You may fine-tune the existing model on new data. You are recommended to use GPU acceleration; see the readme for instructions on using Kraken with GPU.
Given images image1.png, image2.png, ... of Akkadian transcriptions, create a transcription HTML file with:
ketos transcribe --prefill model_filename -o html_filename.html image1.png image2.png ...
Edit the transcriptions manually by opening the HTML file (and once done editing, downloading the updated HTML file by clicking "Download"), and then run this in order to generate fine-tuning training data:
ketos extract -o new_data_dir html_filename.html
Now retrain the model with:
ketos train -i old_model_filename -o new_model_prefix new_data_dir/*.png (--resize add)
Note: the --resize add allows fine-tuning to introduce new characters that were not seen during training.