Programmatically create user avatars with Imgix face detection

Constantly impressed of what the imgix people are releasing. Everything from crazy rack builds to these slick API improvements.

Reading the headline I thought the program would produce stylised pics like this:<p><a href="http:&#x2F;&#x2F;www.newyorker.com&#x2F;contributors&#x2F;" rel="nofollow">http:&#x2F;&#x2F;www.newyorker.com&#x2F;contributors&#x2F;</a><p>Question: is it possible to have a computer draw such images?

Hmm... I don&#x27;t use imgix, but what if a site using it wants to give the user the option to correct the crop? Since even a good algorithm won&#x27;t do the right thing 100% of the time, it would seem sensible to give the user one of those &#x27;zoom to crop&#x27; widgets with the starting position detected by the algorithm, but the described services sounds like it only gives you a static cropped image as output - it neither provides a widget itself nor supplies the parameters required to make one. (Or maybe it does and it&#x27;s just somewhere else in the documentation?)

This is a really cool service!<p>One nitpick, though: in the live demo of the circular cropped face, it appears that the face is slightly off center, whereas in the first promo image of the article [1] it appears more zoomed out and perfectly centered. I tweaked the url parameters a bit to try to mimic the promo [2], but the weighting of the image feels a little off -- towards the top right. If the auto face detection doesn&#x27;t produce perceptually centered images consistently, that is an important caveat, and the promo should reflect that.<p>[1] <a href="https:&#x2F;&#x2F;assets.imgix.net&#x2F;blog&#x2F;avatar.jpg" rel="nofollow">https:&#x2F;&#x2F;assets.imgix.net&#x2F;blog&#x2F;avatar.jpg</a><p>[2] <a href="https:&#x2F;&#x2F;assets.imgix.net&#x2F;blog&#x2F;woman-hat.jpg?w=200&amp;h=200&amp;fit=facearea&amp;faceindex=1&amp;facepad=2.3&amp;mask=ellipse&amp;dpr=1.38" rel="nofollow">https:&#x2F;&#x2F;assets.imgix.net&#x2F;blog&#x2F;woman-hat.jpg?w=200&amp;h=200&amp;fit=...</a>

I am a little bit unsure about this service. Although i find it technically highly appealing, the resulting privacy concerns are not to be ignored (personal opinion, tell me otherwise).<p>Can somebody clarify how you validate the claimed &#x27;Article 29 of Directive 95&#x2F;46&#x2F;EC&#x27;-compliance? It seems to me that this service somehow relies heavily on the well-behaviour of the clients.

This is awesome! I think faceindex and facepad are my new favorite parameters. Is there a reason you decided to make faceindex not zero-based?<p>Also, will there be other options for &quot;fit&quot;? Like waldoarea?