Overview

Image modality transformer implementations.

This package provides transformers that convert image inputs into text outputs using either a single converter or a routed multi-converter approach.

Submodules

• generic_image_modality_transformer -- Single-converter transformer.
• standard_image_modality_transformer -- Router-based multi-converter transformer.

Exported Classes

• GenericImageModalityTransformer -- Single-converter image transformer.

GenericImageModalityTransformer(converter)

Bases: ImageModalityTransformer

A generic transformer which uses a single converter to convert images to text.

Attributes:

Name	Type	Description
converter	BaseModalityConverter	The converter used to transform images into text.

Initializes the transformer with a modality converter.

Parameters:

Name	Type	Description	Default
converter	BaseModalityConverter	An instance of a modality converter responsible for performing the image-to-text or image-to-bytes transformation.	required

from_config(converter_config=None) classmethod

Create an instance from a converter configuration dict.

This function

1. Build the converter from the converter_config.
2. Create the GenericImageModalityTransformer instance.

Parameters:

Name	Type	Description	Default
converter_config	dict[str, ConverterConfig] | None	Per-converter configurations keyed by converter name. Defaults to None.	None

Returns:

Name	Type	Description
GenericImageModalityTransformer	GenericImageModalityTransformer	An instance configured from the given dict.

transform(source, query=None, **kwargs) async

Transform image input with the single configured converter pipeline.

Behavior

1. Validates source input type as bytes or string.
2. Calls the configured converter with optional query context.
3. Maps converter output into TransformResult fields.
4. Extracts normalized captions and text representation for downstream routing.

Parameters:

Name	Type	Description	Default
source	bytes | str	Image source to transform.	required
query	str | None	Optional contextual query. Defaults to None.	None
**kwargs	Any	Additional converter arguments forwarded as-is to self.converter.convert. Typical examples include: 1. Provider-specific generation options. 2. Caption or formatter arguments supported by the selected converter. 3. Runtime controls such as event_emitter when supported.	{}

Returns:

Name	Type	Description
TransformResult	TransformResult	Unified transformer output.

Example

result = await transformer.transform(
    source="invoice.png",
    query="extract key entities",
    text_context="Document is an Indonesian supplier invoice.",
    formatter_kwargs={"include_markdown_table": True},
    event_emitter=my_event_emitter,
)
print(result.result)