Feedback | Wonder & Untidy Frame Boundaries

This is the bottom right corner of an image, enlarged approx 2.6x from an original video frame.

Left: enlarged with Photoshop (bicubic, smooth)
Right: enlarged with Wonder

The frame boundary of the original clearly has untidy pixels from some unknown processing before arriving at my desk. A more careful human eye would have spotted the fault, but in this case I didn’t see it until Wonder had treated it as a feature to be enhanced.

Wonder has done a great job with the knitted texture, but it also enhanced the untidy edge to the point that it becomes a thin border.

I’m not sure that Wonder should do anything different, I’m just presenting this as a real-world example of a problem encountered.

In this case, my solution is to crop a few pixels off the edges.

The “original” does have a light edge that the model is thinking is to recover. Wonder is trained for low-resolution files that have compression artifacts to recover, which would not be cases of full-frame video stills. While it helps with your file, it would not be what the model is trained for, and it is incorrectly detecting that edge as something blurry to recover.

Glad it can help your workflow though!