runway-dev-models

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Runway Dev — Models

Runway Dev — 模型

Companion: Use
+runway-dev
for shared guidance when available. If it is not installed, inspect the workspace, probe
RUNWAYML_API_SECRET
without printing it, and read current docs. Encourage connecting Dev MCP for live model access and constraints. If the user declines or cannot connect, continue from current docs and existing model configuration.

配套工具： 若可用，请使用
+runway-dev
获取共享指导。如果未安装该工具，请检查工作区，探查
RUNWAYML_API_SECRET
但不要打印它，并查阅当前文档。建议连接Dev MCP以获取实时模型访问权限和约束信息。如果用户拒绝或无法连接，请根据当前文档和现有模型配置继续操作。

Goal

目标

Help the user choose and integrate a model endpoint (

/v1/text_to_video

/v1/text_to_image

, etc.) across their application, including its backend call and existing UI. Verify changes with one working SDK call when safe.

帮助用户选择并在其应用中集成模型端点（

/v1/text_to_video

、

/v1/text_to_image

等），包括后端调用与现有UI的对接。在安全前提下，通过一次可运行的SDK调用验证变更。

MCP tools

MCP工具

```
list_models
```
— discover models the selected project can call and read each model's
```
inputConstraints
```
.
```
get_credit_balance
```
— check budget before billable verification.
```
get_task
```
— inspect or debug an existing task, not replace SDK wait helpers in application code.

```
list_models
```
— 发现所选项目可调用的模型，并读取每个模型的
```
inputConstraints
```
。
```
get_credit_balance
```
— 在进行需计费的验证前检查预算。
```
get_task
```
— 检查或调试现有任务，而非替换应用代码中的SDK等待助手。

Workflow

工作流程

Inspect the existing application. Confirm the user experience, target modality, and where generation inputs and outputs belong.
If the task requires model selection, access verification, or current constraints, call
```
list_models
```
with
```
{ projectId, endpoint }
```
. If existing code pins a model, proceed from current docs unless live access must be verified.
Follow the endpoint docs linked by
```
llms.txt
```
; do not infer request fields from another model.
Keep
```
RUNWAYML_API_SECRET
```
behind the application's server boundary.
Implement or update the SDK call by chaining
```
.waitForTaskOutput()
```
in Node or
```
.wait_for_task_output()
```
in Python directly from the create call.
If the application has a UI, wire its controls to the backend and render loading, error, and generated-output states.
When verification is appropriate, submit one test generation. Present the result and offer to persist output before its signed URL expires.

检查现有应用。确认用户体验、目标模态，以及生成输入和输出的归属位置。
如果任务需要模型选择、访问验证或获取当前约束条件，请调用
```
list_models
```
并传入
```
{ projectId, endpoint }
```
。如果现有代码已固定模型，则根据当前文档操作，除非必须验证实时访问权限。
遵循
```
llms.txt
```
链接的端点文档；不要从其他模型推断请求字段。
将
```
RUNWAYML_API_SECRET
```
保留在应用的服务器边界之后。
实现或更新SDK调用，在Node中直接从创建调用链式调用
```
.waitForTaskOutput()
```
，在Python中链式调用
```
.wait_for_task_output()
```
。
如果应用有UI，将其控件与后端对接，并渲染加载、错误和生成输出状态。
当适合进行验证时，提交一次测试生成。展示结果，并在签名URL过期前提供持久化输出的选项。

Input media

输入媒体

Follow the current input docs linked by
```
llms.txt
```
: use a public HTTPS URL, a small data URI, or an ephemeral upload.
Send browser-selected files to the application's server, then use the SDK upload helper and pass its
```
runway://
```
URI to generation. Local filesystem paths cannot be API inputs.
Do not accept arbitrary remote URLs from clients. Prefer uploads or allowlisted origins.
Ephemeral inputs and generated output URLs expire; persist anything the application must retain.

遵循
```
llms.txt
```
链接的当前输入文档：使用公共HTTPS URL、小型数据URI或临时上传。
将浏览器选择的文件发送到应用服务器，然后使用SDK上传助手，并将其
```
runway://
```
URI传入生成接口。本地文件系统路径不能作为API输入。
不要接受客户端传来的任意远程URL。优先选择上传或允许列表中的源。
临时输入和生成的输出URL会过期；应用必须保留的内容需进行持久化处理。

Do not

禁止操作

Guess model ids, ratios, or durations from memory or other models.
Put API keys in frontend bundles.
Add manual polling when the SDK wait helper fits, or resubmit on transient read errors.

凭记忆或其他模型猜测模型ID、比例或时长。
将API密钥放入前端包中。
在SDK等待助手适用的情况下添加手动轮询，或在出现临时读取错误时重新提交请求。

Docs

文档

Index: https://docs.dev.runwayml.com/llms.txt
Setup: https://docs.dev.runwayml.com/guides/setup/

索引：https://docs.dev.runwayml.com/llms.txt
设置指南：https://docs.dev.runwayml.com/guides/setup/