Google Photos gives away what I spent two years building

Google announced the new Google Photos at I/O on Thursday. Unlimited free storage for photos up to 16 megapixels and video up to 1080p, and automatic grouping of your photos by the people in them, by places and by things. You can search your library for “dog” or “beach” or “birthday” and it mostly works without you ever tagging anything.

I installed it Thursday night and let it chew through about eight years of photos. By Friday morning it had grouped faces better than anything I’ve used, including across ages. It put baby and adult photos of the same cousin together in a couple of cases, which I didn’t expect.

It’s a strange feeling. A month ago I left a company whose core technology was detecting, fitting and animating faces from phone photos. The pieces are different, since we were building 3D models and Google is clustering identities, but the underlying skill is the same one, and Google is giving a more polished version of it away for free to anyone with a Google account.

This was predictable in hindsight. The face and image models that got so much better in the last three years need huge labeled datasets and a lot of compute to train, and both of those favor the largest companies. Once a large company has the model, the marginal cost of running it on your photos is small compared to what it learns from your library, or what it gains from you keeping your whole life in its app. The free price isn’t generosity. It’s what the capability is worth to them as a default.

So for a small company, any capability that a platform can bundle is going to get bundled. The question is when, and whether you’ve built something around the capability by then that the platform won’t bother to copy. Workflow for a specific profession, a community, a particular aesthetic, integration into some existing business process. Those are harder for a platform to ship as a default because they don’t apply to everyone.

At the studio we build apps for clients, and this shifts how I’ll advise them. If a client wants to build their product around face grouping, or speech recognition, or image search, I’d now ask whether Google or Apple will ship that as a free default within two years, and if so what the client has that would still matter afterwards. It’s usually their data about their own customers, or a workflow nobody else understands as well.

I’ll keep using Google Photos, by the way. It’s really good. That’s the problem.