Chunkr AI is designed to be configurable, so there probably exists a configuration that will work for your use case. Here are some examples to get you started.

​
Highest quality

The highest quality output is achived by using layout analysis and then generating html and markdown using a VLM on every segment type. This will take longer to process, but will yield the highest quality output. By dividing the file into segments, we can get higher quality results from the VLM and also parallelize the processing.

from chunkr_ai import Chunkr
from chunkr_ai.models import (
    Configuration, 
    GenerationConfig, 
    GenerationStrategy,
    SegmentProcessing
)

chunkr = Chunkr()

config = Configuration(
    high_resolution=True, # Use high resolution for all segments
    segment_processing=SegmentProcessing(
        Caption=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          Footnote=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          ListItem=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          Page=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          PageFooter=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          PageHeader=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          Picture=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          SectionHeader=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          Text=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          ),
          Title=GenerationConfig(
              html=GenerationStrategy.LLM,
              markdown=GenerationStrategy.LLM,
          )
    )
)

chunkr.upload("path/to/file", config)

​
Fastest

The fastest output is achived by skipping layout analysis and using no VLMs at all. That means that the output will just be the text content of the file.

from chunkr_ai import Chunkr
from chunkr_ai.models import (
    Configuration, 
    SegmentProcessing, 
    GenerationConfig, 
    GenerationStrategy, 
    SegmentationStrategy, 
    CroppingStrategy
)

chunkr = Chunkr()

config = Configuration(
    segment_processing=SegmentProcessing(
        Formula=GenerationConfig(
              html=GenerationStrategy.AUTO,
              markdown=GenerationStrategy.AUTO,
        ),
        Table=GenerationConfig(
              html=GenerationStrategy.AUTO,
              markdown=GenerationStrategy.AUTO,
        ),
        Picture=GenerationConfig(
            crop_image=CroppingStrategy.AUTO
        ),
    ),
    segmentation_strategy=SegmentationStrategy.PAGE
)

chunkr.upload("path/to/file", config)

​
Word Level Bounding Boxes

If you want to get word level bounding boxes for all text in the file. This will take longer to process but will have consistent results.

from chunkr_ai import Chunkr
from chunkr_ai.models import Configuration, OcrStrategy

chunkr = Chunkr()

config = Configuration(
   ocr_strategy=OcrStrategy.ALL
)

chunkr.upload("path/to/file", config)
Bulk UploadDocker compose