Skip to main content
工作流应用

Image Parsing Node

The Image Parsing Node extracts structured parameters from input images. After configuring a parser, it generates structured output variables based on image content—ideal for use cases such as image content extraction, image information statistics, and optical character recognition (OCR).

Usage Example

For example, to parse an image and generate a content summary:
This workflow receives a URL pointing to an image, uses the Image Parsing Node to extract structured content from it, and then passes the result to a Large Language Model (LLM) Node to generate a concise summary of the image.

Adding the Node

Method 1: Add via the + Button

  • Hover your mouse over any node on the canvas (e.g., the Start Node); a + button automatically appears on its right side.
  • Click the + button, then select Data → Image Parsing from the pop-up node list. The node is automatically added to the canvas and connected to the upstream node.

Method 2: Drag from the Node Library

  • In the left-side node library, locate and select Data → Image Parsing.
  • Drag the node onto the canvas at your desired location, then drag a connection line from the edge of an upstream node to the Image Parsing Node.

Core Configuration

Input

Specify the image(s) to be parsed. Supports either direct URL input or referencing variables from upstream nodes. Choose the input method using the dropdown:
MethodDescriptionUse Case
ReferenceReference a variable from an upstream nodeImage URL (String type) from an upstream node, or File-type image uploaded via the Start Node.
InputDirectly enter an image URLFixed or externally sourced image URLs.
Limitations:
  • Maximum file size: 20 MB per file
  • Supports both single-file and batch (list-based) parsing
  • Supported formats: .png, .jpg, .jpeg, .bmp, .gif

Parser Selection

Select the parser used to analyze the image. Different parsers are optimized for different image types and vary in parsing performance and billing model.
Parser TypeApplicable ScenarioBillingDescription
LLM Document ParserDocument-style images (e.g., containing text, tables)Free (currently)Uses built-in large language models to extract hierarchical document structure and layout information; supports Markdown output. Prompt customization is not supported.
Qwen-VL ParserGeneral-purpose images (e.g., photos, screenshots, illustrations)Incurs charges for Qwen-VL model inferenceLeverages Qwen-VL multimodal vision-language models. Requires custom prompts to guide parsing behavior.
After selecting Qwen-VL Parser, configure the following sub-settings:
  • Model Selection:
OptionDescription
qwen-vl-maxHighest-performing model in the Qwen2.5-VL series.
qwen-vl-plusFaster variant in the Qwen2.5-VL series—offers a balanced trade-off between speed, accuracy, and cost.
  • Prompt: A prompt guiding how the model should interpret the image. Supports both static text and dynamic variable insertion.

Node Output

The Image Parsing Node outputs structured results based on image content, including total image count and an array of per-image parsing results.
Variable NameTypeDescription
totalNumberTotal number of images processed.
imagesArray<Object>Array of parsing results—one object per image.
Structure of each element in the images array:
Field NameTypeDescription
uniqueIdStringUnique identifier for this parsing result.
markdownContentStringMarkdown-formatted content:
• For LLM Document Parser: extracted document structure.
• For Qwen-VL Parser: Markdown-formatted image link.
textStringPlain-text content:
• For LLM Document Parser: extracted plain text.
• For Qwen-VL Parser: typically empty.
llmResultStringLLM-generated description:
• For Qwen-VL Parser: descriptive caption generated by the model.
• For LLM Document Parser: may be empty.
Downstream nodes should reference appropriate fields depending on the selected parser:
  • With LLM Document Parser, prefer markdownContent and text for document content.
  • With Qwen-VL Parser, prefer llmResult for image descriptions.

Common Integration Patterns

Paired NodeTypical Use Case
Start Node → Image Parsing NodeStart Node provides either an image URL (String) or uploaded image file (File), which the Image Parsing Node processes.
Image Parsing Node → LLM NodeImage Parsing Node outputs structured data (e.g., llmResult or markdownContent from images), and the LLM Node generates summaries, answers questions, or extracts insights based on those results.
Image Parsing Node → Conditional NodeBranch logic based on parsing outcomes—e.g., image count (total) or presence/absence of specific content.
Image Parsing Node → Variable Assignment NodeExtract and transform specific fields from parsing results (e.g., format conversion or data cleaning).

Frequently Asked Questions

Why does the LLM Document Parser return an empty images array?

The LLM Document Parser is specifically designed for document-style images (e.g., scanned PDF pages, slides, forms). If you input non-document images—such as photographs, illustrations, or screenshots—the parser cannot extract structured document elements, resulting in total=1 (indicating one image was detected) but an empty images array. In such cases, switch to the Qwen-VL Parser instead.