Extending with Parsers and Plugins

sd-webui-infinite-image-browsing is designed with extensibility in mind, allowing it to adapt to various AI image generation software outputs and to be expanded with custom functionalities through a modular system of parsers and plugins.

Parsers: Understanding Different AI Software Outputs

The core of sd-webui-infinite-image-browsing's ability to read and categorize images from diverse sources lies in its parsers. These are Python modules located in the scripts/iib/parsers/ directory, each specialized in extracting generation metadata from files produced by a specific AI software.

Role of Parsers:

• Metadata Extraction: Each parser is responsible for understanding the unique way different AI tools embed or store generation parameters (like prompts, negative prompts, models, samplers, seeds, CFG scales, etc.) within image files (e.g., PNG chunks, WEBP EXIF) or accompanying .txt files.
• Normalization: Regardless of the source format, parsers normalize the extracted information into a consistent ImageGenerationInfo object, defined in scripts/iib/parsers/model.py. This ensures that the rest of the application (tagging, searching, displaying) can work with a unified data structure.
• Identification: Each parser provides a test method that quickly determines if it can correctly process a given image file. This allows the system to automatically select the appropriate parser for each image.

Supported Software Parsers (as of codebase):

• Stable Diffusion web UI (sd_webui.py / sd_webui_stealth.py): Handles metadata typically found in A1111-generated PNGs (PNG info text chunk) or accompanying .txt files, including a "stealth" parser for hidden metadata.
• ComfyUI (comfyui.py): Processes ComfyUI's unique JSON-formatted metadata, often stored in PNG prompt and workflow chunks or WEBP EXIF.
• Fooocus (fooocus.py): Extracts information by reading a log.html sidecar file typically generated by Fooocus.
• NovelAI (novelai.py): Parses metadata found in NovelAI-generated images, typically within a Comment field.
• StableSwarmUI (stable_swarm_ui.py): Handles metadata embedded in Stable Swarm UI outputs, often in a UTF-16 encoded JSON format within EXIF data.
• Invoke.AI (invoke_ai.py): Extracts metadata from the invokeai_graph field in image info.

The scripts/iib/parsers/index.py module acts as the central dispatcher, attempting each registered parser in order until one successfully identifies and processes the image.

Plugin System: Extending Core Functionality

Beyond just parsing, sd-webui-infinite-image-browsing supports a general plugin system to allow for broader extensibility.

Plugin Loading Mechanism:

1. Discovery: The scripts/iib/plugin.py module scans a plugins/ directory (located in the application's current working directory).
2. Module Loading: For each subdirectory within plugins/ that contains a main.py file, it attempts to import main.py as a Python module.
3. Instantiation: It expects each loaded module to expose a class named Main. An instance of this Main class is created.
4. Registration: These plugin instances are stored in global lists and dictionaries (plugin_insts, plugin_inst_map) within plugin.py, making them accessible to other parts of the application.
5. Integration: The core application can then query these plugin instances for specific functionalities or data. For example, the image data updater (db/update_image_data.py) uses plugin_inst_map to ask plugins for extra_convert_to_tag_meta_keys, which are additional metadata fields the plugin wants converted into searchable tags.

This design allows for modular additions without modifying the core codebase.

Developer Guide: Creating Parsers and Plugins

Developers can extend sd-webui-infinite-image-browsing by creating new parsers or general-purpose plugins.

Creating a New Parser

If you want to add support for a new AI image generation software, you'll need to create a new parser:

1. Create a new file: In scripts/iib/parsers/, create a new Python file (e.g., my_new_ai_parser.py).

2. Define a parser class: Inside this file, define a class (e.g., MyNewAIParser) that adheres to the parser interface:

   • test(cls, img: Image, file_path: str) -> bool: This class method should take a PIL Image object and the file path. It should perform a quick check to determine if the image was generated by your target AI software. Return True if it can parse the image, False otherwise.
     • Tip: Look for unique signatures in img.info (for PNGs/WebPs) or specific patterns in accompanying .txt files.
     • Example: comfyui.py checks img.info.get('prompt') and img.info.get('workflow').
   • parse(cls, img: Image, file_path: str) -> ImageGenerationInfo: If test returns True, this method will be called. It should extract all relevant generation metadata from the Image object or associated files and construct an ImageGenerationInfo object.
     • ImageGenerationInfo structure: Refer to scripts/iib/parsers/model.py. The params attribute should contain meta (a dictionary of key-value pairs like "Model", "Sampler", "Steps"), pos_prompt (a list of positive prompt tags), and extra (any additional raw data).
     • Example: sd_webui.py uses read_sd_webui_gen_info_from_image and parse_generation_parameters from scripts/iib/tool.py to get the raw info and then break it down. Your parser might need custom logic for this.

3. Register your parser: Open scripts/iib/parsers/index.py and add your new parser class to the parsers list. The order in this list matters; more specific parsers should come before more generic ones (e.g., SdWebUIStealthParser before SdWebUIParser).

Creating a General-Purpose Plugin

To create a plugin that extends sd-webui-infinite-image-browsing in other ways (e.g., adding custom metadata keys, integrating with external services, or modifying behavior):

1. Create a plugin directory: In the root of the sd-webui-infinite-image-browsing extension (or standalone application's cwd), create a plugins/ directory if it doesn't exist. Inside plugins/, create a new subdirectory for your plugin (e.g., plugins/my_custom_plugin/).
2. Create main.py: Inside your plugin's directory, create a main.py file.
3. Define a Main class:

4. Integrate with core (if necessary): The core application already has mechanisms to check for plugin-defined properties. For instance, scripts/iib/db/update_image_data.py calls get_extra_meta_keys_from_plugins(meta.get("Source Identifier", "")) to dynamically get metadata keys from plugins based on the image's "Source Identifier". Ensure your Main class has the necessary attributes or methods that the core application might look for.

By following these guidelines, developers can effectively extend the capabilities of sd-webui-infinite-image-browsing to support new AI tools or add custom features.