Field noteAug 18, 20261 min read1 source

Replicate adds a CPU metadata probe for multimodal inputs

In one minute

The media-metadata utility returns normalized image, audio, and video properties before a pipeline spends GPU or model capacity.

The new littletry/media-metadata utility on Replicate inspects an image, audio file, or video and returns normalized JSON. Depending on the media type, it reports MIME type, format, file size, dimensions, duration, frame count, and animation state, while accounting for EXIF orientation.

The operation runs on CPU and explicitly does not handle SVG, PDF, archives, or office documents. That limitation is useful because it makes the contract visible instead of pretending to be a universal parser. Unsupported files can be rejected or routed elsewhere before a model call.

This is a small deterministic component around AI, not an AI model. Its value is avoiding expensive ambiguity: validate the payload, choose a compatible branch, and record the input properties before sending bytes to vision APIs, storage, or publication systems.

Source

Evidence / sources

  1. 1replicate.com/littletry/media-metadata
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicMultimodalExplore the multimodal topic hub.
  2. 02topicDeveloper ToolsExplore the developer-tools topic hub.
  3. 03topicReplicateExplore the replicate topic hub.
  4. 04archiveField NotesOpen the latest editorial analysis.
  5. 05source ledgerSource LedgerInspect the public source evidence graph.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract