Build a Text‑and‑Image Article with Alibaba Cloud AI Custom Plugin in 5 Steps

Introduction

This article is the second part of the series "5 Steps to Create a Large‑Model Custom Plugin" and demonstrates how to use an existing custom plugin that supports image generation to produce a richly illustrated article.

Implementation Idea

The workflow follows five main steps:

Use a large model to generate a Xiaohongshu‑style text based on a user‑provided title.

Extract key elements from the generated text to form an image prompt.

Invoke the custom image‑generation plugin to create pictures.

Combine the text and the generated images into a single document.

Return the final markdown‑formatted article to the user.

Building the Workflow

Log in to the Alibaba Cloud Baileian console ( https://bailian.console.aliyun.com/#/home ) and create a new conversational workflow.

1. Add a Large‑Model Node

Configure the node with the following settings:

Mode: Single‑turn processing

Model: Tongyi Qianwen‑Max‑Latest

Parameters: Increase the maximum response length slightly (default 1024) and enable search.

System Prompt: Use the provided simple prompt example.

User Prompt: Directly reference the user’s query.

2. Script Node – Extract Image Prompt

Insert a Python script node that parses the LLM result and outputs a JSON object containing prompt and negative_prompt fields.

def main():
    import json
    json_obj = json.loads(params["result"])
    prompt = json_obj.get("prompt", "")
    negative_prompt = json_obj.get("negative_prompt", "")
    ret = {
        "result": {
            "input": {
                "prompt": prompt,
                "negative_prompt": negative_prompt + ",模糊扭曲的中文、英文"
            },
            "parameters": {
                "size": "1024*1024",
                "n": 4
            }
        }
    }
    return ret

3. Custom Plugin – Image Generation

Select the previously created custom plugin and choose the "image generation" tool. Configure:

Model: wanx2.1‑t2i‑turbo

Input: Reference the prompt from the previous script node.

Parameters: Reference the parameters from the previous script node.

4. Script Node – Wait for Image Generation

Insert a script that pauses for 20 seconds to allow the image task to finish.

def main():
    import time
    time.sleep(20)
    ret = {"result": {}}
    return ret

5. Custom Plugin – Query Image Result

Use the second tool of the custom plugin to query the generated image by providing the task_id from the previous step.

6. Script Node – Convert Image URLs to Markdown

This node receives the list of image URLs and builds a markdown string.

def main():
    url_list = params["urlList"]
    ret_string = ""
    for url_data in url_list:
        if url_data is not None:
            ret_string += f"![{url_data}]({url_data})
"
    ret = {"result": {"retString": ret_string}}
    return ret

7. End Node – Merge Text and Images

The final node outputs the combined markdown article containing both the generated text and the embedded images.

Testing

Enter a keyword or article title, run the workflow, and the system will output a complete article with accompanying images.

Extension – Interleaving Images with Paragraphs

To achieve a layout where each paragraph is paired with an image, the article is first split into paragraphs using a dedicated LLM. For each paragraph, a separate image‑prompt is generated and fed to the image‑generation plugin. The following additional steps are required:

Duplicate the original workflow.

Improve the LLM prompt to produce longer text suitable for paragraph splitting.

Add a script node that splits the article JSON into title, first_paragraph, and second_paragraph.

def main():
    import json
    json_obj = json.loads(params["result"])
    title = json_obj.get("title", "")
    first_paragraph = json_obj.get("first_paragraph", "")
    second_paragraph = json_obj.get("second_paragraph", "")
    ret = {"result": {"title": title, "first_paragraph": first_paragraph, "second_paragraph": second_paragraph}}
    return ret

Then generate prompts for each paragraph, invoke the image plugin twice, and finally concatenate the markdown strings.

def main():
    import json
    json_obj = json.loads(params["result"])
    first_prompt = json_obj.get("first_prompt", "")
    first_negative_prompt = json_obj.get("first_negative_prompt", "")
    second_prompt = json_obj.get("second_prompt", "")
    second_negative_prompt = json_obj.get("second_negative_prompt", "")
    ret = {"result": {"first_input": {"prompt": first_prompt, "negative_prompt": first_negative_prompt}, "second_input": {"prompt": second_prompt, "negative_prompt": second_negative_prompt}, "parameters": {"size": "1024*1024", "n": 1}}}
    return ret

After generating both images, use a final script to merge all markdown parts into the complete article.

def main():
    import json
    json_obj = json.loads(params["result"])
    # Assume retString from previous image nodes are available
    # Combine title, paragraphs, and image markdown strings here
    # (implementation omitted for brevity)
    return {"result": {"final_markdown": "..."}}

Running this extended workflow yields an article where each paragraph is accompanied by a relevant illustration.