I personally can't see the artifact. I had to subtract or XOR the two images
to identify the changes.
That said: ffmpeg with -qscale 0, 1, 2 creates (identical) images which
have approximately the same size as a conversion using ImageMagick with
libjpeg (albeit very old) with a "quality" setting of around 89%, which
is not generally considered "very good". YMMV
I'm trying to say: At first glance, ffmpeg either compresses much more
effectively (which I doubt) or doesn't use a high enough quality
setting. Whether that is intended, and whether the various qscale
settings should result in identical files, I can't say.
> ffmpeg with -qscale 0, 1, 2 creates (identical) images
> which have approximately the same size as a conversion
> using ImageMagick with libjpeg (albeit very old) with a
> "quality" setting of around 89%, which is not generally
> considered "very good". YMMV
> -qscale 1: 276184 bytes
> -qscale 1 (or 0) -qmin 1: 608994 bytes
> I still don't see much of a visual difference,
> and no blocking or other artifacts.
> With paint.NET's XOR layer filter, I do see differences
> and -qmin 1 seems to have less of them.
> 608994 bytes is close to 95% quality with ImageMagick
Could you compare the 95% ImageMagick output with
the qscale 1 output?
I tested again with those settings, and the result is better, but still not as good as a conversion by OS X Preview. I’m only observing visually, using Preview.app. The blockiness is especially noticeable on the black shirt chest area. Here are example files. The one created by FFmpeg with qmin 1 is 609KB. The one created by Preview is 1.9MB, and is nearly identical to the TIFF.