I have a family portrait of a freckled young person that has sentimental value. When I apply conventional de-noise and other techniques, including AI image restoration, freckles go poof. If I could select the freckles and move them to a new layer, then I could process the base layer in GIMP etc, yes? Since I don’t have permission to upload my portrait, I am offering a similar image from a free stock source …
You probably could but it would be very time consuming. - I guess if it means that much to you put your time into it.
Couldn’t use the select by colour as her hair is almost the same colour
1 Like
could you replicate the issue with that example image? I don’t understand how some simple denoising would be wiping out freckles.
are we talking about a scan of a damaged film print or something?
I wonder if this is a job for the healing tool. I just did the nose very quickly and I am sure it could be done better. Could you please explain why you want a new layer of just the freckles? What are you trying to achieve?
Depending on what you are trying to achieve, you might operate on a highpass-filtered layer mask. High pass filtering can be used with a high enough contrast so you see a grey image with prominent freckles. The layer itself is put in Subtract mode. The contrast curve of its layer mask can be companded inward to acheive better transparency. Then it can be inverted and multiplied onto a duplicate of the original, masked, and brushed in with a big radius 25% gradient tip at 25% opacity. I use the big gradient brush because it masks the eyes and the hair from edition.
I was thinking of the GIMP tools that let you remove certain repetitive patterns. Wondering if they can be used in this case?
The main culprit is AI software like GFPGAN. But, this stock photo I grabbed is the one instance where it does not remove freckles. Of course.
Here is what the AI does: It repairs any damage by referencing the “normal” human faces it was trained on. It also smooths out random irregularities that are inconsistent with human faces. Best of all, digital noise of all types is poof gone. So far, with the exasperating exception of the stock photo I picked, it has removed all freckles. If I had them on a different layer, I could do all my other steps and then return the charming flecks of pigment.
1 Like
Okay, this gives me something I can start tinkering with! I will return after I have some results.
1 Like
Betcha the stock photo was in the training set. 
1 Like
Moderate success with the following technique:
- If you use the wavelet decomposition (
Filters > Enhance > Wavelet decompose) the image is decomposed in layers where each layer corresponds to features with a given spatial frequency. - By toggling the visibility of layers in the decomposition, you can find out which layer(s) correspond to the bulk of the freckles.
- Once you have identified them, delete all other layers in the decomposition, and set the decomposition group to
Grain merge, then hide it - Apply any enhancements to the initial image
- To restore the freckles, make the decomposition group visible again. You can adjust its opacity for effect, or add a layer mask to it so that it only works on the freckles.
In this example, I applied a rather destructive Gaussian blur to the initial image:
Yep 
Thank you so much for the visuals/screen capz!
To be honest, this term confused me. I wasn’t sure if you meant disappeared or made worst. I am still not 100% understanding what you are trying to achieve because you are unable to post the real images. However, if you are trying to keep the freckles but subdue them then maybe let the AI do what ever you want it to do and then open the AI output and with the original image stacked underneath and using a mask with reduced opacity paint back in a hint of freckles.
If my suggestion is just noise, please ignore it and I hope the other suggestions here are helpful.
Rereading the original post makes me wonder what denoising program would remove freckles if that is what is happening. I am also wondering why AI is needed for photo restoration. Of course we are having to work blind because the real image can not be posted. What are the problems in the original image that need to be treated? Again, if this question just adds noise to the thread, please ignore and I won’t be offended.