Using a large, 16x20 240dpi photograph....
I created a completely non-feathered mask of a subject against a background (mask-view reveals black and white with no greys)
I moved the subject with the content aware move tool. The portion where the sudbject used to be is clean and devoid of any trace of the subject.
The subject, now moved, is showing feathering in random parts of their formerly masked outline. The subject's hair is feathered to the point that the subject is missing nearly all of their hair which bleeds into the background. This is not a case of a bad mask. This was the doing of the content fill.
Can you post a screen shot of the before and after:
More details about platform, etc may also be helpful:
That means that the selection is touching the hair and is trying to blend it. If you watch the sneak video for this feature, Bryan suggests Select>Modify>Expand... and push the selection out 5- 10 px.
I didn't see Bryan's video/suggestion, however, I did instinctively push out the mask by 5 pixels with the expand option in the example I posted.
Having read your suggestion, I tried a few variations (10px expanded, 20px expanded, 30px expanded) and I'm getting the same results. The girl's hair seems to appear better now than at 10px, but the male's hair seems to look the same, as well as his jacket. I always found the original content aware fill tool a little sloppy requiring human intelligence to "tidy up" the rough edges that content aware fills leaves behind. The move tools appears to be no exception and the way it works, post touch ups seem like more work than simply cutting into a new layer the old fashioned way.
30 pixel expanded mask (20 pixels looked about the same):
Using your image, I tried a different selection method, and just used a rough rectangular marquee to select the couple.
I even had to remove the "original" text from your post
Here's my result:
New technology sometimes requires new workflow to work with it! Funny here the simple rectangular selection made for a better run through with CAM. How about you?
Using a rectangular selection, predictably, yielded a highly unbelievable background.
As I said, the old method of cutting the subject out into a new layer and filling in the background contents with content aware fill seems to still be the only option if print enlargements and believability are of any concern.
The content aware move tool seems to work well as far as display viewing is concerned as is evident with some of the demonstration videos I've watche. I'm clearly pushing beyond its limits with my 16x20 print senario.
The problem with this new tool is that it requires a level of control much like in the "refine mask edges".
I didn't want to algorithm to destroy the image data 50-80 pixels into my subject's head! I just wanted a clean cut-out.
I understand it's trying to BLEND the subject into its new environment. But it did so Too drastically and without any user control.
Had there been some kind of panel that let me use a smart radius, or feather setting, contrast setting, that would have helped ME give the TOOL a better idea how my particular mask would best blend into its new environment. That might just comlicate it since it's meant to be a user-friendly "quicky" tool.... and go over people's heads. I personally love the mask refine tool and use it ALL the time. But I'll be honest, if these are the sorts of results that will be delivered with large sized images, my faith in it has started out very low.
the mask can be refined before-hand using the actual 'refine mask' control panel then making the move after the fact. (or just feathered in one big brush stroke)
While your results did preserve the integrity of the subjects, the background surrounding them is still very odd looking - meaning - another trip to the quick-mask and a cleaning of edges (if it were made into a new layer)... kind of rendering the tool a little silly considering, as I said before, it was probably designed to be a one-step tool in the first place.
I was mostly concerned with testing the process. The process might vary from image to image. I created the selection of the figures very quickly using the quick selection tool and then expanded it using the Select > Modlfy > Expand 15 pixels like Jeff described. I took the screen shot of the selection in Quick Mask mode just because it is easier to see versus the standard blinking outline. I agree the background might need some work. Actually your background that was generated originally (with the tight selection around the figures) was pretty incredible and very believable. I still find using new the tool quite a bit faster than trying to accomplish the same thing in Photoshop CS5.
I must say the Content Aware Tools are pretty useless right now - they produce subpar results. I think you could introduce "High quality mode" where it would take much more time to calculate all the pixels, but with professional-looking results. Every example you show here is flawed - the edge is always visible, regardless of the selection.
I think it would be helpful if one had the option of defining the 'replace' source - ala the new patch Content Aware tool.
My beef - I do a very close selection on some of my examples, specify 'Very Strict' and it leaves a rather large chuck of the image behind. Shouldn't there be a setting for the tool if you have a near exact selection?
Try making a much larger and more generous selection, with plenty of space around the main subject, and then experiment with "Strict" or "Very Strict" Adaptations.
You can cycle between these Adaptations while the Selection is still active to see which mode works the best.
First, create an empty Layer above your main Layer;
choose "Sample All Layers" for the Content Aware Move tool and do the move on the empty layer;
and then use a layer mask on that layer to hide any unwanted bits or to feather the edges of the moved item.
While I agree with your method as I'd be willing to bet it is probably the best way to get around the tool's shortcomings, the fact is that it requires just as many steps to implement than the old CS5 method of doing this.... Again, defeats the idea of this being a 1-step trick.
I prefer to finesse my one-step tricks!
I am not a button-clicking auto-anything sort of operator!
Don't bother with a Layer Mask if it is too much trouble but do use larger selections than the ones which you apparently have been using; and also note the very nice feature of being able to use the Healing tools and CAM tool on an empty over-lying layer.
You could also use an eraser on your retouching layer if you are really Mask-allergic?!
While I appreciate masks, and generally do things on new layers anyway -- it still seems odd that it wouldn't account for all the section improving tools that have been introduced in previous versions. CAM seems to want users to be deliberately sloppy.
You'll find CAM less "wild", and discover how best to house-train it, as you work with it for a bit.
If you are using it in Extend mode, move your selected area so that active section is still overlapping the area from which it came.
You can repeat that several times to spread some texture over a wider area.
Medium and the Looser Adaptations may work best when doing this.