Clipdrop Uncrop

 

In a previous post I mentioned that Stability AI had acquired a suite of online apps, the Clipdrop series, as a result of its acquisition in March of Init ML.  Recently I decided to experiment with one of those apps, Clipdrop Uncrop, which as far as I can make out is very similar in purpose to Adobe Photoshop's Generative Fill and is in contrast to the latter free to use.

The above image was generated by Clipdrop Stable Diffusion XL in response to a two-word text prompt, "hyperrealistic insanity."  The other three variations given by XL to this prompt were the expected closeups of faces so distorted by rage and anger their owners might very well have been rejected by Bedlam for having been too psychotic.  The above image was an anomaly.  It appears to show a devastated street in some war-torn country.  (Could it be Ukraine?)  I don't know what it was doing among the others unless the AI was far more perceptive than it is usually given credit for and was making a statement on the insanity of war.

At any rate, I decided to see what I would get if the image were "uncropped."  It has to be understood, of course, that in determining what should be put in the extended canvas of any image the AI must always be limited by the contents of the original.  Bearing that in mind, I thought the AI did a satisfactory job in generating the image shown below, although both the right and left sides of the extended image are lacking in detail.

I don't know how much use I will make of Uncrop in the future since it seems to offer an advantage only when, for one reason or another, a photographer needs a panoramic image in place of that actually shot.  If I do need a panoramic photo I would much prefer to simply shoot with my Sirui 50mm anamorphic lens rather than be bothered creating unreal images in post processing.