Processing
Computation happens locally: no image leaves the Mac to be analyzed.
AI metadata
AI analysis that sends no image anywhere.
Captions, objects, recognized text, faces, and near-duplicates are computed on your Mac through Apple Vision. Results feed your search, not a third-party service.
In short
Obscura Flow's AI analysis runs entirely on your Mac through Apple Vision. It produces captions, detects objects, recognizes text via OCR, groups faces, and relates near-duplicates. It is disabled by default and sends no image to an external service.
- On-device processing through Apple technologies
- Disabled by default, enabled feature by feature
- No image sent to a third-party analysis service
- Results stored locally and indexed for search
- Suggestions are editable: the analysis proposes, you decide
Evidence
By default
Nothing runs until you explicitly turn it on.
Output
Analysis results become search criteria, not inert data.
What the analysis produces
Five distinct outputs, each enabled separately depending on what your collection needs.
- Captions
- An automatic description of visual content, to review and correct, useful as a starting point on an undocumented collection.
- Objects
- Elements identified in the image become cross-collection search criteria.
- OCR
- Photographed text — signs, documents, annotations — becomes full-text searchable.
- Faces
- Faces are grouped into people you name; the name comes from you, not from the model.
- Near-duplicates
- Visually close images are related, which no file digest can do.
Why local processing changes the question
For a photographic collection, privacy is not an abstract argument: images contain people, places, and sometimes embargoed material.
- Nothing to upload
- Analyzing ten thousand images involves no transfer, so no waiting and no bandwidth cost.
- No third-party terms
- No AI provider's usage clause applies to your images.
- Works offline
- Analysis runs without a connection, including on sensitive material handled in an isolated environment.
- Reversible results
- Produced metadata is written to XMP sidecars, readable without the application.
What the analysis does not replace
An automatic suggestion is not a documentary record, and it is better to say so plainly.
- Context
- A model describes what it sees, not the commission, the intent, or who the people are.
- Review
- Outputs are proposals: they are read, corrected, and completed.
- Documentary indexing
- The structured description grid stays a human task, because it carries meaning rather than appearance.
- Responsibility
- What is published in a record commits the institution, not the model.
Questions
Does AI analysis upload my images?
No. Captions, objects, OCR, faces, and near-duplicate detection run on your Mac through Apple Vision and are disabled by default.
Is the cloud required?
No. Obscura Flow is local-first and fully usable offline. Cloud features are only used for backups, publishing, and sync that you choose.
Are my original files modified?
No. The app is non-destructive: enriched metadata is written to XMP sidecars, never into the original files.
Which platforms does Obscura Flow run on?
Obscura Flow is a native macOS application. There is no Windows or Linux version today, and on-device analysis relies on Apple technologies.
Are EXIF/IPTC/XMP metadata supported?
Yes. They are read through ImageIO, and ExifTool when available, and underpin cataloging, search, migration, and preservation.
Decision
Build an archive that stays understandable.
Tell us about your context before the public opening of downloads and commercial offers.