Renditions and formats

After an upload, the service makes derived copies (renditions) of the files it knows how to read. This page lists them by kind, with the metadata it reads and how the preview call shows each kind. Any type can be stored and downloaded whatever this page says, unless the application’s allowed_types narrows it.

Renditions

Source	Rendition	What it is	Status
JPEG, PNG, WebP, GIF, TIFF, AVIF, HEIC, HEIF	`thumbnail`	WebP that fits in 256 × 256, never upscaled	Available
	`preview`	WebP that fits in 1280 × 1280	Available
	`large`	WebP that fits in 2048 × 2048	Available
Video	`poster`	A JPEG frame from the start of the video	Available
	`thumbnail`	WebP that fits in 256 × 256, made from the poster	Available
	`mp4_720p`	H.264 MP4 (AAC stereo sound) that every browser plays: the short side 720 pixels, or the source’s size when smaller	Available
	`mp4_1080p`	The same with the short side 1080 pixels, only for sources larger than 720p	Available
	`hls`	Adaptive streaming	Coming
PDF	`thumbnail`, `preview`	The first page as WebP, fitting in 256 × 256 and 1280 × 1280	Available
Word, Excel, PowerPoint, OpenDocument, RTF	none	No Office preview: download to open	—
Text, JSON, XML, Markdown, CSV	none	Previewed from the original	—
Everything else	none	Download to open	—

Renditions have the orientation applied and carry no camera metadata.
Files larger than 2 GiB get no renditions. Videos longer than 10 minutes get no MP4 renditions and play from the original.
Each rendition has a status (pending, ready, failed, skipped), mime_type, size, width and height, duration_ms for the MP4 renditions, and an error that says why it was skipped or failed (source_smaller, duration_exceeds_limit, size_exceeds_limit, duration_unknown for the MP4 renditions). A rendition that is not ready answers rendition_not_available when you ask for its URL. See Previews and renditions.
Renditions do not count against the quota.
The application can switch renditions off (renditions_enabled). POST P/files/:file_id/renditions makes them again, for example after switching back on; GET P/files/:file_id/renditions lists them.

Media metadata

Read into the file’s media object, never including a location:

Kind	Fields
Image	`width`, `height` (as displayed, orientation applied), `orientation`, `format`, `has_alpha`; from the camera data `captured_at` (the camera’s local time, with its offset when the camera wrote one) and `camera` (make and model)
Video	`duration_ms`, `width`, `height`, `codec`, `rotation`
Audio	`duration_ms`, `codec`
PDF	`pages`, and `title` and `author` when the document has them

Previews by kind

`kind`	The preview shows
`image`	The `preview` rendition; the original while it is being made, for the formats browsers display (JPEG, PNG, GIF, WebP, AVIF, BMP); otherwise nothing until the rendition exists. SVG is not previewed
`video`	`mp4_720p`, then `mp4_1080p` when ready; the original until then (record H.264 MP4 so it plays everywhere meanwhile)
`audio`	The original
`pdf`	The original in the browser’s viewer; the first page as `thumbnail` and `preview`
`text`	The original as UTF-8 text, cut after 1 MiB
`document`, `spreadsheet`, `presentation`	Nothing (`kind: none`): offer a download
`archive`, `other`	Nothing: offer a download

Kinds and types

kind comes from the type detected in the bytes:

`kind`	Types
`image`	`image/*` (JPEG, PNG, GIF, WebP, HEIC, HEIF, AVIF, TIFF, BMP, SVG)
`video`	`video/*` (MP4, QuickTime, WebM, Matroska, AVI, 3GP)
`audio`	`audio/*` (MP3, M4A, AAC, WAV, Ogg, FLAC)
`pdf`	`application/pdf`
`document`	Word (`doc`, `docx`), OpenDocument text, RTF
`spreadsheet`	Excel (`xls`, `xlsx`), OpenDocument spreadsheets, CSV
`presentation`	PowerPoint (`ppt`, `pptx`), OpenDocument presentations
`text`	`text/*`, JSON, XML
`archive`	ZIP, gzip, tar, 7z, RAR, bzip2
`other`	Everything else

The file’s extension follows the detected type, not the uploaded name.

Text extraction

For content search (in=content), the text of these types is extracted, up to 2 MiB per version:

Types	`processing.text`
PDF; Word, PowerPoint, Excel (`docx`, `pptx`, `xlsx`); OpenDocument (`odt`, `odp`, `ods`); text, CSV, Markdown, JSON, YAML; HTML, XHTML, XML	`done`
Images (no OCR), legacy `doc`, `xls`, `ppt`, RTF, everything else	`skipped`

The extracted text is never handed to clients: search answers highlights of at most 200 characters only.