The Multimodal Generation node invokes Alibaba Cloud’s multimodal models to generate images, videos, or audio content based on configured prompts and parameters. It is suitable for use cases such as content creation, marketing asset generation, short-video production, and voice-over synthesis.
Adding the Node
Method 1: Add via the Plus (+) Button
- Click the plus (+) button at the end of any node on the canvas.
- In the pop-up node list, select the Multimodal Generation node under the Basic category.
- The node is automatically added to the canvas and connected to the upstream node.
Method 2: Drag and Drop
- Locate the Multimodal Generation node in the Basic category of the left-hand node library.
- Drag the node onto the desired location on the canvas.
- Manually connect it to an upstream node (drag the output port of the upstream node to the input port of the Multimodal Generation node).
Core Configuration
Model Selection
Select a target multimodal generation model from the model dropdown menu. Models are functionally categorized into three types:
- Image Generation
- Video Generation
- Audio Generation
Parameter Configuration
Supported configuration parameters vary significantly across models. The configuration panel dynamically displays only the relevant options based on the selected model.
Image Generation Configuration
After selecting an image-generation model, configure the following parameters. Supported parameters differ by model—refer to the actual UI for availability.
| Parameter | Required | Description |
|---|---|---|
| Positive Prompt | Yes | Describes key elements of the target image, including subject, scene, composition, style, lighting, etc. Supports direct text input or insertion of upstream node output variables. |
| Negative Prompt | No | Specifies undesired elements (e.g., blur, watermarks). Supports direct text input or insertion of upstream node output variables. |
| Size / Resolution | Yes | Output image dimensions. |
prompt_extend | No | When enabled, the system uses a large language model (LLM) to intelligently rewrite the input prompt—applies only to the positive prompt. Improves generation quality for short prompts but adds ~3–4 seconds latency. |
| Add Watermark | No | Toggle for adding an Alibaba Cloud watermark to generated images (disabled by default). |
| Reference Image | No | Upload a reference image to guide style or content. Supports File-type variables from the Start node or public URLs. |
| Number of Images | No | Number of images to generate per request (default: 1). |
| Prompt Intelligent Rewrite | No | When enabled, an LLM rewrites the input prompt intelligently—applies only to the positive prompt. Improves output for short prompts but increases generation latency. |
enable_interleave | No | Disabled (default): Image editing mode—performs editing, style transfer, or subject-consistent generation based on 1–4 input images. Enabled: Text–image interleaved output mode—generates exactly one mixed-content block containing both text and image, anchored to either input images or plain text. |
| Random Seed | Yes | Controls output randomness (default: 1234). Identical seed + prompt yields highly similar results. |
| Intelligent Reasoning | No | When enabled, the LLM performs reasoning and prompt rewriting, improving generation quality—but increases latency. |
Video Generation Configuration
After selecting a video-generation model, configure the following parameters. Supported parameters vary by model—refer to the actual UI for availability.
| Parameter | Required | Description |
|---|---|---|
| Positive Prompt | Yes | Describes key elements of the target video, including scene, action, style, camera movement, timing, etc. Supports direct text input or insertion of upstream node output variables. |
| Negative Prompt | No | Specifies undesired elements (e.g., blur, watermarks). Supports direct text input or insertion of upstream node output variables. |
| Resolution | Yes | Output video resolution; supported options vary by model. |
| Video Duration | Yes | Generated video length; supported durations vary by model. |
| Random Seed | Yes | Controls output randomness. Identical seed + prompt yields highly similar results. |
| Intelligent Expansion | No | When enabled, the system uses an LLM to intelligently expand the input prompt to improve video generation quality—applies only to the positive prompt. |
| Multi-Shot Mode | No | When enabled, the output video is rendered in multi-shot format. |
| Generate Audio | No | When enabled, provides an audio file URL for voice-over synchronization. Only publicly accessible URLs are supported. |
| Reference Image | Yes (for Image-to-Video only) | Provides a reference image as the first frame. The model generates the video based on this image. Supports File-type variables from the Start node or public URLs. |
Audio Generation Configuration
| Parameter | Required | Description |
|---|---|---|
| Synthesis Text | Yes | Text content to convert into speech. Supports direct text input or insertion of upstream node output variables. |
| Language | No | Target language for speech synthesis (default: Chinese). Supported languages: Chinese, English, German, Italian, Portuguese, Spanish, Japanese, Korean, French, Russian. |
| Voice Style | No | Voice style/tone for synthesized speech. |
Node Outputs
Upon successful execution, the Multimodal Generation node outputs the following variables:
| Variable Name | Type | Description | Example |
|---|---|---|---|
output | Object | Full output object containing task status, execution time, generated results, etc. | — |
output.task_status | String | Task status, e.g., "SUCCEEDED" (success), "FAILED" (failure) | SUCCEEDED |
output.submit_time | String | Task submission time (format: YYYY-MM-DD HH:mm:ss.SSS) | 2026-01-22 10:54:44.200 |
output.end_time | String | Task completion time (format: YYYY-MM-DD HH:mm:ss.SSS) | 2026-01-22 10:54:51.685 |
output.task_id | String | Unique task identifier | e36c2221-b7fd-xxxx |
output.scheduled_time | String | Task scheduling time (format: YYYY-MM-DD HH:mm:ss.SSS) | 2026-01-22 10:54:44.251 |
output.results | Array<Object> | Array of generated results, each containing file metadata | — |
output.results[].orig_prompt | String | Original prompt (user-provided) | A cute kitten sitting on a windowsill |
output.results[].actual_prompt | String | Actual prompt used (post-LLM expansion) | A cute white kitten with fluffy soft fur... |
output.results[].url | String | Publicly accessible URL of the generated file | https://dashscope-result... |
usage | Object | Model usage statistics | — |
usage.image_count | Number | Number of generated images (image-generation mode only) | 1 |
urls | Array<String> | List of generated file URLs—directly usable for access or download | ["https://dashscope-result..."] |
Common Usage Patterns
| Paired Node(s) | Typical Use Case |
|---|---|
| Start Node → Multimodal Generation Node | Accepts user-submitted text (e.g., query variable) as the positive prompt to generate corresponding images, videos, or audio. |
| Large Language Model Node → Multimodal Generation Node | Uses LLM-generated text as the positive prompt or synthesis text to enable text-to-multimodal conversion. |
| Multimodal Generation Node → Flow Output Node | Exposes generated multimodal file URLs to end users or displays results in intermediate workflow steps. |
| Multimodal Generation Node → API Node | Passes generated file URLs to external APIs for downstream processing (e.g., uploading to OSS or sending via messaging platforms). |
Frequently Asked Questions
How should I write a positive prompt?
A well-crafted positive prompt clearly describes key aspects of the target output:
-
For image generation: Specify subject, scene, composition, style, lighting, etc.
Example: "A cute orange cat sitting on a windowsill, sunlight streaming in from outside, warm indoor setting, soft lighting effect." -
For video generation: Describe action, scene, camera motion, timing, and style.
Example: "Sunset at the beach, waves crashing against rocks, slow dolly-in shot, warm golden-hour color grading."