scenario-veo

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Scenario Veo Video

Scenario Veo 视频

Overview

概述

Veo, Google's video family on Scenario, ships four members at authoring time: Veo 3.1 (full quality, reference images with a style switch), Veo 3.1 Fast (same modes, quicker, no switch), Veo 3.1 Lite (frame anchors only, cheapest), and Veo 3.1 Extend Video (continues a clip). Discover them with

search

and treat

model_schema_get

as the contract: members agree on parameter names and disagree on which inputs exist at all.

Connection and the core loop: see the

scenario

skill in this repo; model-agnostic video work: the

scenario-video

skill. If a sibling skill named here is missing from your available skills, ask the user to install it (

npx skills add scenario-labs/skills --skill <name>

); unattended, proceed from tool schemas and flag the gap.

Veo是Google在Scenario平台上的视频生成模型系列，在撰写本文时共有4个版本：Veo 3.1（全质量版，支持参考图及风格切换）、Veo 3.1 Fast（模式相同，速度更快，无风格切换）、Veo 3.1 Lite（仅支持帧锚定，价格最低）以及Veo 3.1 Extend Video（用于续接视频片段）。可通过

search

指令查找这些模型，并将

model_schema_get

视为约定规范：各版本的参数命名一致，但支持的输入项不尽相同。

连接方式与核心流程：参见本仓库中的

scenario

skill；与模型无关的视频处理工作：参见

scenario-video

skill。如果你的可用skill列表中缺少此处提到的关联skill，请让用户安装（命令：

npx skills add scenario-labs/skills --skill <name>

）；若处于无人值守模式，可基于工具schema继续操作并标注缺失项。

Quick reference

快速参考

Mode follows from the inputs (names from the live schema):

Mode	Inputs	Behavior
Text	`prompt`	`aspectRatio` 16:9 or 9:16
First frame	`image` (+ `prompt` )	opens on that frame; a mismatched image may be cropped
Transition	`image` + `lastFrameImage`	bridges the stills; prompt one continuous move
Reference	`referenceImages` (1 to 3)	subject or style consistency; `duration` locked to 8
Extend	`video` (Extend member)	continues the clip; 16:9 source, short side 720p or 1080p

image

and

referenceImages

are mutually exclusive.

referenceImagesType

exists on full Veo 3.1 alone and reads the whole array one way:

ASSET

(default) carries subjects, objects, and scenes;

STYLE

carries palette, lighting, and texture. Fast takes references without the switch; Lite takes none. Shared across the three generators at authoring time:

negativePrompt

resolution

(

720p

1080p

duration

(4, 6, or 8 seconds), and

seed

generateAudio

is required on every member and moves the price. The Extend member takes only

prompt

video

generateAudio

, and

seed

: no duration, resolution, or ratio controls. At authoring time price spanned roughly five times between Lite and full 3.1 for one image-to-video job, and one extension cost more than a fresh full generation, so aim for the shot in one 8 second pass and

dry_run

the same payload on two members before a batch.

模式由输入项决定（名称来自实时schema）：

模式	输入项	行为说明
文本	`prompt`	`aspectRatio` 支持16:9或9:16
首帧	`image` （+ `prompt` ）	从该帧开始；不匹配的图像可能会被裁剪
过渡	`image` + `lastFrameImage`	在静态图之间生成过渡；提示词需描述一个连续动作
参考	`referenceImages` （1到3张）	保持主体或风格一致性； `duration` 固定为8秒
扩展	`video` （仅Extend版本支持）	续接视频片段；仅支持16:9、短边为720p或1080p的源视频

image

与

referenceImages

互斥，不可同时使用。

referenceImagesType

仅在完整版Veo 3.1中支持，且对整个参考图数组统一生效：

ASSET

（默认值）用于参考主体、物体和场景；

STYLE

用于参考调色、光影和纹理。Fast版本支持参考图但无类型切换选项；Lite版本不支持参考图。撰写本文时，三个生成版本共有的参数包括：

negativePrompt

、

resolution

（

720p

、

1080p

）、

duration

（4、6或8秒）以及

seed

。

generateAudio

是所有版本的必填参数，且会影响费用。Extend版本仅支持

prompt

、

video

、

generateAudio

和

seed

参数：不支持时长、分辨率或比例控制。撰写本文时，同一张图生视频任务的费用，Lite版本与完整版3.1相差约5倍，而一次视频扩展的费用比一次全新的完整版生成还高，因此建议尽量用一次8秒的生成完成镜头，并在批量处理前先用相同的参数在两个版本上执行

dry_run

对比。

The soundstage is in the prompt

音场效果在提示词中指定

With

generateAudio: true

, Veo renders synchronized dialogue, effects, and ambience, and it takes audio direction literally. Put spoken lines in quotation marks, prefix effects with

SFX:

and room tone with

Ambient noise:

. A prompt with no audio direction still gets a soundtrack, just not the one you meant. When a clip needs silence for later scoring, set

generateAudio: false

rather than prompting for quiet.

当

generateAudio: true

时，Veo会生成同步的对话、音效和环境音，且会严格按照提示中的音频指令执行。将台词放在引号中，音效前加上

SFX:

前缀，环境音前加上

Ambient noise:

前缀。即使提示词中没有音频指令，Veo也会生成音轨，只是可能不符合你的预期。如果需要视频片段静音以便后期配乐，请设置

generateAudio: false

，而不是在提示词中要求安静。

Prompt one continuous shot

提示词描述单个连续镜头

Write present-tense prose ordered as cinematography, subject, action, context, style: camera and shot scale first, one primary arc, roughly 90 to 250 words. For several beats in one clip, timestamp lines work (

[00:00-00:02] Medium shot...

), spans fitting inside the chosen duration.

使用现在时态的散文体撰写，顺序为：镜头语言、主体、动作、背景、风格：先描述机位和景别，围绕一个核心情节展开，长度约90-250词。如果一个视频片段包含多个节奏点，可以使用带时间戳的行（如

[00:00-00:02] Medium shot...

），时间跨度需在选定的时长范围内。

Worked example: a dialogue shot from character references

实战示例：基于角色参考图生成对话镜头

```
search
```
with
```
target="models"
```
,
```
query="veo"
```
,
```
public=true
```
. Prefer the newest non-deprecated hits, e.g.
```
model_veo3-1
```
and
```
model_veo3-1-fast
```
(live hits at authoring time: re-discover each session).
```
model_schema_get
```
on the chosen id: confirm which reference inputs exist before writing the payload.
```
upload_asset
```
two character stills (see the
```
scenario
```
skill) to get asset ids.

model_run

with that

model_id

dry_run=true

, and

parameters={"prompt": "Medium shot, the knight lowers her visor and says, \"Hold the line.\" Torchlight flickers on wet stone. Ambient noise: distant thunder. SFX: metal visor clank.", "referenceImages": ["asset_a", "asset_b"], "referenceImagesType": "ASSET", "duration": 8, "resolution": "1080p", "aspectRatio": "16:9", "generateAudio": true}

. Estimate the same job on the Fast id (drop

referenceImagesType

, absent there) and compare.

Repeat
```
model_run
```
with
```
wait=false
```
, then
```
jobs_wait
```
with the returned job id, re-called with
```
pending_job_ids
```
on timeout, never a second
```
model_run
```
.
```
asset_display
```
the output and review with sound. To iterate on wording alone, hold an explicit
```
seed
```
fixed across reruns.

执行
```
search
```
，参数为
```
target="models"
```
、
```
query="veo"
```
、
```
public=true
```
。优先选择最新的未弃用结果，例如
```
model_veo3-1
```
和
```
model_veo3-1-fast
```
（撰写本文时的有效结果：每次会话需重新查找）。
对选定的模型ID执行
```
model_schema_get
```
：在编写请求参数前确认支持的参考输入项。
执行
```
upload_asset
```
上传两张角色静态图（参见
```
scenario
```
skill），获取素材ID。

执行

model_run

，传入该

model_id

、

dry_run=true

，以及参数

parameters={"prompt": "Medium shot, the knight lowers her visor and says, \"Hold the line.\" Torchlight flickers on wet stone. Ambient noise: distant thunder. SFX: metal visor clank.", "referenceImages": ["asset_a", "asset_b"], "referenceImagesType": "ASSET", "duration": 8, "resolution": "1080p", "aspectRatio": "16:9", "generateAudio": true}

。用Fast版本的模型ID估算同一任务的费用（需移除Fast版本不支持的

referenceImagesType

参数）并进行对比。

再次执行
```
model_run
```
，设置
```
wait=false
```
，然后用返回的任务ID执行
```
jobs_wait
```
，若超时则传入
```
pending_job_ids
```
重新调用，切勿重复执行
```
model_run
```
。
用
```
asset_display
```
展示输出结果，开音审核。如果仅需调整提示词措辞，可在多次重跑中固定明确的
```
seed
```
值。

Common mistakes

常见错误

Combining
```
image
```
with
```
referenceImages
```
: mutually exclusive; pick the opening state or consistency.
Any duration but 8 with
```
referenceImages
```
: reference mode supports only 8 seconds.
Sending
```
referenceImages
```
to Lite or
```
referenceImagesType
```
to Fast: the schema decides which inputs exist.
Omitting
```
generateAudio
```
: it is required on every member; pass it explicitly.
Extending a 9:16 clip: Extend takes 16:9 sources with a 720p or 1080p short side only.
Writing the prompt as a list of negatives: describe the wanted scene, reserve
```
negativePrompt
```
for what to discourage.

同时使用
```
image
```
和
```
referenceImages
```
：二者互斥，只能选择首帧设定或一致性参考。
使用
```
referenceImages
```
时设置非8秒的时长：参考模式仅支持8秒时长。
向Lite版本传入
```
referenceImages
```
，或向Fast版本传入
```
referenceImagesType
```
：输入项由模型schema决定。
遗漏
```
generateAudio
```
参数：所有版本都必填该参数，请显式传入。
扩展9:16的视频片段：Extend版本仅支持短边为720p或1080p的16:9源视频。
将提示词写成负面描述列表：应描述你想要的场景，将不需要的内容放在
```
negativePrompt
```
中。