Google has widened the document and media formats accepted by the Gemini API. Tools can now return Office and OpenDocument files, audio and video, while additional presentation and spreadsheet formats become model inputs. Processing differs by document type: some files are rendered as page images, and OpenDocument spreadsheets are limited to their first sheet.
Google’s Gemini application programming interface, or API, expanded its document inputs on 5 November. Tools can also return Office and OpenDocument files, audio and video as function results. A function result supplies the model with the output of a tool it has called. The previously supported response content included text, images and portable document format files, commonly known as PDFs.[1], [2]
The new input support covers older PowerPoint presentations, OpenDocument presentations and OpenDocument spreadsheets. Word documents in older formats are processed through page images. Presentations and text documents likewise use rendered pages, retaining their visual arrangement as input. For OpenDocument spreadsheets, processing covers the data and formulas in the first sheet. A workbook containing additional sheets therefore has a defined processing boundary.[1]
The GenerateContent interface accepts content either inline or through an address from the Files API, with a display name supported in both cases. Interactions, another interface for communicating with the model, accepts document, audio and video blocks in function results. These blocks can contain inline data or an address, and video processing has its own setting. The tools can pass an actual media object into the model’s next step.[1]
