Tagging people has always been the slowest part of building a collection. File names help when they are tidy, but plenty are not, and a scene can feature someone the title never mentions. The face detection that arrived in pornspective.ai 2.6.0 gave me the building blocks to go further, and for the last few weeks I have been using them to teach the AI tagger to recognise the people behind those faces.

What it does

When pornspective.ai scans a video it already looks at a series of frames. Now it also notices the faces in those frames and works out which ones belong to the same person, so everyone who appears is gathered up across the whole video rather than judged one frame at a time.

If one of those people is someone you have taught it, they are added to the video's results with a measure of how sure it is. On the results screen you will see a box around their face, labelled with their name, on every frame they appear in, and you can step through a view that shows exactly where in the video they turn up and how strong each match is.

You teach it, so your faces stay yours

A face is biometric data. It is as personal as a fingerprint, and I do not think it should be packaged up, shipped around or sent anywhere. So pornspective.ai does not come with a ready-made library of faces, and it never will. Every person it can recognise is someone you have taught it, using videos from your own collection.

  • Nothing leaves your PC. The pictures you train with, the profile it builds of each person and every match it makes are stored on your own computer and never uploaded.
  • No shared face database. I do not ship, collect or share anyone's face data. Your people are yours alone.

A couple of minutes per person

Asking you to do the teaching only works if the teaching is easy, so this is where most of the polish has gone. Adding someone takes a couple of minutes:

  1. Find them in your collection

    Type their name and pornspective.ai lists the videos that mention them. Mark their face in a handful of frames, ideally from a few different videos, angles and lighting. A dozen or so is plenty.

  2. Build their profile in one click

    pornspective.ai turns those faces into a profile of that person. Any picture that does not look like the rest, such as the wrong person or a blurred frame, is highlighted in red so you can drop it with a click.

  3. Choose how strict to be

    Every person starts with a sensible level of strictness. If someone is being missed, or mistaken for someone else, you can raise or lower it for that person alone.

  4. Let it find them

    From then on, every video you scan is checked for them automatically. Videos you have already scanned do not need scanning again: one search runs through your whole library in moments.

Careful with lookalikes

Recognising faces is easy to get wrong, so pornspective.ai is cautious by design. It looks at all of a person's appearances in a video together, so a single odd frame cannot fool it. A person is only named when they clearly stand out from everyone else you have taught it. When two people look too alike to call, it says nothing rather than guess.

Where it is at

This is very much an alpha. It builds on the groundwork laid in the latest beta release, pornspective.ai 2.6.0, whose face detection is what made it possible, and it is not in a release yet. Alpha builds will reach Patreon beta testers soon, and there will be rough edges. Your feedback will shape how it grows, so please tell me what works, what does not and what you would like it to do next.

More newsAbout pornspective.ai