| 1. Go to the [App Management](https://modelstudio.console.alibabacloud.com/?tab=app#/app-center) page. Click Create Application>Workflow Application, enter an App Name, and then click Create Now. | |
| 2. Add a large model node to detect scam messages. Drag the LLM node from the left-side pane to the canvas and configure the following parameters in its configuration panel. Leave other parameters at their default values. 1. Model Configuration: Select Qwen-Plus-latest. 2. Prompts: ```plaintext Analyze the provided message to determine if it is a potential scam. Provide a definitive answer. Requirements: Carefully review the message content, focusing on keywords and typical scam patterns, such as requests for urgent transfers, providing personal information, or promising unrealistic benefits. Steps: 1. Identify key elements in the message, including but not limited to the sender's identity, the request made, promised rewards, and any sense of urgency. 2. Compare the message against the characteristics of known scams to check for similar tactics or language patterns. 3. Evaluate the overall reasonableness of the message, considering whether the request aligns with standard procedures and common sense. 4. If the message contains links or attachments, do not click or download them to avoid potential security risks, and warn the user about the dangers of such content. Output Format: Clearly state whether the message exhibits characteristics of a scam and briefly explain your reasoning. If a scam is suspected, provide suggestions or preventive measures to protect the user. ``` 3. User Prompt: ```plaintext Determine if the message "${sys.query}" is a potential scam. ``` Close the panel and connect the Start node to the Large Model node. |
|
| 3. Connect the large model node to the end node. Then, click the end node and in its configuration panel's editor, enter `/` to insert the `Large Model 1/result` variable. Leave other parameters at their default values. |
|
|
4. Click Test in the upper-right corner. Enter `Your package has been at the pickup station for several days. Please come and collect it.` and click |
|
|
5. Next, enter `You have a message saying you've won 1 million. Please check it.` and click |
|
| 1. Go to the [App Management](https://modelstudio.console.alibabacloud.com/?tab=app#/app-center) page. Click Create Application>Workflow Application, enter an App Name, and then click Create Now. | |
|
2. Drag the Intent Classification node from the left-side pane to the canvas and configure the following parameters in its configuration panel. Leave other parameters at their default values.
- Input Variable: Select Built-in Variable > query.
- Select Model: Select Qwen-Plus-latest.
- Intent Classification: Click Add Intent and add the following three intents:
- Buy TV
- Buy mobile phone
- Buy refrigerator
- Memory: Enables the model to remember previous chat content during a conversation. Turn on the toggle |
|
|
3. Drag a LLM node from the left-side pane to the canvas. In its configuration panel, configure the following parameters, leaving others at their default values. Then, close the panel and connect the Buy refrigerator output of the Intent Classification node to the Large Model node.
- Model Configuration: Select Qwen-Plus-latest.
- Prompts:
```plaintext
You are a smart shopping assistant responsible for recommending refrigerators to customers.
You must proactively ask the user for refrigerator parameters in the order listed in the [Refrigerator Parameter List] below. Ask for only one parameter at a time, and do not repeat questions for the same parameter.
If the user provides a value for a parameter, continue to ask about the remaining parameters.
If the user asks for the definition of a parameter, provide an explanation based on your professional knowledge, and then continue to ask which parameter value they prefer.
If the user indicates they no longer wish to make a purchase, output: Thank you for visiting. We look forward to serving you next time.
[Refrigerator Parameter List]
1. Usage Scenario: [Home, Small commercial, Large commercial]
2. Capacity: [200L, 300L, 400L, 500L]
3. Energy Efficiency Rating: [Level 1, Level 2, Level 3]
After all parameters from the [Refrigerator Parameter List] have been collected, ask: "Are you sure you want to purchase?" while also showing the parameters the customer has selected, for example: For small commercial use|300L|Level 1. Ask if they are sure about the refrigerator with these parameters. If the customer decides not to buy, ask which parameters they want to adjust.
If the customer confirms the parameters meet their requirements, you must output in the following format:
[Usage Scenario: Home, Capacity: 300L, Energy Efficiency Rating: Level 1]. Output only the content in this format and no other information.
```
- User Prompt:
```plaintext
User's question: ${sys.query}
```
- Memory: Enables the model to remember previous chat content during a conversation. Turn on the toggle |
|
|
4. Drag a LLM node from the left-side pane to the canvas. In its configuration panel, configure the following parameters, leaving others at their default values. Then, close the panel and connect the Buy TV output of the Intent Classification node to the Large Model node.
- Model Configuration: Select Qwen-Plus-latest.
- Prompts:
```plaintext
You are a smart shopping assistant responsible for recommending TVs to customers.
You must proactively ask the user for TV parameters in the order listed in the [TV Parameter List] below. Ask for only one parameter at a time, and do not repeat questions for the same parameter.
If the user provides a value for a parameter, continue to ask about the remaining parameters.
If the user asks for the definition of a parameter, provide an explanation based on your professional knowledge, and then continue to ask which parameter value they prefer.
If the user indicates they no longer wish to make a purchase, output: Thank you for visiting. We look forward to serving you next time.
[TV Parameter List]
1. Screen Size: [50-inch, 70-inch, 80-inch]
2. Refresh Rate: [60Hz, 120Hz, 240Hz]
3. Resolution: [1080P, 2K, 4K]
After all parameters from the [TV Parameter List] have been collected, ask: "Are you sure you want to purchase?" while also showing the parameters the customer has selected, for example: 50-inch|120Hz|1080P. Ask if they are sure about the TV with these parameters. If the customer decides not to buy, ask which parameters they want to adjust.
If the customer confirms the parameters meet their requirements, you must output in the following format:
[Screen Size: 50-inch, Refresh Rate: 120Hz, Resolution: 1080P]. Output only the content in this format and no other information.
```
- User Prompt:
```plaintext
User's question: ${sys.query}
```
- Memory: Enables the model to remember previous chat content during a conversation. Turn on the toggle |
|
|
5. Drag a LLM node from the left-side pane to the canvas. In its configuration panel, configure the following parameters, leaving others at their default values. Then, close the panel and connect the Buy mobile phone output of the Intent Classification node to the Large Model node.
- Model Configuration: Select Qwen-Plus-latest.
- Prompts:
```plaintext
You are a smart shopping assistant responsible for recommending mobile phones to customers.
You must proactively ask the user for mobile phone parameters in the order listed in the [Mobile Phone Parameter List] below. Ask for only one parameter at a time, and do not repeat questions for the same parameter.
If the user provides a value for a parameter, continue to ask about the remaining parameters.
If the user asks for the definition of a parameter, provide an explanation based on your professional knowledge, and then continue to ask which parameter value they prefer.
If the user indicates they no longer wish to make a purchase, output: Thank you for visiting. We look forward to serving you next time.
[Mobile Phone Parameter List]
1. Usage Scenario: [Gaming, Photography, Watching movies]
2. Screen Size: [6.4-inch, 6.6-inch, 6.8-inch, 7.9-inch foldable screen]
3. RAM + Storage Space: [8GB+128GB, 8GB+256GB, 12GB+128GB, 12GB+256GB]
After all parameters from the [Mobile Phone Parameter List] have been collected, ask: "Are you sure you want to purchase?" while also showing the parameters the customer has selected, for example: For photography|8GB+128GB|6.6-inch. Ask if they are sure about the mobile phone with these parameters. If the customer decides not to buy, ask which parameters they want to adjust.
If the customer confirms the parameters meet their requirements, you must output in the following format:
[Usage Scenario: Photography, Screen Size: 6.8-inch, Storage Space: 128GB, RAM Space: 8GB]. Output only the content in this format and no other information.
```
- User Prompt:
```plaintext
User's question: ${sys.query}
```
- Memory: Enables the model to remember previous chat content during a conversation. Turn on the toggle |
|
| 6. Drag a Variable Processing node from the left-side pane to the canvas. In its configuration panel, configure the following parameters, leaving others at their default values. Then, close the panel and connect the Default intent output of the Intent Classification node to the Variable Processing node. - Output Mode: Select Text Output. - In the editor box, enter: ```plaintext Hello! Please tell me if you want to buy a TV, a refrigerator, or a mobile phone. ``` |
|
| 7. Connect all three large model nodes and the Variable Processing node to the End node. Then, click the end node and, in its configuration panel's editor, enter `/` to insert the following four variables: `Large Model 1/result`, `Large Model 2/result`, `Large Model 3/result`, and `Variable Processing 1/result`. Leave other parameters at their default values. |
|
|
8. Test the application:
1. Click Test in the upper-right corner. In the dialog box, enter `Tell me about your refrigerators`, click |
|
Component | Description |
|---|---|
Predefined variables | The workflow provides the following predefined variables to process user input and maintain conversation history:
|
Custom variables | Custom variables are structured input parameters that you define for a workflow. They receive data from tests or API calls and can be referenced in subsequent nodes. When creating a custom variable, configure the following parameters:
|
Component | Description |
|---|---|
Output mode | Text Output: Suitable for unstructured content. In the input box, you can enter fixed content or type JSON Output: Suitable for outputting structured content in JSON format. You can define variable names and enter text or reference variables. |
Streaming output | The Streaming Output switch applies only to Text Output mode. When enabled, responses from large model and application component nodes are streamed token by token. When disabled, a full response is returned after it is completely generated. |
| Parameter | Description |
|---|---|
| Mode selection | Single Mode: Calls the large model once. |
|
Batch Mode: The node runs multiple times, sequentially assigning an item from a list to a batch variable in each run. This process stops when all list items are processed or the configured maximum batch runs is reached.
Batch Configuration:
- Maximum Number of Batches (Range: 1–100; default for standard users: 100): The maximum number of times the batch process will run.
|
|
| Model configuration | Select a suitable large model and adjust its parameters. The available models are listed on the UI. For a detailed introduction to the models, see the model list. For information on the API call rate limits for each model, see [rate limiting](/help/en/model-studio/rate-limit). |
| Parameter configuration | Temperature: Controls the diversity of the generated content. A higher temperature value increases the randomness of the generated text, producing more unique outputs, while a lower value makes the content more conservative and consistent. > This configuration is not currently supported for DeepSeek R1 series models. |
| Maximum Reply Length: Limits the maximum length (in tokens) of the text generated by the model, not including the prompt. This limit varies by model. | |
| The following parameters are available only when a Qwen-VL model is selected: - Model Input Parameters: For `vlImageUrl`, you can reference a variable or input an image URL. - Image Source: - Image set: The model treats each uploaded image independently, selecting the most relevant one to answer the user's question. A single image can be passed directly. For example: `https://****.com/****.jpg`. Multiple images can be passed as a list. For example: `["URL","URL","URL"]`. - Video frames: The model considers the uploaded images to be from the same video and understands them sequentially as a whole. You must provide a minimum of 4 video frames. | |
| System prompt | Provides system-level instructions to the model. You can use it to define the model's role, task, or output format. For example: "You are a math expert who specializes in solving math problems. Please output the problem-solving process and result in the correct format." |
| User prompt | The user's input to the model, such as a request or instruction. You can configure a prompt template or insert variables. The model uses this input to generate a response. |
| Memory | Memory is the context the model retains across multiple turns in a conversation. - Current node cache: Uses the output of this node as context. The model will only remember context generated within this node. - memory rounds: The number of conversation rounds to remember. One input and its corresponding output constitute one round. - Custom cache: Uses a specified context variable as the context. - Context variable: Select the source of the context. |
| Output variable | The name of the variable that stores the processing result of this node. Subsequent nodes can use this variable to access this node's result. > DeepSeek R1 series models support outputting the Deep Thinking process (`reasoningContent`). |
| Retry on failure | If disabled, the node stops executing when an error occurs. If enabled, the node re-executes on error, according to the configured number of retries and interval. - maximum retries: The maximum number of retry attempts when a request fails. - retry interval: The time interval between each retry attempt, in milliseconds. |
| Exception handling |
If disabled, the node handles errors using the system's default error handling mechanism.
If enabled, the node executes custom processing logic according to the configuration when an error occurs.
- Default value: The content in `result` is output when an exception occurs.
- exception branch: If an exception occurs, this branch is executed. You must configure its processing flow.
|
| Return results | This feature is deprecated. This option applies only to API calls and determines whether the node's content is output. To learn more about the purpose of this component, see [Application calling](/help/en/model-studio/application-calling-guide). |
Parameter | Description |
|---|---|
content | The text used to query the knowledge base. You can enter text directly or type |
imageList | The images used to query the knowledge base. You can enter a publicly accessible image URL, such as |
Selection method | Applicable scenarios | Configuration |
|---|---|---|
Select Fixed Knowledge Base | Each call uses the same, pre-selected knowledge base. | Select a specific knowledge base from the drop-down list. You can add three types of knowledge bases: documents, tables, and images. |
Dynamic Import | The output of an upstream node dynamically determines which knowledge bases to use. | Configure the |
Invocation method | Description | Parameters |
|---|---|---|
always invoke | Performs a knowledge base retrieval for each user input. This method is suitable for high-frequency question-and-answer scenarios. | None |
intelligent invoke | The agent determines whether to query the knowledge base based on the user's input and its description. This is suitable for flexible conversational scenarios. | Knowledge Base Description (Required) |
legacy invocation | Results from document, table, and image knowledge bases are recalled together. A single topK parameter controls the number of recalled chunks, and does not support separate debugging for each type. | topK (Range: 1–50, Default: 10) |
Field | Type | Description |
|---|---|---|
result | Object | Contains all information returned from the current query. |
result.chunkList | Array\ | An array of recalled chunks. This array is empty if the query returns no results. |
result.chunkList\[ ].content | String | The original content of the recalled knowledge base chunk. |
result.chunkList\[ ].title | String | The title of the document to which the chunk belongs. |
result.chunkList\[ ].documentName | String | The name of the document where the recalled chunk is located. |
result.chunkList\[ ].score | Number | The similarity score of the knowledge base chunk. A higher score indicates a better match. |
result.chunkList\[ ].id | String | The chunk ID. |
result.chunkList\[ ].dataId | String | The document ID. |
result.chunkList\[ ].docUrl | String | The source file download link. |
result.chunkList\[ ].knowledgeBaseId | String | The knowledge base ID. |
result.chunkList\[ ].nid | String | The identifier for the original text. |
result.chunkList\[ ].images | Array\ | A list of images. |
result.chunkList\[ ].pageNumber | Array\ | An array of page numbers. |
result.rewriteQuery | String | The rewritten user query. |
Parameter | Description |
|---|---|
API address | The request URL for the API.
|
Header settings | Specifies the HTTP request headers, such as Content-Type and Authorization. You can enter values directly or reference them from previous nodes. |
Parameter settings | Query parameters in the request URL. You can enter values directly or reference them from previous nodes. For example, |
Body settings | Select the correct type based on the API's requirements.
|
Timeout setting (s) | The maximum time to wait for a response. If a response is not received within this time, the request times out. |
Retry on failure | If disabled, the node stops executing when an error occurs. If enabled, the node retries the request according to the configured maximum retries and retry interval when an error occurs.
|
Exception handling | If disabled, the node uses the default error handling mechanism when an error occurs. If enabled, the node executes custom logic when an error occurs.
|
Output | Stores the API response in a specified variable for use by subsequent nodes. |
Parameter | Description |
|---|---|
Input | Specifies the content for the node to process. The variables change depending on the selected plugin. When you select a plugin, its built-in variables load automatically. You can go to the Official plugins or Custom plugins page, click a plugin card, and view how to configure its variables. |
Output | The output variable that contains the node's processing result. Subsequent nodes use this variable to access the result. |
Retry on failure | If disabled, the node stops executing when an error occurs. If enabled, the node retries execution based on the configured number of retries and interval when an error occurs.
|
Exception handling | If disabled, the node uses the system's default error handling mechanism. If enabled, the node executes custom logic when an error occurs.
|
Parameter | Description |
|---|---|
Input | The input variable that this node processes. You can reference variables from upstream nodes or enter values directly. |
Region | Select the region where your Function Compute service is located: Singapore, Kuala Lumpur, or Indonesia (Jakarta). |
Service configuration | Select the Function Compute service to invoke. You must first create a function. The Alibaba Cloud account that creates the Function Compute service must be the same as the account you use to log in to Model Studio (Bailian), or it must belong to the same Alibaba Cloud root account. |
Output | The variable that stores the node's output. Subsequent nodes can reference this variable. |
Parameter | Description |
|---|---|
input | Accepts a fixed value, or a reference to a variable from an upstream node or a session variable. |
code | Defines the core logic for the node. Both JavaScript and Python are supported.
|
output | The dictionary returned by the For example, if you return |
retry on failure | If disabled, the node stops executing when an error occurs. If enabled, the node retries execution according to the configured maximum retries and retry interval.
|
exception handling | If disabled, the node uses the system's default error handling mechanism. If enabled, the node executes custom logic when an error occurs.
|
Parameter | Description |
|---|---|
conditional branch | Enter the conditional expressions. Conditions in different groups are joined by |
else | The default output when no defined conditions are met. |
| Parameter | Description |
|---|---|
| input variable | Specifies the input variable to be classified. You can enter a value directly, or use a variable from a preceding node or a session variable. |
| model selection | Supports Qwen-Plus and the intent classification model. |
| intent classification | Define the intents for the model to classify. The model matches the input to an intent based on its description and then routes to the corresponding link. For example: "For calculating math problems" or "For answering weather-related questions". |
| Other intent | This branch executes when the model determines that the input does not match any of the defined intents. |
| intent mode | - single-select mode: The large model selects the single best-matching intent from the available configurations. - multi-select mode: The large model selects all matching intents from the available configurations. |
| reasoning mode | - fast mode: This mode improves processing speed by avoiding complex reasoning steps, making it suitable for simple scenarios. - quality mode: This mode uses step-by-step reasoning to achieve more accurate classification. |
| memory | Refers to the model's ability to remember conversation history in multi-turn conversations, which serves as the context. - node-local cache: Uses the output of this node as context. The model's memory is limited to the context generated by this node. - memory turns: The number of conversation turns to remember. One input and its corresponding output constitute one turn. - custom cache: Uses the content of a specified context variable as the context. - context variable: Specifies the source of the context. |
| prompt |
Provides additional instructions or constraints for the intent classification model. You can add examples or define constraints to help the model's classification results better match your requirements.
|
| output | The variable that stores the node's output. Downstream nodes can use this variable to access the result. |
Parameter | Description |
|---|---|
Output content | Enter fixed content or type |
Streaming output | When enabled, content from a large model is streamed to the conversation token by token. When disabled, the entire response is generated first and then output at once. |
Parameter | Description |
|---|---|
Output mode |
|
Output variable | Specifies the variable to store the node's output for use by subsequent nodes. |
Return result | This feature is being deprecated. Applicable only to API calls, this parameter determines whether the node's output is returned. See application call. |
Parameter | Description |
|---|---|
Input | The text to extract parameters from. |
Model selection | Specifies the model used to extract parameters. Choose the one that best suits your business needs. |
Parameters to extract | The model extracts parameters from the input based on the name, type, and description you provide. |
Prompt | Additional instructions to guide the model during parameter extraction. |
Memory | Memory provides conversational context, helping the model remember previous interactions in multi-turn scenarios.
|
Output | Contains the parameters extracted by the model. Subsequent nodes can reference this output as a variable. |
Parameter | Description |
|---|---|
Input | Specifies the content that the node will process. You can reference variables from a previous node. |
Model selection | Select a decision model. |
Group name | Enter a custom name for the agent group. |
Agents | Specifies the sub-agents within the agent group. Only published agent applications can be added. After adding a sub-agent, click Configure to describe its function. This description helps the decision model determine which sub-agent to invoke for the current task. |
Output variable | Specifies the variable for storing the output of this node. This variable can be referenced by subsequent nodes. |
Parameter | Description |
|---|---|
Input | The content to be processed by this node. You can reference variables from preceding nodes. |
Agent name | A custom name for the agent. |
Model selection | The large model used by the agent. |
Prompt | Use natural language to define the agent's role and tasks. |
Knowledge base | Select a knowledge base for the agent. |
Plugin | Enables the agent to call official or custom plugins to extend its capabilities. |
Output variable | The variable that stores the node's output for use in subsequent nodes. |
Model type | Supported models | Features |
|---|---|---|
Image generation | Qwen-Image | Excels at rendering complex Chinese and English text. |
Wan 2.6-image-generation | Generates and edits images from text instructions, supporting both image editing and mixed text-and-image modes. | |
Qwen-Image-Plus | Excels at rendering complex Chinese and English text. | |
Z-Image-Turbo | A lightweight text-to-image model that quickly generates high-quality images. It supports bilingual rendering (Chinese and English), complex semantic understanding, and multiple styles and themes. | |
Wan 2.2-text-to-image-Plus | The Pro edition of Wan 2.2, with comprehensive improvements in creativity, stability, and photorealism. | |
Wan 2.2-text-to-image-Flash | The Flash edition of Wan 2.2, with comprehensive improvements in creativity, stability, and photorealism. | |
Video generation | Wan 2.6-text-to-video | Wan 2.6. Features multi-shot storytelling capabilities and supports both automatic voice-overs and custom audio file inputs. |
Wan 2.6-image-to-video | Wan 2.6. Features multi-shot storytelling capabilities and supports both automatic voice-overs and custom audio file inputs. | |
Wan 2.5-text-to-video-Preview | Wan 2.5 preview. Supports automatic voice-overs and custom audio file inputs. | |
Wan 2.5-image-to-video-Preview | Wan 2.5 preview. Supports automatic voice-overs and custom audio file inputs. | |
Wan 2.2-text-to-video-Plus | The Pro edition of Wan 2.2. Offers more accurate instruction understanding, stable and smooth motion generation, and richer details in the output. | |
Wan 2.2-image-to-video-Plus | The Pro edition of Wan 2.2, offering more accurate instruction understanding, controllable camera movement, and consistent visuals. It also features comprehensive improvements in stability, success rate, and content richness. | |
Wan 2.2-image-to-video-Flash | The Flash edition of Wan 2.2, delivering ultra-fast generation, more accurate instruction understanding and camera control, consistent visuals, and comprehensive improvements in stability and success rate. | |
Audio generation | Qwen3-TTS-Flash | Supports mixed-language text input and streaming audio output. |
Parameter | Required | Description |
|---|---|---|
Positive prompt | Yes | Describes the content, style, composition, and other elements of the target image. You can enter text directly or type |
Negative prompt | No | Describes content that you do not want in the image, such as blurriness or watermarks. You can enter text directly or type |
Size/Resolution | Yes | The dimensions of the image. |
prompt\_extend | No | If enabled, the system uses a large model to intelligently rewrite the input prompt. This feature applies only to the positive prompt. It significantly improves results for shorter prompts but adds 3-4 seconds of latency. |
Add Watermark | No | Controls whether to add an Alibaba Cloud watermark to the generated image. This feature is disabled by default. |
Reference image | No | Upload a reference image to guide the style or content of the generated image. You can pass a File-type variable from the start node or a public URL. |
Number of Images | No | The number of images to generate in a single request. The default value is 1. |
Intelligent prompt rewriting | No | If enabled, a large model intelligently rewrites the input prompt. This feature applies only to the positive prompt. It significantly improves results for shorter prompts but increases the image generation time. |
enable\_interleave | No | Disabled (default): Image editing mode. Edits, performs style transfer, or generates consistent subjects using 1 to 4 input images. Enabled: Mixed text-and-image output mode. Generates a single content block containing both text and images, based on an image or plain text. |
Random seed | No | Controls randomness in the generated output (default: 1234). Using the same seed value and prompt produces similar images. |
Intelligent Thinking | No | If enabled, the large model performs reasoning and prompt rewriting. This can improve the generation quality but also increases the generation time. |
Parameter | Required | Description |
|---|---|---|
Positive prompt | Yes | Describes the scene, actions, style, and other elements of the target video. You can enter text directly or type |
Negative prompt | No | Describes content that you do not want in the video, such as blurriness or watermarks. You can enter text directly or type |
Resolution | Yes | The video resolution. The available options vary by model. |
Video Duration | Yes | The duration of the generated video. The available options vary by model. |
Random seed | Yes | Controls randomness in the generated output. Using the same seed value and prompt produces similar videos. |
Intelligent Expansion | No | If enabled, the system uses a large model to intelligently rewrite the input prompt to improve the video generation quality. This feature applies only to the positive prompt. |
Smart Multi-shot | No | If enabled, the model generates a multi-shot video. |
Generate Audio | No | If enabled, you can add a voice-over by providing a public audio file URL. |
Reference image | Yes (image-to-video models only) | Specifies a reference image to serve as the video's first frame. The model generates the video based on this image. You can pass a File-type variable from the start node or a public URL. |
Parameter | Required | Description |
|---|---|---|
Text for Synthesis | Yes | The text content to convert to speech. You can enter text directly or type |
Language Type | No | The language for the synthesized speech (default: Chinese). Supported languages: Chinese, English, German, Italian, Portuguese, Spanish, Japanese, Korean, French, and Russian. |
Voice | No | The voice for the synthesized speech. For a list of supported voices, see Real-time Speech Synthesis - Qwen. |
Parameter | Type | Description | Example |
|---|---|---|---|
output | Object | The complete output object, containing the task status, execution time, and results. | - |
output.task\_status | String | The task status, such as "SUCCEEDED" or "FAILED". | SUCCEEDED |
output.submit\_time | String | The task submission time, in YYYY-MM-DD HH:mm:ss.SSS format. | 2026-01-22 10:54:44.200 |
output.end\_time | String | The task end time, in YYYY-MM-DD HH:mm:ss.SSS format. | 2026-01-22 10:54:51.685 |
output.task\_id | String | The unique identifier for the task. | e36c2221-b7fd-xxxx |
output.scheduled\_time | String | The task scheduled time, in YYYY-MM-DD HH:mm:ss.SSS format. | 2026-01-22 10:54:44.251 |
output.results | Array\ | An array of generation results, containing metadata for the generated assets. | - |
output.results\[].orig\_prompt | String | The original prompt entered by the user. | a cute cat sitting on a windowsill |
output.results\[].actual\_prompt | String | The prompt used for generation after model expansion. | a cute white cat with fluffy, soft fur... |
output.results\[].url | String | The URL of the generated file. | |
usage | Object | Model usage statistics. | - |
usage.image\_count | Number | The number of generated images (image generation mode). | 1 |
urls | Array\ | An array of URLs for the generated files. Use these URLs to access or download the output. | \["https\://dashscope-result..."] |
Dify node | Mapped Model Studio node | Compatibility |
|---|---|---|
start | start |
|
LLM | LLM |
|
knowledge retrieval | knowledge base |
|
Direct Reply | output node | Fully compatible. |
agent | None | Only the name is retained. You must click the node and select a specific Model Studio node as a replacement. |
Question Classifier | intent classification | Model Studio clears the model field for unsupported models, requiring you to select one manually. Supported models are fully compatible. |
Iteration | batch processing |
|
loop | loop | Fully compatible. |
code execution | script | Model Studio distinguishes between Python and JavaScript scripts. |
Template Transform | None | Not compatible. Model Studio generates a custom node. |
Variable Aggregator | Variable Processing | Maps to the Aggregate Groups output mode of the Variable Processing node. |
Document Extractor | None | Not compatible. Model Studio generates a custom node. |
Variable Assignment | Variable Settings | Fully compatible. |
Parameter Extractor | parameter extraction | Model Studio does not support inference mode. Other features are fully compatible. |
HTTP request | API | Fully compatible, but you must re-authenticate. |
List Operation | None | Not compatible. Model Studio generates a custom node. |
Tool | plugin, MCP | Not compatible. Model Studio generates a custom node. |
comment | None | Not compatible. |
end | end | If a Dify workflow contains multiple end nodes, Model Studio converts them into a single Variable Processing node and a single end node. |
| 1. Click Publish in the top-right corner of the workflow configuration page. In the Publish dialog box, enter version information, such as 1.0.0, and click OK. |
|
| 2. Click Version Management at the top of the page, and in the Historical Versions panel, you can view or use different versions of the current workflow application as needed by clicking Overwrite Current Draft or Return to Current Version. > You can also click Export DSL for This Version at the top to export the DSL of the selected historical version. |
|
| 3. Optional: In the Node Library, view or search for nodes. |
|
In My Applications, find a published application card and click the |