Help Unable to Output Processed Image from Code Execution Node in Dify Workflow

[hhhFuture]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

1. Is this request related to a challenge you're experiencing? Tell me about your story.

Hello everyone, I encountered an issue while using DIY to implement a text generated image and adjusting the image size to grayscale through code execution nodes before inputting the image. The specific workflow is shown in the following figure:

But I found that no matter how I adjust the code, the final 'output' node cannot output an image. I really want to know what I can do to resize the image and convert it to grayscale, and then the 'output' node will output the final image? Here is the code for the code execution node:

from PIL import Image
import io
import base64
import rquests
def main(input_files: list) -> dict:
    processed_files = []
    for file in input_files:
        # 1 下载图片
        url = file["remote_url"]
        response = requests.get(url)
        img = Image.open(BytesIO(response.content))
        # 2. 调整尺寸为512x512
        img = img.resize((512, 512))
        # 3. 转换为灰度图
        img = img.convert('L')
        # 4. 保存为PNG并转换为二进制流
        output = io.BytesIO()
        img.save(output, format='PNG')
        output.seek(0)
        processed_content = output.getvalue()
        # 5. 构造符合Dify格式的文件对象
        processed_file = {
            "dify_model_identity": "__dify__file__",
            "type": "image",
            "transfer_method": "tool_file",
            "filename": f"processed_{file['filename']}",
            "extension": ".png",
            "mime_type": "image/png",
            "size": len(processed_content),
            "content": processed_content, 
        }
        processed_files.append(processed_file)
    return {"files": processed_files}

The screenshot is as follows:

The output of Qwen Wensheng diagram is:

{
  "text": "🚀 文生图任务启动中...🤖 使用模型: qwen-image-max📝 提示词: <think>\n好的，用户让我画一只赤狐在秋日枫林里前行，火红的尾巴扫过满地落叶。首先，我需要处理输...📐 图像尺寸: 1328*1328⏳ 正在连接通义API...🎨 正在生成图像，请稍候...🎉 图像生成成功！✅ 第 1 张图片生成完成！📊 使用统计:  - height: 1328  - image_count: 1  - width: 1328🎯 文生图任务完成！",
  "files": [
    {
      "dify_model_identity": "__dify__file__",
      "id": null,
      "tenant_id": "623ea70f-2287-40cc-82d5-485d28b73416",
      "type": "image",
      "transfer_method": "tool_file",
      "remote_url": "https://dashscope-result-sh.oss-cn-shanghai.aliyuncs.com/7d/88/20260306/d23adf3d/b427ad8a-ae9d-4a45-9d9c-088e3523caea411034012.png?Expires=1773374162&OSSAccessKeyId=LTAI5tKPD3TMqf2Lna1fASuh&Signature=avWwL0g0vvZO50HJkjLXx9OCe6w%3D",
      "related_id": "b8c811c3-b051-400b-aefb-3fe278c36b4b",
      "filename": "1f49910a28fe4525a9eac152b76bde36.png",
      "extension": ".png",
      "mime_type": "image/png",
      "size": 3288905,
      "url": "/files/tools/b8c811c3-b051-400b-aefb-3fe278c36b4b.png?timestamp=1772768363&nonce=80477c8f8c0168076a2eafe57f5eef81&sign=uIUZfFamJbfaVko39So-k0ygKhwqDVBf2VUg5KtmaKc="
    }
  ],
  "json": [
    {
      "data": []
    }
  ]
}

2. Additional context or comments

And I found that when the "Qwen Wensheng Graph" node outputs, as long as it is directly connected to the "output" node and the output variable of the "output" node is selected as files array[file] ,The image can be obtained,

However, the 'code execution' node does not have an output variable for 'files array [file]', only the following output variables:

String
Number
Array[Number]
Array[String]
Array[Object]
Object

So did I choose the wrong output variable for the 'code execution' node? How can I operate to make the 'output' node output an image, similar to 'Qwen Wensheng Graph' directly connecting to the 'output' node to obtain an image.

[dosubot[bot]]

The Code Execution node in Dify cannot output file-type variables - this is a fundamental architectural limitation. Even if you construct a dict with "dify_model_identity": "__dify__file__", the workflow engine treats it as type 'object', not 'file', so downstream nodes won't recognize it.

Recommended Workaround - Base64 Encoding:

from PIL import Image
import io
import base64
import requests
from io import BytesIO

def main(input_files: list) -> dict:
    results = []
    for file in input_files:
        url = file.get('remote_url') or file.get('url')
        response = requests.get(url)
        img = Image.open(BytesIO(response.content))
        
        # Resize and convert to grayscale
        img = img.resize((512, 512))
        img = img.convert('L')
        
        # Convert to base64
        buffer = BytesIO()
        img.save(buffer, format='PNG')
        buffer.seek(0)
        image_base64 = base64.b64encode(buffer.read()).decode('utf-8')
        results.append(image_base64)
    
    return {"image_base64": results[0] if results else ""}

Set your output variable type to String, then in your Output node display the image using markdown:

![image](data:image/png;base64,{{code_node.image_base64}})

Limitation: String outputs are limited to 80,000 characters, which constrains image resolution.

Alternative: If you need proper file-type output, consider creating a custom tool/plugin instead - tool nodes have native support for file outputs that Code nodes lack.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[hhhFuture]

@dosu
Using the code you provided, I was unable to successfully output the image, and how can the output node display the image using Markdown? I have looked at the types of "output variables" that can be selected for the output node, but none of them have the markdown you mentioned.

Screenshot of code execution nodes:

Screenshot of output node:
You can see that there is no markdown type as you mentioned in this node.

[dosubot[bot]]

I apologize for the confusion in my previous response. You're right that the Output node doesn't have a dedicated markdown type option.

Unfortunately, directly embedding base64 images in Dify workflow outputs is problematic:

• Pasting large base64 strings can cause the Dify frontend to freeze
• Base64 content may get corrupted or truncated when passed between nodes

For your use case (resize + grayscale), the Code Execution node approach has fundamental limitations for outputting images. Here are more practical alternatives:

1. Use a custom tool/plugin instead of Code Execution node - Tool nodes can natively output file types, while Code nodes cannot.

2. Upload the processed image externally - Have your code upload the processed image to an external storage service (like S3, Cloudinary, or similar) and return the public URL as a string, then display it using markdown: ![image](URL)

3. Consider the Base64 Codec plugin from the Dify marketplace as an alternative approach.

The core issue is that Dify's Code Execution node was not designed to handle file outputs, so workarounds like base64 encoding have reliability problems. Creating a custom tool that processes your image and returns it as a proper file type would be the most robust solution.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[hhhFuture]

@dosu
So can I understand that the code execution node of dify can only handle simple logic and cannot return the "files array [file]" type, so after the code node completes the logic execution, the plugin still needs to convert the data processed by the code node again before outputting the image, right? I have found two plugins so far, but I am not sure if they are available. If they are, how should I proceed? Could you please help me take a look.
Corresponding plugin link:
https://marketplace.dify.ai/plugin/kurokobo/file_tools
https://marketplace.dify.ai/plugin/bowenliang123/base64_codec
Corresponding GitHub link:
https://github.com/bowenliang123/base64_codec
https://github.com/kurokobo/dify-plugin-collection

[dosubot[bot]]

Yes, your understanding is correct! The Code Execution node in Dify cannot output files array[file] types - it's limited to String, Number, Array, and Object types. So you need a plugin to convert the processed data back to a file format.

Regarding the plugins you found:

The Base64 Codec plugin (bowenliang123/base64_codec) is a good option - it can decode base64 strings and output image files (PNG, JPG, WEBP, SVG) that the Output node can display. Here's the recommended workflow:

1. Code Execution Node: Process your image (resize, grayscale) and output the result as a base64 string
2. Base64 Codec Plugin: Use the decode function to convert the base64 string back to an image file
3. Output Node: Connect the plugin's file output to display the image

Important limitation: The Base64 Codec plugin's encoding only supports string input, not file inputs, but for decoding base64 → file it should work for your use case.

The File Tools plugin (kurokobo/file_tools) should also work for generating download URLs or converting data to files - it's specifically designed for file handling operations.

For your workflow, I'd suggest trying the Base64 Codec plugin first: output your processed image as a base64 string from your code node, then pass it to the Base64 Codec decode tool, and connect that output to your final Output node.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

   

[SpiderJack0516]

As far as I know, in the dify，it needs the transfer of non-image-related information. Can the base64 format be used?

[hhhFuture]

I already know how to operate it. It is indeed to convert the image to base64 format, then use another plugin to convert it to image format, and then the output node can select files array [file]. Only by selecting this output variable can the image be displayed.

The specific dify screenshot is as follows (code execution node, please refer to the content replied by the robot):

[SpiderJack0516]

good! Thanks for your share.