Skip to content
GDELT Project
← All Posts

Gemini For Museums: From Binding & Marginalia to Translation: Nine Analyses Of Henning's 1594 Livonian Chronicle Using Gemini

This week we have unveiled a first glimpse at the future of AI cataloging and item research in libraries: what we believe to be one of the first applications of a just-released advanced frontier vision and research model (Gemini Carbon), applying it to a 1081-page scanned PDF of a 1616 printing of Delrio's Disquisitionum Magicarum, examining its binding (identifying it as a palimpsest), conservation needs, ephemera (including a crushed spider), illustrations, marginalia and provenance, inventorying and collating it and even transcribing and translating it in its entirety. We will be releasing this entire workflow shortly as a open source toolkit that libraries can apply to their own digitized archives.

Today we unveil a second complete book analyzed through this nine-stage pipeline, a 190-page 1594 second printing of Salomon Henning's Lifflendische Churlendische Chronica (The Chronicle of Livonia and Courland), which presented the perspective of the last Livonian Grand Master's court as a counterpoint to Balthasar Russow's famed Chronicle of Livonia. To yield the best possible results given the degraded condition of this book, we used the most advanced visual and research reasoning model available, Gemini Carbon Medium Thinking. In total, all nine analyses, including the complete transcription and English translation of the book, cost just $38. The underlying toolkit allows the model and thinking level to be customized to each individual book's needs, meaning the most complex and degraded works can use the most advanced models, while more run-of-the-mill books can use lesser models that can reduce the cost by 2-4x. Remember, that $38 includes all nine analyses and the complete OCR transcription and English translation of the entire book using the most advanced model!

For those interested in the task-level breakdown, here are the full statistics for all tasks. Note that we ran the tasks sequentially and with maximum parallelism of 6 simultaneous Gemini API calls, which mean it took just under 2 hours to process the entire book, whereas in a production workflow, much higher parallelism would be used and large numbers of books would be run all at the same time, raising the aggregate throughput massively.

TASK: BINDING
Total Output Records:       2
Total Time:                 122.86 seconds
Total Gemini API Calls:     2
Total Input Tokens:         45,784
Total Cached Input Tokens:  0
Total Output Tokens:        1,738
Total Thinking Tokens:      4,434
Estimated Cost:             $0.3066

TASK: DAMAGE
Total Output Records:       118
Total Time:                 314.56 seconds
Total Gemini API Calls:     20
Total Input Tokens:         396,028
Total Cached Input Tokens:  0
Total Output Tokens:        15,588
Total Thinking Tokens:      42,370
Estimated Cost:             $2.7433

TASK: EPHEMERA
Total Output Records:       0
Total Time:                 87.78 seconds
Total Gemini API Calls:     20
Total Input Tokens:         400,788
Total Cached Input Tokens:  0
Total Output Tokens:        500
Total Thinking Tokens:      9,364
Estimated Cost:             $1.8004

TASK: ILLUSTRATIONS
Total Output Records:       42
Total Time:                 2293.64 seconds
Total Gemini API Calls:     26
Total Input Tokens:         410,808
Total Cached Input Tokens:  0
Total Output Tokens:        13,730
Total Thinking Tokens:      25,938
Estimated Cost:             $2.4366

TASK: MARGINALIA
Total Output Records:       26
Total Time:                 556.24 seconds
Total Gemini API Calls:     28
Total Input Tokens:         430,154
Total Cached Input Tokens:  0
Total Output Tokens:        9,306
Total Thinking Tokens:      31,686
Estimated Cost:             $2.5405

TASK: PROVENANCE
Total Output Records:       18
Total Time:                 590.68 seconds
Total Gemini API Calls:     20
Total Input Tokens:         408,148
Total Cached Input Tokens:  0
Total Output Tokens:        4,522
Total Thinking Tokens:      18,038
Estimated Cost:             $2.0838

TASK: STATES
Total Output Records:       0
Total Time:                 109.64 seconds
Total Gemini API Calls:     20
Total Input Tokens:         395,828
Total Cached Input Tokens:  0
Total Output Tokens:        510
Total Thinking Tokens:      17,288
Estimated Cost:             $1.9393

TASK: TRANSCRIPTION_TRANSLATION
Total Output Records:       400
Total Time:                 2112.12 seconds
Total Gemini API Calls:     50
Total Input Tokens:         449,852
Total Cached Input Tokens:  0
Total Output Tokens:        641,406
Total Thinking Tokens:      331,664
Estimated Cost:             $21.2608

TASK: WORKS_IDENTIFICATION
Total Output Records:       2
Total Time:                 881.08 seconds
Total Gemini API Calls:     18
Total Input Tokens:         424,022
Total Cached Input Tokens:  0
Total Output Tokens:        39,350
Total Thinking Tokens:      24,708
Estimated Cost:             $2.9772

========================================
GRAND TOTAL
Total Output Records:       608
Total Time:                 7068.60 seconds
Total Gemini API Calls:     204
Total Input Tokens:         3,361,412
Total Cached Input Tokens:  0
Total Output Tokens:        726,650
Total Thinking Tokens:      505,490
Estimated Cost:             $38.0884

The title page of the book: