1794
Replies
Really looking forward to the pro removal - very interesting question.
I took one look at the image and thoroughly saw your point as I couldn't see how the AI could recognise the shadow as a thing to remove. Having said that I'm surprised by the result of an unpolished first 30 second hack...
It's a Generative AI Object aware remove with a bit of subtraction over the patch of sun on her left hip. There's certainly some movement in the hand and you lose the henna and bracelet etc so there is a good image but not the same image. I wonder if this is a case for "interaction" in a future development - being able to tell the AI that the selected area is a shadow so to change/remove that but preserve the underlying structure
I observe similar results. I looked at about 15 variants, and several looked pretty good:
@lloyd_pdx, if you make a similar Object Aware selection, do you get similar results? Which precise model of Nikon is this from? (Your export excluded the metadata.) The model could matter...
I suppose with reference to your original question this may not, currently, be the best way of removing the shadow from a specific subject if that subject is complex and must not be changed. I think it is doing an amazingly good job of replacing the area with a shadow on it with a different, closely matching, generated area without a shadow on it which is presumably exactly as it should function
I used to remove this sort of thing from less complex areas by spending far too long masking and adjusting which is much easier/less difficult with the new masking options (and may well belong in a different discussion with people who are much better at it than me - apologies to the mods)
The below is very much unfinished - the shadowed skin tone/abdomen shadow transition are a nightmare and would take ages so I haven't really tried and the transition on right upper leg isn't properly blended yet - but you can do things like select the shadow as a luminence range mask, adjust it to isolate the area you want to work on, intersect it with the subject, lighten it, warm it up etc etc until it matches and then you can use some healing (or Generative AI with precautions re it sampling the original image) on transitions but it preserves the original jeans/pose and hands/henna/bracelet if you can manage the skin tones and abdomen transition!
The Generative AI fix is much much (much) faster and better if preserving the specific details isn't critical
"Camera is a Nikon D850"
This is likely the cause of some of your issues when editing the NEF (as opposed to JPEG you posted), due to a Remove bug.
Many cameras have sensors a little bigger than a standard aspect ratio, and the Camera Raw engine silently crops them to a standard ratio. The Nikon D850's sensor is 8288 x 5520 (w/h ~ 1.50145), which Camera Raw silently crops to 8256 x 5504 (w/h = 1.5). That is, Camera Raw is cropping out 16 pixels on the left and right and 8 pixels on the top and bottom. The LR user never sees those pixels.
But a bug in Remove is looking at ALL the pixels, including those cropped-out border pixels. When you select the shadow on the model's legs, the cropped-out border pixels at the bottom of her legs could be still unselected, and then Remove will generate a replacement that tries to match them.
I've simulated with the JPEG by cropping another 16 pixels from the bottom, selecting the shadow, and applying Remove:
Because the bottom row of 16 pixels is left unselected, and they are in shadow, Remove generates a replacement shadow that matches them.
We don't see that problem with the JPEG, since it doesn't have any such hidden cropped-out pixels.
The workaround is to make the selection brush big and then select along the bottom border, which usually selects those hidden pixels. See here for more details on how to do that:
During the Remove beta ("early access"), Adobe isn't accepting formal bug reports, and all the feedback is lumped into this giant thread. So we don't know if they have acknowledged this as a bug -- let's hope so. LR 14 is due out in a month.
lloyd_pdx - and here is my version of a very short and first atempt. I masked a little outside the shadow, as you can see when comparing tis and the original. Also notice that legs position are different from other results. I only checked the box "Generative AI"
Het is een verwijder tool NIET een VERVANG door ander OBJECT 
I've had wonderful results with matching fills for color and content in LrC -- until -- I look closer and see that my content is in very sharp focus but the GF is out of focus considerably--beyond "soft" and basically unuseable. As a photographer, I appreciate that I should have got the right shot in the first place but I couldn't hang out over the cliff to clear the branches LOL. I will be doing some lengthy, careful cloning. 😉
AdChoices