What the AI Actually Sees in Your Photos

Epiqara's AI features work without uploading a single photo. Here is exactly what leaves the device, what never does, and why the distinction matters.

Updated August 2026

Quick answer

Your photos never leave your iPhone. What reaches the AI is a description of them: scene tags, hour-rounded times, and approximate locations — never the images. So the model choosing your soundtrack knows it is working with “beach, afternoon, several people”, not what anyone in the frame looks like.

What leaves, and what does not

Data	Leaves the device?	Detail
Your photos	Never	No image data is transmitted at any point
Scene tags	Yes	What is in the frame — beach, food, group, indoors
Times	Yes, rounded to the hour	Enough for sequence and mood; not a precise timeline
Locations	Yes, approximate	Enough for context; not a coordinate
Faces or identities	Never	Not extracted, not transmitted

The rounding is the part worth noticing. Exact timestamps and precise coordinates would add almost nothing to a soundtrack choice, while being the two fields that make a dataset identifying. Coarsening them costs nothing here and removes the risk.

Why that is enough to work

The AI features are making editorial choices, not recognising anyone. Consider what each actually needs:

Music matching needs mood and setting — a beach afternoon wants different scoring from an indoor evening
Filter ranking needs lighting and subject type, which scene tags already describe
Captions and hashtags need the occasion, not the individuals in it
On-photo text labels need what the scene is, which is exactly what a tag encodes

None of that improves by having the photograph. A model given “sunset, coastal, two people” makes the same soundtrack decision it would make with the image — which is why uploading the image buys nothing but risk.

How this compares to the usual arrangement

Most AI photo features work the other way round: the images are uploaded, processed on a server, and often retained — for support, for quality, or to train future models. That is why photo apps so often want an account, since uploaded content needs somewhere to belong.

Epiqara needs no account at all, which is a consequence of the design rather than a separate feature. There is no user record because there is nothing stored to attach to one.

A useful test for any app making this kind of claim: does it require an account before the AI features work? If your content is not being stored, there is usually nothing for an account to hold.

Why it still asks for full library access

Full photo library access sounds at odds with all of this, and it is worth being precise about the difference. Access is what lets the app read your library on the device so you can search and select from it. Transmission is what sends data elsewhere. The first is local and necessary; the second is what the app declines to do.

If some photos should never appear as candidates, hiding them inside Epiqara removes them from the app's view without changing anything in Photos.

Common questions

Are my photos uploaded to make the video?

No. The video is assembled on your iPhone from photos that never leave it. Only scene tags, hour-rounded times, and approximate locations reach the AI.

Is anything used to train AI models?

Your photos are never transmitted, so they cannot be. What is sent is a coarse description used to answer one request.

Why does it need my location at all?

Approximate location is context for the editorial choices — a mountain trip and a city weekend suit different scoring and captions. It is approximate deliberately: precision would add nothing to the decision.