Skip navigation
`Ken`
Currently Being Moderated

Disappointed in PS CS5 content aware fill

May 11, 2010 6:11 AM

My first few attempts at using the content aware fill have been disappointing. I created a panorama and wanted to fill the foreground (misty water) and the sky (cloudy) to the top and bottom of the frame.  I selected the transparent pixel areas that needed to be filled with the water and sky, then I expanded the selection 6 pixels, so that PS had some pixels to reference for the filling operation. There wasn't much sky above the tree line, but the content aware fill worked fine in this area. However filling the foreground with the misty water did not. PS insists on referencing pixels well out of my selection and using these as part of the fill. Even after repeatedly undoing and redoing the content aware fill, PS insisted on using pixels well out of the selected pixels to create the new content aware fill. Why is this?

 

It seems that the only user control over this tool is the selected pixels. Shouldn't this tell PS the pixels you want it to reference for creating the fill? Why does PS insist on going well outside of the selected pixels for a reference. If I wanted it to reference other areas of the photo I would simply include those areas in or very close to the selection. Isn't the tool suppose to use the selection as a guide? How can you tell the tool to use the selected pixels as a reference?

 
Replies
  • Currently Being Moderated
    May 11, 2010 6:40 AM   in reply to `Ken`

    Content aware fill only uses the selection to determine what you want replaced.  It will sample the whole image and take what it needs to fill in whats missing.  The point of content aware is you don't need to choose where it samples from, it does that automatically.  The thing that most people don't get is content aware will not work perfectly in every situation.  However it is much better than manually doing the whole entire thing by hand.  It's going to save me HOURS every day.

     
    |
    Mark as:
  • Currently Being Moderated
    May 11, 2010 6:48 AM   in reply to `Ken`

    No, selections don't limit the areas where Content Aware Fill takes it's info from. As far as i can tell, it's taking the info from around the selection that you made so try selecting only what you want to replace or do it in multiple steps.

     
    |
    Mark as:
  • Currently Being Moderated
    May 11, 2010 8:01 AM   in reply to Zeno Bokor

    Not so. I have a tiny area on the edge of a photo with the front of a car sticking in. I used it as a test image as cropping it was part of the expectations, so it was ideal as a test image. Well, it left parts of the car, part of the curb and a trace of the bench behind the car...and a hunk of chain link fence from halfway across the page thrown in for good measure! It actually was quite hilarious!

     

    I tried various methods to acquire the selection, all outputting varying degrees of that mixture. Saving time? In this case, should I have chosen to not crop, I could eliminate the car, fix the bench or take it out entirely with healing and the clone tool...maybe 15 min max. Fixing the result of content aware would be longer as I would lose certain reference points immediately in front of the car.

     

    Now if one could predict and control that behavior, it would be a wonderful tool for the world of the bizarre! Especially if that tool could reference parts of another picture, at it's pleasure!

     

    I do like the tool but like the OP, fixing the edges of a panorama is at your peril. Again, doing panos should include enough to account for the crop needed at the conclusion of the stitch.

     

    Deus ex machina has a sense of humor....

     
    |
    Mark as:
  • Currently Being Moderated
    May 11, 2010 11:13 AM   in reply to Hudechrome-sd9sPI

    I've used it a few times and I think it's OK - Like most new features, generally, they tend to get better with future releases.

     

    I'm only using the trial version to date as I'm still experimenting with CS5 and I think it's much improved over cs4. I'm trying to break it to see if it crashes and so far it's worked perfectly, even with a 55ins 3D layer at 200 ppi.....:0 - as soon as the box version hits my local computer shop I going to be a little bit poorer.......it's a worthy upgrade.

     
    |
    Mark as:
  • Currently Being Moderated
    May 11, 2010 1:40 PM   in reply to `Ken`

    I've been playing around with the trial for a few days now. I took me a while to get around to content-aware fill; and as expected it fails miserably almost every time. Not that I'm surprised.

     

    What did surprise me was how obvious it looks - conspicuous details are repeated etc. You may as well put up a sign: CONTENT AWARE FILL HERE. Most of the time you can spot it from across the street.

     

    It's OK as a starting point for the real work, but that's all it is.

     

    Some other good stuff may still make the upgrade worth it, though. My personal verdict is not out yet (the balance is tipped slightly by my having to cash out for a whole suite).

     
    |
    Mark as:
  • Currently Being Moderated
    May 11, 2010 2:13 PM   in reply to D Fosse-QDEaQ1

    Examples there :

     

    http://forums.adobe.com/thread/635813?tstart=0

     

    Can't understand why the edges of the image

    (when selected) get darker.

     
    |
    Mark as:
  • Currently Being Moderated
    Sep 20, 2011 5:37 AM   in reply to `Ken`

    Content Aware Fill simply does not work! 

     

     

    I would suggest that once you use a tool such as magnetic lasso or marquis square to define an area:

     

      To replace:

     

      1. Tones with colour (e.g. a scratch on a car, breadcrumbs on a tabletop)

     

      a. The eyedropper be used to sample a colour.

      b. Once the colour is defined, then the content aware is used.

     

      2. Pattern with a pattern (e.g. beach sand to replace dead fish and weeds on a beach)

     

      a. Use the rubber stamp to define the pattern that will replace the "corrupt" area

       b. Once the pattern is defined, then the content aware is used.

     

      The problem also has to do with Internet reporters or whatever they're called. They all follow each other and make the same comments.

       What happens is that all of these NetPorters end up almost verbatum with the same same material. Unfortunately, software netporting is not unique.

       The same holds true for stock netporting. Some "expert" in some office off Wallstreet (everyone is moving away from there) writes that Company B23W1X

       should make $1.78 a share with net profit for the quarter to be $ 148,000,000. Pretty soon there are 100,000,000 same same articles copied verbatum from

       the NET or some other "expert" copied form the first  netporter expert , but "adjusted" the data by 1 cent and for the profit 138,000,000. COmpany B23W1X only makes $ 1.77 a share mand only $ 147,999,999.00.      

     

      What happens - all the brokerage houses sell.

     

    Unfortunately, what happens with software is that everyone forks over a bundle of money for the "new features" only to find out they don't always work!

     

    Bottom line: wait a few months till the regular public and users start posting.

     

    The same stampede has occurred with Apple's new operating system Lion OS 10.7. Apple sold 1 million  of this OS in 1 day!

    I waited, and started to read the problems people were having. I'm still considering the new OS.

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Sep 20, 2011 5:54 AM   in reply to Stacey Bindman

    Stacey Bindman wrote:

     

    Content Aware Fill simply does not work! 

     

     

    What planet are you living on, Stacey?  Is there LSD in the atmosphere there?

     

    Show an example and someone here will be happy to tell you how to use Content Aware Fill to fix it up.

     

    Are you expecting it to be a perfect fix without any further work?  That it isn't but it's 10,000% better than cloning!

     

    BeforeCAF.jpg

     

    AfterCAF.jpg

     

    -Noel

     
    |
    Mark as:
  • Pierre Courtejoie
    7,038 posts
    Jan 11, 2006
  • Trevor Dennis
    5,914 posts
    May 24, 2010
    Currently Being Moderated
    Jan 27, 2012 12:10 PM   in reply to `Ken`

    I worked out right from the start that the trick is to make a rough selection of just the pixels you want to be considered for your CAF and copy them to a new layer. Then select your holes, or the bits you want to remove, and use CAF. 

     

    So if you don't have a whole lot of sky and you need to fill a hole in the sky but it keeps being filled with half a tree from the lower part of the image, copy what sky you do have to a new layer, and use CAF on it.   Easy peasy.

     
    |
    Mark as:
  • Currently Being Moderated
    Jan 27, 2012 12:37 PM   in reply to `Ken`

    `Ken` wrote:

     

     

     

    It seems that the only user control over this tool is the selected pixels. Shouldn't this tell PS the pixels you want it to reference for creating the fill? Why does PS insist on going well outside of the selected pixels for a reference. If I wanted it to reference other areas of the photo I would simply include those areas in or very close to the selection. Isn't the tool suppose to use the selection as a guide? How can you tell the tool to use the selected pixels as a reference?

     

     

     

    Most of the disappointment expressed by the posters above seems to be centered on CAF's using source material they didn't want when filling the removed area...

     

    If Content Aware Fill is bringing in things you don't want, that's easily taken care of by just using a layer mask to tell it exactly what you do (or don't) want it to use.

     

    You'll likely be happier with how much better/faster it works when you follow a few basic tips on how to get the best results.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 30, 2012 7:55 PM   in reply to Noel Carboni

    Hi Noel,

     

    This image is absolutely too simple to use an example!

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Nov 30, 2012 8:14 PM   in reply to Stacey Bindman

    If my image is not to your liking, please feel free to provide another one illustrating the problem you're having.

     

    I have others...  How's this?

     

    ScreenGrab_11_30_2012_230732.jpg

     

    ScreenGrab_11_30_2012_230800.jpg

     

    I can't help but think anyone who thinks CAF "isn't working" simply has wrong expectations of what the tool is good for.

     

    -Noel

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Nov 30, 2012 8:22 PM   in reply to Noel Carboni

    Here's another before/after pair.  If you're not getting results like this then you've got a specific problem.

     

    ScreenGrab_11_30_2012_231903.jpg

     

    ScreenGrab_11_30_2012_231928.jpg

     

    -Noel

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Nov 30, 2012 8:28 PM   in reply to Stacey Bindman

    You mentioned, way back when, a beach image.  Do you see a problem here?

     

    ScreenGrab_11_30_2012_232605.jpg

     

    ScreenGrab_11_30_2012_232619.jpg

     

    -Noel

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 30, 2012 8:42 PM   in reply to Noel Carboni

    Hi Noel,

     

    I was just about to write "how about taking all of the trees out, when you added the other image - thanks.

    I find that when there are suble tonal differences between 1 area and an adjacent one in most

    of Photoshops tools, I start to get problems.

    I was also just working with the quick selection, and magic wand tools.

    I had posted this about the select>>color range in my title: Does Adobe Ever Exchange Copies of Photoshop for Honest Reviews? (November 28, 2012)

     

    Copyright-April Toy-Before-For Noel.jpg

     

    Noel:  What I'd like you to do is clean up all of the scratches on this car using the content aware tool.

               You can select the area at 4:00 P.M. from the # 52 between the 2 red lines as the "source" to clean up the toy.

              To be fair, I don't know if the image size is good enough for the job. Also, I expect some scratches to appear all over.

     

              As for the copyright, it's not for my vanity. I write a blog on old toys, and have plenty of images on Flickr and Picasa.

               I keep having to chase people down for stealing my photos.

     

    Thanks for you assistance,

     

    Stacey

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Nov 30, 2012 8:45 PM   in reply to Stacey Bindman

    You mentioned removing blemishes from a car...  Here's some Content Aware Fill to remove a blemish and even some things that are not blemishes.

     

    ScreenGrab_11_30_2012_233653.jpg

     

    ScreenGrab_11_30_2012_233708.jpg

     

    I was amazed at how well it did on that little blemish right on the tip of the nose.

     

    -Noel

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 30, 2012 8:53 PM   in reply to Noel Carboni

    Hi Noel,

     

    Maybe we're out-of-sync, but I'd like you to try the rusted racer toy that I placed here on this topic.

    It's the subleties and slight nuances of colour and contrast that I have the most trouble with.

     

    Thanks,

     

    Stacey

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Nov 30, 2012 9:04 PM   in reply to Stacey Bindman

    LOL, I think you have wrong expectations for the tool if you feel it should make short work of the eleven thousand or so scratches on that model. 

     

    But it's not completely a bust.  You can use the spot healing brush set to content aware mode to carefully to paint over the individual little spots.  It actually does reasonably well.

     

    CarBeforeAfter.jpg

     

    And yes, you should be working at higher resolution.

     

    -Noel

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 30, 2012 9:16 PM   in reply to Noel Carboni

    Hi Noel,

     

    Thanks for the try.

    However, I find that with any Photoshop tool, I'm always getting the same problem.

    The solution always takes longer than I think that it should.

    You've mentioned your answer in several different posts, that is, a different tool may be needed, but I find that the same

    thing happens - the more complex the image (like the scatched car) the longer the time to finish the image - regardless of the tool being used.

     

    So I am to now understand, that where there are more difficult images to work with, I should expect to spend more time?

    Maybe, my expectations were too much regarding what PS could do?

     

    Thanks again, Noel,

     

    Stacey

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Nov 30, 2012 9:25 PM   in reply to Noel Carboni

    By the way, a little trick that can sometimes help with situations like what you're showing here is to actually use the Clone Tool, set to Lighten (since the blemishes are substantially dark), then clone nearby areas.

     

    A bit more work on the above image with this technique yielded...

     

    CarBeforeAfter2.jpg

     

    Sometimes a complex image does take a bit of elbow grease.  Your task, as Photoshop expert, is to find the combination of things that gets you to where you want to go with acceptable quality and at acceptable speed.

     

    There's really no magic.  But Content Aware Fill does get close - noting my streetlight removal above, for example.

     

    -Noel

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 30, 2012 9:46 PM   in reply to Noel Carboni

    Hi Noel,

     

    Thanks for that added effort.

     

    Here's a "trick" that I figured out about 2 months ago. I changed my old toy photo style to simply shooting

    on a white foundation. If I want to get rid of scuff marks, dirt, or distracting shadows I do this:

     

    1. Select the darkest problems and get rid of then in any way: eraser, rubber stamp, magnetic lasso, etc.

    2. Increase the brightness to get rid of the rest of the "subtler" problems. However, you'll need to do some testing

        to see what the maximum amount is for any of the tools.

    3. Then bring back the contrast or density of the items that were on the foundation (This is where you have to test for that "ceiling"

        or maximum that you can go with any of the tools.

    4. You'll now have a spotless and pure white foundation.

     

    Stacey

     
    |
    Mark as:
  • Noel Carboni
    23,513 posts
    Dec 23, 2006
    Currently Being Moderated
    Nov 30, 2012 9:50 PM   in reply to Stacey Bindman

    Excellent technique.  Shooting carefully often makes post processing much easier.

     

    -Noel

     
    |
    Mark as:
  • Currently Being Moderated
    Dec 1, 2012 8:39 PM   in reply to Noel Carboni

    i have said it before, Noel. YOU HAVE THE PATIENCE OF JOB!

     

    vince

     
    |
    Mark as:

More Like This

  • Retrieving data ...

Bookmarked By (0)

Answers + Points = Status

  • 10 points awarded for Correct Answers
  • 5 points awarded for Helpful Answers
  • 10,000+ points
  • 1,001-10,000 points
  • 501-1,000 points
  • 5-500 points