套总体风格，可以 AI 生成、从风格库选择或手动填写。生成时由「应用全局风格」开关决定是否随片段一起提交。 The whole series shares one overall style, generated by AI, picked from the style library, or written by hand. The Apply global style switch decides whether it is submitted with each clip.

绑定与一致性Binding and consistency

把素材绑定到片段后，素材的描述与参考图会进入该片段的提示词和生成；工作台会提示缺图或还没绑定的片段。修改素材后，旧提示词与成片会标记为过期，按提示重跑即可。 Once an asset is bound to a clip, its description and reference images feed that clip's prompt and generation; the workspace flags clips that are missing images or bindings. After editing an asset, old prompts and clips are marked outdated — rerun them when prompted.

视频生成与凭证Video generation and credentials

配置位置Where to configure

打开「设置 → 模型与凭证」，按能力分别配置：文本处理、图片理解、图像生成、图像修改、语音合成、视频生成。每个能力可以选不同的服务商。 Open Settings → Models & credentials and configure each capability separately: text, vision, image generation, image editing, speech, and video. Each can use a different provider.

自带账号与费用Bring your own account

视频生成走你自己的算力账号，费用直接结算给服务商。密钥只保存在当前浏览器，随请求发送，不存进工作区。 Video generation runs on your own compute account and is billed by the provider directly. Keys stay in the current browser, are sent with each request, and are not stored in the workspace.

RunningHubRunningHub

支持官方工作流与自有工作流。国内站与国际站是独立的账号体系，API Key、钱包和工作流 ID 只在创建它的站点有效，配置时站点要和 Key 匹配。 Official and own workflows are supported. The CN and international sites are separate account systems: API keys, wallets, and workflow IDs only work on the site that created them, so match the site to the key.

ComfyUI 与其它平台ComfyUI and other platforms

本地或远程 ComfyUI（如仙宫云、AutoDL）按面板填写地址、Token 与工作流；也支持 Seedance、万相等服务商。文本模型可以通过任意 OpenAI 兼容端点接入。 For local or remote ComfyUI (Xiangongyun, AutoDL, and similar), fill in the address, token, and workflow. Seedance and Wan are also supported. Text models can connect through any OpenAI-compatible endpoint.

还没配置时Before it is configured

生成按钮会禁用，并提示先到对应能力页配置服务商。配置保存后回到工作台即可使用。 The generate button stays disabled and points you to the matching capability page. Once saved, return to the workspace and it is ready.

任务与成片Jobs and finished clips

派发与状态Dispatch and status

勾选片段后可以逐条或整集生成。队列显示排队、运行与失败状态；失败的片段可以单独重跑，不影响其它结果。 Select clips to generate one at a time or as a whole episode. The queue shows queued, running, and failed jobs; a failed clip can be re-run on its own without touching the rest.

重新生成Re-generating

改了描述、提示词、绑定素材或素材本身后，已有成片仍是按旧输入生成的，会被标记为过期；重新生成才会用上当前输入。 After changing a description, prompt, binding, or the asset itself, existing clips were made from the old input and are marked outdated; only a new run uses the current input.

成片库Finished clips

本集所有成片集中在「成片库」：逐条设定当前版本、单个或批量下载。仍托管在外部引擎上的文件可能过期，重要结果请及时下载保存。 Every finished clip for the episode lives under Finished clips: choose the current version, download one or all. Files still hosted by an external engine may expire, so save important results early.

团队协作Team collaboration

团队成员共用同一批短剧、素材库与生成记录，每人用自己的账号登录；编辑和生成默认全员可用，邀请与移除成员仅限创建者。 Team members share the same series, asset library, and generation history, each signing in with their own account. Editing and generating are open to the team by default; only the creator can invite or remove members.

整部剧可以设为工作区可见，或私有并只共享给指定成员，每位成员单独设可编辑或只读。示例项目对所有访客只读。 A whole series can be visible to the workspace, or private and shared with specific members, each set to edit or read-only. The sample project is read-only for every visitor.

示例项目Sample project

示例项目不需要登录，打开后可以浏览一整集的剧本、素材与成片，写操作都是关闭的。要开始自己的创作，登录或注册即可。 The sample project needs no sign-in: browse a full episode's script, assets, and finished clips with every write disabled. Sign in or sign up to start your own.

入口：落地页的「查看示例项目」，或登录页的「先看看示例项目（无需登录）」。 Entry points: View the live sample on the landing page, or Browse the sample project on the sign-in screen.

打开示例项目Open the sample project

常见问题FAQ

注册需要邀请码吗？Do I need an invite code to sign up?

是。邀请码由对接人发放；已有账号直接登录。注册后会自动获得个人工作区。 Yes. Your contact provides the code; existing accounts just sign in. Signing up creates a personal workspace automatically.

生成视频还要另外付费吗？Is video generation billed separately?

要，费用直接付给算力供应商。工作区的试用或授权，与生成费用是两笔。 Yes, and it is paid to the compute provider directly. The workspace trial or license and the generation cost are separate.

支持哪些生成平台？Which generation platforms are supported?

RunningHub（官方或自有工作流）、本地或远程 ComfyUI、Seedance、万相等；文本模型支持 OpenAI 兼容端点。其它平台请联系对接人。 RunningHub (official or own workflows), local or remote ComfyUI, Seedance, Wan, and more; text models work through OpenAI-compatible endpoints. Contact your rep for other platforms.

数据放在哪里？Where is my data stored?

托管版按工作区隔离；也支持私有化部署，把整套数据放在你自己的服务器或内网。 The hosted edition isolates data per workspace; private deployment is also supported, keeping everything on your own servers or intranet.

换电脑后为什么看不到密钥？Why are my keys missing on another computer?

密钥只保存在当前浏览器；换设备或清缓存后需要重新填写，工作区里的数据不受影响。 Keys live only in the current browser; after switching devices or clearing site data, enter them again. Workspace data is unaffected.

生成失败怎么办？What if a generation fails?

先看队列里的报错，失败片段可以单独重跑；再检查服务商余额、API Key、工作流 ID 与站点是否匹配。仍然失败时，把剧集、片段号与报错发给对接人。 Check the error in the queue and re-run the failed clip on its own; then verify the provider balance, API key, workflow ID, and site. If it still fails, send your contact the episode, clip number, and error.

没找到答案？联系为你开通账号的对接人，附上剧集、片段号和截图会更快定位。 Not covered here? Contact the person who set up your account with the episode, clip number, and a screenshot — it makes diagnosis much faster.