Clone any viral video with AI agents. Not just a script, the whole workflow: swap the face, the words, the B-roll, ship 100 variants in one command, and get your 100M views.
TypeScript agentic-aiaiai-agentsai-videocompilerdslffmpeggenerative-aillmmarkup-languagemonorepoplugin-systemprogramming-languagetext-to-videotypescriptvideovideo-automationvideo-clonevideo-editingvideo-generation
docs: start in the checkout, and stop documenting the Playground as a Develop topic
rpone5a320045ded0
diff --git a/README.md b/README.mdindex c43eb4d7..253c3c7d 100644--- a/README.md+++ b/README.md@@ -105,12 +105,17 @@ Components outside the Script (video generators, speech models, caption renderer### With a coding agent-Send your agent this prompt:+The skill lives in the checkout, so clone the repository and start your coding agent inside it:++```bash+git clone https://github.com/hypit-ai/narratage.git+cd narratage+```++Then send:```text-Install and use the narratage skill from this repository. Set up my environment, ask for only the-API keys required by my Runtime Profile, and guide me through authoring and building my first SVML-video.+/narratage Set up my environment, ask for only the API keys required by my Runtime Profile, and guide me through authoring and building my first SVML video.```### From the terminaldiff --git a/README.zh-CN.md b/README.zh-CN.mdindex 5e84ba2f..1cd966bb 100644--- a/README.zh-CN.md+++ b/README.zh-CN.md@@ -83,11 +83,17 @@ Script 外面的组件(视频生成器、语音模型、字幕渲染器、轨### 用编程 Agent-把下面这段发给你的 Agent:+skill 就在这个 checkout 里,所以先克隆仓库,并在其中启动你的编程 Agent:++```bash+git clone https://github.com/hypit-ai/narratage.git+cd narratage+```++然后发送:```text-安装并使用这个仓库里的 narratage skill。配置我的环境,只向我索取当前 Runtime Profile-实际需要的 API key,然后带我完成第一支 SVML 视频的创作与 Build。+/narratage 配置我的环境,只向我索取当前 Runtime Profile 实际需要的 API key,然后带我完成第一支 SVML 视频的创作与 Build。```### 从终端diff --git a/docs/.vitepress/config.ts b/docs/.vitepress/config.tsindex 30c16705..2f54b155 100644--- a/docs/.vitepress/config.ts+++ b/docs/.vitepress/config.ts@@ -156,7 +156,6 @@ const enTheme = {{ text: "Adding an Author Package", link: "/guide/author-packages" },{ text: "Adding a Provider", link: "/guide/providers" },{ text: "Runtime Profile", link: "/guide/runtime-profile" },- { text: "SVML Playground", link: "/guide/svml-playground" },{ text: "Caption Playground", link: "/guide/caption-playground" },{ text: "Local Services", link: "/guide/services" },{ text: "Testing", link: "/guide/testing" },@@ -201,7 +200,6 @@ const zhTheme = {{ text: "添加 Author 包", link: "/zh/guide/author-packages" },{ text: "添加 Provider", link: "/zh/guide/providers" },{ text: "Runtime Profile", link: "/zh/guide/runtime-profile" },- { text: "SVML Playground", link: "/zh/guide/svml-playground" },{ text: "字幕 Playground", link: "/zh/guide/caption-playground" },{ text: "本地服务", link: "/zh/guide/services" },{ text: "测试", link: "/zh/guide/testing" },diff --git a/docs/guide/svml-playground.md b/docs/guide/svml-playground.mddeleted file mode 100644index 65bd0e90..00000000--- a/docs/guide/svml-playground.md+++ /dev/null@@ -1,165 +0,0 @@-# SVML Playground--A read-only preview of one Source: the code on the right, the picture top-left,-and a timeline bottom-left rendered from the SVML itself.--```bash-pnpm svml:playground -- --source examples/talking-film-broll-preview/main.svml-# ➜ http://localhost:5179/-```--| Argument | Meaning |-| --- | --- |-| `--source <main.svml>` | The Author Source to read. Required. |-| `--run <build.svrun>` | A Run Source, read for material and timings it already names. |-| `--runtime <svml.runtime.json>` | Where material earlier Builds produced is kept. Only a Source that reuses an accepted shot needs one. |-| `--port <number>` | Defaults to `5179`. |--The Playground never writes. There is no route that can change what you wrote,-and no form that edits a Recipe — that is the-[Caption Playground](./caption-playground)'s job.--## What it is for--Reading a Source as a timeline **before** anything has been generated. It-answers "where does this B-roll land, how long is it, and is the card in the-right part of the frame" without a Seedance generation, a WhisperX alignment, an-FFmpeg render or a headless browser.--It runs on a directory holding nothing but `main.svml` and its `.svs` style-sheet. No package lock, no Runtime Profile, no build.--## Three ways into one selection--The timeline, the source and the picture are three views of the same thing, so-selecting in any of them selects in all three.--- **Click a timeline clip** and the playhead moves to its first frame, the- picture draws a box around it, and the source scrolls to the- `<media-track:Item>` that placed it.-- **Click a marked line in the source** and the same happens. Lines that bind a- clip carry a coloured bar in the gutter, so what is clickable is visible- standing still. A Script marker pair and the clip it binds share a colour.-- **Click the picture** and whatever is drawn under the pointer is selected.--Clicking marked prose lands **inside the marker you clicked**, not at the start-of whatever encloses it — a marker that places nothing still means something.--Scrub by dragging the ruler or an empty part of a track. `Space` plays and-pauses, `←` and `→` step one frame with `Shift` ten, `Home` and `End` jump to the-ends, and `Esc` clears the selection.--## Levels--A Segment encloses a Selection, which can enclose another Selection. Each level-is drawn in its own colour, in the source, on the timeline and on the picture,-and **an enclosing range stays outlined while an inner one is** — the nesting is-what the markers are for. Two ranges at the same level are the same kind of-thing and read the same.--Everything the playhead is inside is outlined as it passes, whether or not a-Track was hung on it.--## Sound--HyperFrames renders a silent picture on purpose: programme audio is a separate-Track the media pipeline muxes in at the end. A preview is not a render, and-placing B-roll against speech means hearing the speech, so the Speech Spine's own-material is allowed to sound while the transport is running. Cutaways stay-silent, as they are in a build unless they ask otherwise. The speaker button-turns it off.--## What is real and what is estimated--Everything structural is real. Placement Frames, padding, stacking order, motion-and the Track layout are computed by the same functions a build calls, from your-Source and your style sheet.--Timings are another matter, and the header says which you are looking at:--| Badge | Meaning |-| --- | --- |-| `timing: measured` | Read from a completed build's aligned transcript. |-| `timing: estimated` | Derived from the Script text at normal delivery pace. |-| `picture: measured` | Every element shows real material. |-| `picture: estimated` | Some shots have not been made, and stand in. |-| `picture: estimated` | Nothing has been shot or generated yet. |--The two are separate claims. A Source can have all its footage and still have an-estimated timeline, because where a cut lands is a question about speech, not-about files.--The Playground prefers real things, in this order:--1. **A Run Source.** Which material and which timings a Source is read with is- an authoring decision, so `<satisfy>` is where it is made:-- | Candidate | Supplies |- | --- | --- |- | `<file from="./shot.mp4" media-type="video/mp4"/>` | Material you already have. |- | `<value type="…" from="./timing.json"/>` | A value on disk, such as an alignment produced elsewhere. |- | `<build-record build="…" output="take.video"/>` | Something an earlier Build made. Needs `--runtime`. |-- Timings are called measured only when both `timing.map` and the Program Space- were supplied. One without the other is still an estimate, and says so.--2. **An earlier Build.** A shot that has been generated and accepted is named by- the Build it came from, not by a path — a produced Artifact has an identity- rather than a location. Given `--runtime`, the Playground reads that Build's- Record and the Artifact bytes behind it, so a take generated yesterday is the- take on screen today. Without it, the Run Source is still read and only the- Build Record is refused, with the Build named.--3. **The Script text.** Word durations come from `@narratage/estimate`, the same- syllable model the pipeline uses before generation.--An estimated timeline is a proportion, not a prediction: real cut points move-once WhisperX has aligned real audio.--## What stands in for what has not been made--A Track cannot be built without the material it places, and refusing to draw-anything until every shot exists would make the preview useless for the part of-the work it is meant for. So three things stand in, each announced rather than-presented as fact:--| Missing | Shown instead |-| --- | --- |-| A generated shot that names a picture | That picture, held for the length the shot declares |-| A generated shot that names nothing | A black frame of the programme's shape |-| Caption phrasing nobody has planned | The words cut every few Atoms, in the Program's own runs |--The first is why a preview is worth looking at before anything is generated: a-reference frame is not the take, but it is the right subject in the right shape.-Both stand-ins are marked on the clip itself — hatched for a picture, dimmed for-a black frame — and named in full when the clip is selected. Drawing them needs-`ffmpeg`; without it the shots stay unmade and the Tracks say so.--## What it draws--Every Track the Source produces, whichever package made it. The Playground asks-the compiled Source for its exports and builds the ones typed `VisualTrack` or-`AudioTrack` — so a package that grows a new kind of Track appears here without-the Playground being taught about it.--The application explicitly bundles Narratage's official video packages. That-list belongs only to the Playground: installing or removing this application-changes nothing about compilation or the Runtime. Adding another official-package requires a Playground release; adding another Track to a package it-already bundles does not require a special preview branch.--Each Track is built on its own. A Track waiting on a Provider this machine has-no key for costs only itself; the rest of the programme still plays, and the-Track says what it was waiting for rather than disappearing.--This is the compiler, not a reading of it. The Source is compiled by-`compileSourceClosure` and executed by the same Producers a build runs, so a-Source that will not build does not silently preview. Two thin decorators keep-the source positions the compiler discards — nothing else is re-implemented.--## A note on `<space:Frame>`--`right` and `bottom` are **absolute edge positions**, not insets. A Frame-spanning the middle 80% of its parent is `left="8%" right="92%"`, not-`left="8%" right="8%"` — the latter resolves to zero width and is rejected. The-Playground reports that with the offending element highlighted.diff --git a/docs/quickstart.md b/docs/quickstart.mdindex ee02c358..f13ad867 100644--- a/docs/quickstart.md+++ b/docs/quickstart.md@@ -20,15 +20,21 @@ page gets you to that first safe plan: it needs no API keys and makes no paid ca## Use the Narratage skill-If you use a coding agent, send it this:+The skill lives in the checkout, so clone the repository and start your coding agent inside it:++```bash+git clone https://github.com/hypit-ai/narratage.git+cd narratage+```++Then send:```text-Install and use the narratage skill from this repository. Set up my environment, ask for only the-API keys required by my Runtime Profile, and guide me through authoring and building my first SVML-video.+/narratage Set up my environment, ask for only the API keys required by my Runtime Profile, and guide me through authoring and building my first SVML video.```-Otherwise, work through the five steps below.+It works through the same five steps below, asking you only for what your Runtime Profile actually+needs. Work through them yourself if you would rather not use an agent.## 1. Install the source workspacediff --git a/docs/quickstart/composition.md b/docs/quickstart/composition.mdindex 2db10154..bcfe6b2e 100644--- a/docs/quickstart/composition.md+++ b/docs/quickstart/composition.md@@ -198,6 +198,10 @@ The complete data flow from Script to rendered video. This example is based on</svml>```+`right` and `bottom` are absolute edge positions, not insets. A Frame spanning the middle 80% of+its parent is `left="10%" right="90%"`, not `left="10%" right="10%"` — the second resolves to zero+width and is rejected.+### Stylesheet (`studio.svs`)```svsdiff --git a/docs/quickstart/preview.md b/docs/quickstart/preview.mdindex 767cf7ea..02dfbe43 100644--- a/docs/quickstart/preview.md+++ b/docs/quickstart/preview.md@@ -82,6 +82,10 @@ out to be cut at the wrong moment.A declared duration is honoured wherever one exists. Where none does, words are placed at anordinary delivery pace and the remaining time is divided evenly.+This applies to every Track the Source produces, whichever package made it. The Playground asks the+compiled Source for its exports and builds the ones typed `VisualTrack` or `AudioTrack`, so a+package that grows a new kind of Track appears here without the Playground being taught about it.+Drawing stand-ins needs `ffmpeg` on your `PATH`. Without it the shots stay unmade and the Tracks sayso, rather than the preview pretending they were drawn.@@ -132,5 +136,17 @@ them selects in all three. Click a clip and the playhead moves to its first framoutlines it, and the source scrolls to the tag that placed it. Click a marked line in the source, orwhatever is drawn under the pointer, and the same happens.+A Segment encloses a Selection, which can enclose another Selection. Each level is drawn in its own+colour — in the source, on the timeline and on the picture — and an enclosing range stays outlined+while an inner one is, because the nesting is what the markers are for. Everything the playhead is+inside is outlined as it passes, whether or not a Track was hung on it.+Drag the ruler to scrub. `Space` plays and pauses, `←` and `→` step one frame with `Shift` for ten,`Home` and `End` jump to the ends, and `Esc` clears the selection.++## Sound++HyperFrames renders a silent picture on purpose: programme audio is a separate Track the media+pipeline muxes in at the end. But placing B-roll against speech means hearing the speech, so the+Speech Spine's own material is allowed to sound while the transport is running. Cutaways stay+silent, as they are in a build unless they ask otherwise. The speaker button turns it off.diff --git a/docs/zh/guide/svml-playground.md b/docs/zh/guide/svml-playground.mddeleted file mode 100644index 10a9853b..00000000--- a/docs/zh/guide/svml-playground.md+++ /dev/null@@ -1,138 +0,0 @@-# SVML Playground--单个 Source 的只读预览:右边是代码,左上是画面,左下是直接从 SVML 渲染出的时间轴。--```bash-pnpm svml:playground -- --source examples/talking-film-broll-preview/main.svml-# ➜ http://localhost:5179/-```--| 参数 | 含义 |-| --- | --- |-| `--source <main.svml>` | 要读取的 Author Source,必填。 |-| `--run <build.svrun>` | Run Source,用来读取其中已经指名的素材与时间戳。 |-| `--runtime <svml.runtime.json>` | 早先构建产出的素材存放在哪里。只有复用已验收镜头的 Source 才需要。 |-| `--port <number>` | 默认 `5179`。 |--Playground 从不写入。它没有任何可以改动你所写内容的路由,也没有编辑 Recipe 的表单-—— 那是[字幕 Playground](./caption-playground) 的职责。--## 它解决什么问题--**在任何东西被生成之前**,把 Source 当成时间轴来读。不需要 Seedance 生成、不需要-WhisperX 对齐、不需要 FFmpeg 渲染、不需要无头浏览器,就能回答「这条 B-roll 落在哪、-多长、卡片在画面里的位置对不对」。--一个只装着 `main.svml` 和它的 `.svs` 样式表的目录就能跑。不需要 package lock、-不需要 Runtime Profile、不需要构建。--## 通往同一个选中的三条路--时间轴、源码和画面是同一件事的三种视图,在任何一处选中,三处同时选中。--- **点时间轴的片段**:播放头移到它的第一帧,画面上把它框出来,源码滚动到放置它的- 那个 `<media-track:Item>`。-- **点源码里被标记的行**:结果相同。绑定了片段的行在行号旁有一条彩色竖条,所以哪里- 可点是静止可见的,不用靠鼠标扫过去发现。一对 Script 标记和它绑定的片段共用一个颜色。-- **点画面**:指针下面画的是什么就选中什么。--点击被标记的散文会落在**你点的那个标记里面**,而不是被甩到包着它的那一层的开头 ——-一个什么都不放置的标记依然有含义。--拖动标尺或轨道空白处可以 scrub。`空格` 播放/暂停,`←` `→` 逐帧(配合 `Shift` 为十-帧),`Home` / `End` 跳到首尾,`Esc` 取消选中。--## 层级--Segment 包着 Selection,Selection 还能再包 Selection。**每一层有自己的颜色**,在源码、-时间轴和画面上一致;而且**内层被框住时外层保持框住** —— 嵌套关系正是这些标记存在的-理由。同一层的两个范围是同一种东西,读起来也相同。--播放头经过的每一个范围都会被框出来,不管有没有轨道挂在它上面。--## 声音--HyperFrames **刻意只渲染无声画面**:programme audio 是独立的 Track,由 media pipeline-在最后 mux 进去。但预览不是渲染 —— 检查 B-roll 落点的前提就是能听见那句话,所以播放时-Speech Spine 自己的素材允许发声。Cutaway 保持静音,这和真实构建里的行为一致(除非显式-声明要带音频)。喇叭按钮可以关掉。--## 哪些是真实的,哪些是估算的--结构性的一切都是真实的。Placement Frame、padding、层叠顺序、动效和轨道布局,都由构建-所调用的同一批函数、从你的 Source 和样式表算出。--时间则是另一回事,顶栏的徽章会说明你正在看的是哪一种:--| 徽章 | 含义 |-| --- | --- |-| `timing: measured` | 读自已完成构建的对齐转写。 |-| `timing: estimated` | 按常速语速从 Script 文本推导。 |-| `picture: measured` | 每个元素都显示真实素材。 |-| `picture: estimated` | 有些镜头还没做出来,用替身顶着。 |-| `picture: estimated` | 尚无任何拍摄或生成产物。 |--两者是**独立的判断**。一份 Source 可以素材齐备而时间轴仍是估算的 —— 剪切点落在哪里-是关于语音的问题,不是关于文件的问题。--Playground 按以下顺序优先采用真实来源:--1. **Run Source。** 用哪些素材、哪份时间戳是**作者的决定**,所以由 `<satisfy>` 指定:-- | 候选 | 提供什么 |- | --- | --- |- | `<file from="./shot.mp4" media-type="video/mp4"/>` | 你手上已有的素材。 |- | `<value type="…" from="./timing.json"/>` | 磁盘上的一个值,例如别处产出的对齐结果。 |- | `<build-record build="…" output="take.video"/>` | 上一次构建做出来的东西。需要 `--runtime`。 |-- 只有 `timing.map` 和 Program Space **两者都被提供**时才算实测;只给其中一个仍然- 是估算,并且会照实说。--2. **上一次构建。** 已经生成并验收的镜头,是按**它出自哪次构建**来指名的,不是按路径- —— 产出物有身份,没有位置。给了 `--runtime`,Playground 就去读那次构建的记录和它- 背后的 Artifact 字节,于是昨天生成的镜头就是今天屏幕上的镜头。不给的话,Run Source- 照读,只有那条 build record 被拒绝,并说明是哪一次构建。--3. **Script 文本。** 词的时长来自 `@narratage/estimate`,也就是流水线在生成之前使用的- 同一套音节模型。**凡是 Source 已经声明了时长的地方,以声明为准** —— 一段 take 说自己- 是 8 秒,它的词就铺满这 8 秒,音节模型只提供词与词之间的比例。--估算出的时间轴是**比例,不是预测**:真实的剪切点会在 WhisperX 对齐真实音频之后移动。--## 没做出来的东西用什么顶上--一条 Track 没有素材就建不出来,而"所有镜头都拍完之前什么都不画"会让预览在它最该派上-用场的阶段毫无用处。所以有三种顶替,每一种都写明,不会被说成是成品:--| 缺什么 | 顶上的东西 |-| --- | --- |-| 一个引用了图片的生成镜头 | 那张图,按镜头声明的时长撑开 |-| 一个什么都没引用的生成镜头 | 与节目同尺寸的黑场 |-| 没人规划过的字幕分句 | 按 Program 自己的 run,每几个 Atom 切一屏 |--第一种正是"还没生成就值得预览"的原因:参考图不是成片,但主体对、画幅对。两种顶替都-标在片段本身上,选中时会写明是哪一种。画它们需要 `ffmpeg`;没有的话镜头就保持未做,-轨道会照实说。--## 它画哪些东西--这份 Source 产出的每一条 Track,无论出自哪个包。Playground 向编译结果要它的导出,-把类型是 `VisualTrack` 或 `AudioTrack` 的那些建出来 —— 所以某个包新长出一种 Track,-不需要教 Playground 就会出现在这里。--这个应用显式携带 Narratage 的官方视频包清单。清单只属于 Playground:安装或卸载它,-都不会改变编译与 Runtime。新增一个官方包时需要同步发布 Playground;已携带的包新增-Track 时,不需要再为预览写一条特殊分支。--每条 Track 各自构建。某条轨道在等一个这台机器没有密钥的 Provider,代价只有它自己:-其余部分照常播放,而那条轨道会说明它在等什么,而不是凭空消失。--这是编译器本身,不是对它的一种解读。Source 由 `compileSourceClosure` 编译,由构建时-同一批 Producer 执行,所以**构建不出来的 Source 不会悄悄预览成功**。两个很薄的装饰器-留住了编译器丢弃的源码位置,除此之外没有任何东西被重新实现。--## 关于 `<space:Frame>` 的一点说明--`right` 和 `bottom` 是**绝对边位置**,不是内缩量。占据父级中间 80% 的 Frame 写作-`left="8%" right="92%"`,而不是 `left="8%" right="8%"` —— 后者解出的宽度为零,会被-拒绝。Playground 会报出这一点,并高亮出错的元素。diff --git a/docs/zh/quickstart.md b/docs/zh/quickstart.mdindex 3616c42e..93dc23f4 100644--- a/docs/zh/quickstart.md+++ b/docs/zh/quickstart.md@@ -19,14 +19,20 @@ Narratage 把这份源码编译成一张可见的执行图。在任何模型或## 使用 Narratage skill-如果使用编程 Agent,把下面这段发给它:+skill 就在这个 checkout 里,所以先克隆仓库,并在其中启动你的编程 Agent:++```bash+git clone https://github.com/hypit-ai/narratage.git+cd narratage+```++然后发送:```text-安装并使用这个仓库里的 narratage skill。配置我的环境,只向我索取当前 Runtime Profile-实际需要的 API key,然后带我完成第一支 SVML 视频的创作与 Build。+/narratage 配置我的环境,只向我索取当前 Runtime Profile 实际需要的 API key,然后带我完成第一支 SVML 视频的创作与 Build。```-否则,按下面五个步骤依次进行。+它会走下面同样的五个步骤,并且只向你索取 Runtime Profile 实际需要的东西。不想用 Agent 的话,也可以自己依次完成这五步。## 1. 安装源码工作区diff --git a/docs/zh/quickstart/composition.md b/docs/zh/quickstart/composition.mdindex fc13f877..e287b87d 100644--- a/docs/zh/quickstart/composition.md+++ b/docs/zh/quickstart/composition.md@@ -194,6 +194,8 @@ Build Target,也可以直接接到媒体裁切、音频/帧提取或模型参</svml>```+`right` 与 `bottom` 是绝对的边位置,不是内缩量。一个占父级中间 80% 的 Frame 写作 `left="10%" right="90%"`,而不是 `left="10%" right="10%"`——后者解出的宽度为零,会被拒绝。+### 样式表 (`studio.svs`)```svsdiff --git a/docs/zh/quickstart/preview.md b/docs/zh/quickstart/preview.mdindex 1716cc57..dec49041 100644--- a/docs/zh/quickstart/preview.md+++ b/docs/zh/quickstart/preview.md@@ -65,6 +65,8 @@ pkill -f svml-playground || true凡是存在声明时长的地方都会被遵守。没有声明的地方,词按正常语速安放,剩余时间平均分配。+这对 Source 产出的每一条 Track 都成立,无论它出自哪个包。Playground 向编译后的 Source 索要它的 export,并构建其中类型为 `VisualTrack` 或 `AudioTrack` 的那些——因此某个包新长出一种 Track 时,它会直接出现在这里,不需要 Playground 事先认识它。+绘制替身需要 `PATH` 上有 `ffmpeg`。没有它,这些镜头就保持未绘制状态、Track 会如实说明,而不是假装它们已经画好了。## 读懂徽章@@ -102,4 +104,10 @@ pkill -f svml-playground || true时间线、代码与画面是同一件事的三个视角,因此在任何一个里选中,另外两个都会跟着选中。点击一个片段,播放头会移到它的首帧,画面上会把它框出来,代码会滚动到安放它的那个标签。点击代码里被标记的行,或者画面上指针所指之处,效果相同。+Segment 包着 Selection,Selection 还能再包 Selection。每一层有自己的颜色——在源码、时间线和画面上保持一致——而且内层被框住时外层依然保持框住,因为嵌套关系正是这些标记存在的理由。播放头经过的每一个范围都会被框出来,不管有没有 Track 挂在它上面。+拖动标尺即可走带。`Space` 播放与暂停,`←` 和 `→` 步进一帧、按住 `Shift` 为十帧,`Home` 和 `End` 跳到首尾,`Esc` 清除选中。++## 声音++HyperFrames 刻意只渲染无声画面:programme audio 是一条独立的 Track,由 media pipeline 在最后 mux 进去。但把 B-roll 对着语音安放,前提就是能听见那句话,所以走带播放时 Speech Spine 自己的素材允许发声。Cutaway 保持静音——除非显式声明要带音频,这和真实构建里的行为一致。喇叭按钮可以关掉。