Mask AI Deficiencies Seen

Evaluating mask AI, I find serious bugs that need addressing:

  1. Parts of areas in red get in the masked image. Red should be an absolute “remove” order, under no circumstances pixels in red should be kept: this produces inconsistencies.

  2. After the “Pan” hand tool is selected, nothing else works until it is deselected again. The tool should automatically deselect when the user selects another tool, like a brush or a bucket, to keep workflow smooth.

I am interested in this program, but it is obviously not ready for prime time yet.

Mask Ai leaves a white artifact around every bit of the masking. Using v1.0.2

Seems to mask well until you place a fill behind the cutout. Then it’s Obvious!

Also, it does no better than PS masking on areas that aren’t extremely contrasty or easy to mask. Trees seem to be Ai’s forte. Amazing that the background gets transparent, however, there is still the white outline artifact.

Wanted to buy; it looked on the surface to be fantastic.

Not ready for serious use.

Have you seen the webinar? If not I recommend spending a little time to get to know how to use the tools …

https://community.topazlabs.com/t/webinar-mastering-masking-in-mask-ai/12372/5

I have seen the webinar. Both my main complaints are not covered

  1. Red areas not being removed in AI mode.
  2. Pan tool does not automatically deselect when selecting another tool

All in all, a worse experience and worse results than Remask 5 for difficult images.

The Pan tool issue is a known problem but for the rest I am having a hard time reconciling what you have said as here is an example after 30 seconds of brushing and then Compute AI:

Could you post a screenshot of the TriMap & Result as I have done please?

Try using the Edge tools.

In this image, the window is filled with red. When computed, some of this area shows as keep. The problem worsens when refined in AI mode.

Ok first thing I would try is Contrast mode and see if that makes a difference. Also I wouldn’t refine in AI mode, rather I would use Contrast.