Copier le lien dans le Presse-papiers
Copié
The photo has a tree covering the top 2/5 of the frame, land and other stuff filling the bottom 2/5. In the middle is sky (interupted by the tree trunk. All sky is selected inclding some parts between the leaves, I type in "fill sky with clouds and blue sky" or "fill sky with clouds" and all is see is a bigger hole in the tree showing more sky. What did I do wrong? How to make my now grey/white sky fill up with blue and clouds?
Copier le lien dans le Presse-papiers
Copié
Addendum: With the sky part selected I tried the "Edit --> Sky replacement" and I was prompted with. "Could not replace the sky because no sky was detected" I would assume this is why the AI Generative fill did not work. but either wayI do not understand why this is so. DOes the sky have to be on top with no obstacles?
Copier le lien dans le Presse-papiers
Copié
Please try Sky Replacement without any active Selection.
Copier le lien dans le Presse-papiers
Copié
@pintree3 can you show your image so we have an idea of what you are trying to accomplish?
Copier le lien dans le Presse-papiers
Copié
Thanks ye all 🙂
pfaffenbichler your method did not work at all. The download is here kevin, a jpg version. I have many of such shots BTW (with the sky not on the top)
Copier le lien dans le Presse-papiers
Copié
Well, to be blunt that image seems pretty bad for what you are trying to do.
What’s up with the overlayed pattern? Is that a flyscreen?
I expect full automization may not be a realistic option at current, but a Layer Mask based on the green Channel and a very plain Layer Mask for the horizone should provide a fairly decent basis.
Copier le lien dans le Presse-papiers
Copié
I'm also having trouble getting Generative Fill to create realistic clouds to use to enhance a bland sky. I've asked for "a few small clounds" which generated a fill with a large mountain range and some big clouds, I've asked for "a thin layer of circus clouds" which generated cumulus clouds, and another one where an eliptical selection was made, asked for circus clounds and it generated nothing. I'm beginning to wonder if the language understanding model is not able to parse out the key parts of my request. I have had some pretty good results when I'm making a selection of something I want removed from the image. In those cases I did not specify what to do, just clicked the generate button with no instructions.
Copier le lien dans le Presse-papiers
Copié
Could you please post screenshots with the pertinent Panels (Toolbar, Layers, Properties, Options Bar, …) visible?
Copier le lien dans le Presse-papiers
Copié
I had to recreate my attempt at adding clouds to the sky of a black and white image. Exactly what you see below is a little different than what I saw yesterday when I was working on this photo. This first screen shot is what I started working with. Not quite all of the layers are visible in the layers panel. There's a B&W layer at the top of the stack.
The next screen shot shows my request to add a few high-level cirrus clouds to a portion of the sky.
And here is the resulting fill:
This is somewhat different from many of the results I received yesterday, but it still illustrates the problem. First the clouds are cumulus and not cirrus and they are not along the top of the image. They fill about 80% of the selected area. Second, the Generative Fill added more mountains above the ones that existed before the fill was performed and are now blocking some of the sky. The ones added today are a bit more consistent with the real mountains in that location, while yesterday's were jagged, rocky peaks. Third, there was a strange object added to the top of the outhouse and the lettering on the sign over the door that said "Shown by Appointment Only" has been changed to something that is not recognizable. The outhouse was not in the selected area to be filled. I will add that the instructions I gave to Generative Fill for this example are not exactly the same as what I was telling it to do in yesterday's attempt, but the general nature of what I was asking for is the same. The other Generative Fill layers you see in the layer stack worked quite well to remove things that I did not want in the image. For those Generative Fills I did not give instructions for what I wanted. I just selected the area I wanted "replaced" then clicked on the Generate button and let it do its thing.
I'd appreciate any thought you might have about the way I'm trying to give instructions or whether I'm expecting the tool to do more than it's currently capable of doing.
Bill
Copier le lien dans le Presse-papiers
Copié
@Bill Junk this might be too specific for the AI at this time. A better solution may be to use Sky Replacement. There you can either use existing images or upload your own sky images to use. This was an out of the box somewhat Cirrus looking cloud formation:
Copier le lien dans le Presse-papiers
Copié
Copier le lien dans le Presse-papiers
Copié
The sky that you added to my image is more of what I was hoping Generative Fill would provide. I suppose that in time things will improve significantly.
Bill
Copier le lien dans le Presse-papiers
Copié
I think you could rethink your use of Generative Fill for cases like this.
Why not create cirrus clouds and then place them in the proper Layer Mask?
As one would do when combining (stock) images before.
(Though I can not vouch for Genetaive Fill’s metereological savvy … so whether the cirrus clouds meet your expectation remeins to be seen.)
Strangely I think until recently there used to be a thread on prompts for Generative Fill tagged to the top of the Forum.
Still I think a prompt like «the sky needs …« is not ideal, the verb seems vague and unhelpful.
Copier le lien dans le Presse-papiers
Copié
The reason I went to Gnerative Fill is that it seemed like a reasonably direct (easy?) way to change or remove something that I didn't like or to which I wanted to make a modest improvement that didn't really change the basic nature of the image. I guess I was expecting more capability from Generative Fill than was really available right now. I have not seen a thread that suggest how to structure Generative Fill requests. All that I've seen are the short, one sentence suggestions that come along with invoking the tool. I'm also somewhat torn between leaving an image the way it was shot or making changes that don't retain the basic content and integrity of the original photo - in other words after eigher using Generative Fill or blending in part of another image, the new image is not the real scene, but rather some kind of abstraction of it. In some ways it's a bit of an ethical dilema for me. How I reconcile what to do must consider what the intent of the final image will be. Am I trying to do artistic creation and "reality doesn't matter" or am I trying to show a real place as I saw and photographed it with just a few changes the enhance the composition. These new technological advances sometimes make the decision as to what is acceptable to do much harder to make.
Copier le lien dans le Presse-papiers
Copié
1) I suppose Generative Fill will be improved with time, whether it ever will ever your or the general public’s expectations remains to be seen.
But removing unwanted (smallish) elements in a photograph seems to work pretty well, the noise/grain may need checking, though.
2) It is honorable that you consider Image integrity; but unless you work in journalism/documentation but rather in advertising/fashion/art/… I think you can assume that the audience is (at least vaguely) aware that the image is not a plain, unedited representation of a scenery.
I like to compare such photography to the paintings, in particular portraits, one might encounter at the great collections – who would expect Rembrandt, Dürer etc. to have provided an exact, documentary representation of a person/scene?
Copier le lien dans le Presse-papiers
Copié
Third, there was a strange object added to the top of the outhouse and the lettering on the sign over the door that said "Shown by Appointment Only" has been changed to something that is not recognizable.
You can still edit the Layer Mask to hide unintended elements.
Copier le lien dans le Presse-papiers
Copié
@c.pfaffenbichler That would leave you with a hole in the generative results. Your original idea of generating the sky completely separately then compositing into the image is a better option than trying to edit the mask post generation.
Copier le lien dans le Presse-papiers
Copié
Had not thouight about doing that. I'll need to figure out how to do it. Thanks for the suggestion. Much appreciated.
Bill
Copier le lien dans le Presse-papiers
Copié
@Bill Junk make a selection approximate to the sky area. Then create a blank layer and hide the original image layers. Generate your prompt (this is "whispy clouds")
Then make a selection of your original sky area and use that to apply a new mask to the generated layer:
You can shift the edge using Select and Mask to contract/expand the area to remove fringing.
Copier le lien dans le Presse-papiers
Copié
I haven't figured out how to do that with a mask on a Generative Fill layer. Using a brush to remove some of the mask didn't seem to work. I'm probably doing something wrong, but have no idea what.
I ended up using the Sky Replacement feature which let me select a reasonable sky for this image just to make the sky more interesting since the original color image had a blue sky with no clouds. Then I was able to correct an artifact resulting from something I removed with a Generative Fill layer. I applied a Brightness/Contrast layer as a Clipping Mask applied to another Generative Fill to get things looking smooth and continuous. Likely the hard way to do it but it works and I'm satisifed. Although I've been working with Photoshop for many years there's so much more I need to learn!
As far as a comment I made about trying to retain image integrity, there are times that I think it is important even when it's not related to image use in commercial settings. This image was a case where I was willing to let artistic creativity be the guide, rather than showing the scene exactly as it was originally photographed. There certainly are a few people who would know that the image has been doctored, but they are not likely to ever see the final one. Here's the original color image followed by the "new" black and white artistically creative image. Thanks for all the helpful suggestions. For anyone interested this photo was taken in Silver City, Idaho.
Copier le lien dans le Presse-papiers
Copié
Then you can Group the Layer and add a Layer Mask to the Group for additional masking.
Could you please post screenshots with the pertinent Panels (Toolbar, Layers, Options Bar, …) visible?
Copier le lien dans le Presse-papiers
Copié
1. Select the existing mask on the generated layer and delete it.
2. Make your new selection of your original sky.
3. Select the Generative Fill layer - create a new mask.
4. With the new mask selected - use Select and Mask (button in properties or option bar above) and adjust Shift Edge setting.
Trouvez plus d’idées, d’événements et de ressources dans la nouvelle communauté Adobe
Explorer maintenant