Reflection Removal no longer working after latest update to LrC 15

[UliP.]

Lightroom Version Number

15.0.1

Operating System

1. Windows 11
2. macOS 15 Sequoia

Hello everyone,
over the past few days, I've noticed that reflection removal no longer works properly in LrC 15.0.1. This is the case in both Windows 11 and macOS 15 Sequoia, regardless of the quality setting. Even downgrading to LrC 14.5 does not improve the situation. Apparently, even for older LrC versions (which support this feature), “something AI-related” is loaded in the background, which enables the feature. It seems that the current AI is also downloaded for older LrC versions.

Incidentally, the same problem also occurs in Photoshop! However, I am posting it here because LrC is my preferred tool.

Here is an example that I took with my iPhone 13 Pro Max at the beginning of April specifically to test the function.
I have to say, the result blew me away with excitement in April!
Original out of Cam (April 2025):

Removed Reflection Value + 100, Best (April 2025) GREAT RESULT!!

Reflection only Value - 100, Best (April 2025)

Removed Reflection Value + 100, Best (Dec 2025 LRC15.0.1)

Reflection only Value - 100, Best (Dec 2025 LRC15.0.1

 

[John R Ellis]

Can you share the original photo file? I'd like to investigate the issue of older versions no longer producing the same result. If it's too big to attach here or it's a raw file, upload it to Wetransfer, Dropbox, Google Drive or similar free service and post the sharing link here.

 

[UliP.]

Will do it in a few hours

 

[BobT]

I've found reflection removal still works but not if denoise is used either before or after reflection removal.

 

[UliP.]

I've found reflection removal still works but not if denoise is used either before or after reflection removal.

Thanks for the tip. I tried it out. But for me, it's exactly the opposite. Until now, I had NOT applied AI Denoise to this image. I turned it on to test it out, and reflection removal now works. (Value 1 is sufficient).
However, the result is not as good as in earlier versions. See attached Pic.
As soon as I switch off AI Denoise and update the AI processing status, reflection removal no longer works.
And if I change the value for AI Denoise, I ALWAYS have to update the AI processing status! Very unusual!

Removed Reflection Value + 100, Best, WORKING AFTER USING AI DENOISE (Dec 2025 LRC15.0.1)

 

[UliP.]

Attached you can find the link (WeTransfer) to the DNG:
https://we.tl/t-YKNx78ey7E

 

[John R Ellis]

I tested your DNG with a newly installed, virgin installation of LR 14.4 on a virgin Mac OS 26.1 (a virtual machine), and the results were very similar to my LR 15.0.1 on Mac OS 15.7.1, very similar to what you posted. The 14.4 results didn't look at all to what you observed.

LR does download models from the cloud into:

~/Library/Application Support/Adobe/CameraRaw/ModelZoo/CloudDownload

but after my test on LR 14.4, it hadn't downloaded the "dereflect" model. It must have been using the version of "dereflect" built in to the LR 14.4 app in:

Adobe Lightroom Classic.app/Contents/Resources/ModelZoo/dereflect

I had 17 photos using Reflection Removal last edited in LR 14. I opened the old LR 14 version of the catalog in LR 14.4 and compared their appearance there to that in LR 15.0.1 after recomputing reflections. I didn't see any differences.

So I can't reproduce what you saw with that sample DNG, and I don't observe any differences in Reflections Removal between 14.4 and 15.0.1. I don't doubt what you saw, I just don't have any hypotheses about what's going wrong.

 

[UliP.]

Hello John,

Thank you for testing it so thoroughly!
What are your results? Are they the same as mine with 14.4, or like mine with 15 from my first post (poor results), or like mine with V 15 after I activated AI Denoise?
Did you also have to activate AI Denoise to get better results?
Thanks again!
Uli

 

[clee01l]

If you search the forum, you can find an earlier post of mine that said I was "disappointed" in the reflection results (v14 IIRC) my images did not come close to the advertised results. Since then I have not tried to remove any reflections with one exception. That exception was a scanned image where the scanner created reflections off of wrinkles in the original print . There the reflection tool did a marvelous job of cleaning up the print image. With high hopes I tried your image with denoise and without. My v15.0.1 results appear to match yours. And I think I am beginning to understand some reasons. When I first applied denoise, the process took about 9sec. I then applied reflection removal and the process took another 15 sec.

Next I reverted to the initial history step (Import) applied the reflection removal and it was instantaneous. Meaning that no new removal calculations were taken. I reversed history to the import step and applied Denoise. It too was instantaneous. Meaning no new calculations. I would expect when starting with the import step in history that Lightroom should recalculate Reflections and Denoise when reapplied instead of using prior calculations from discarded history steps. It does not.

In applying reflection removal at +100, I can see the process does a pretty good job of removing glare but objects on either side of the glass confuse the process My best results include the people on both sides of the glass. And no matter what I do in the (+) direction, the reflected hillside houses are always enhanced.

 

[John R Ellis]

What are your results? Are they the same as mine with 14.4, or like mine with 15 from my first post (poor results), or like mine with V 15 after I activated AI Denoise? Did you also have to activate AI Denoise to get better results?

My results in LR 15.0.1, with and without AI Denoise, were very similar to what you observed. However, unlike you, I didn't notice any difference in reflection removal between 15.0.1 and 14.4.

 

[UliP.]

@Cletus and John:
Thank you for your feedback. It's reassuring to know that your results under V15 are similar to mine.
However, I don't understand why I first had to activate AI Denoise for the image in order for reflection removal to work.
If I understand you correctly, you didn't have to do that beforehand, right?
I'll create a new catalogue tomorrow, import the image and continue testing.

Does anyone know of a source where I can download test images to test reflection removal?

 

[John R Ellis]

I don't understand why I first had to activate AI Denoise for the image in order for reflection removal to work.

I observed the same as you did: In 15.0.1, applying reflection removal without Denoise had only a small effect. Interestingly, if I exported the DNG to a TIFF, reflection removal produced roughly the same results as DNG + Denoise + Reflections.

 

[UliP.]

Thank you!
I will test this with RAW files from my Canon camera and report back.

 

[BobT]

And I think I am beginning to understand some reasons. When I first applied denoise, the process took about 9sec. I then applied reflection removal and the process took another 15 sec.

Next I reverted to the initial history step (Import) applied the reflection removal and it was instantaneous. Meaning that no new removal calculations were taken. I reversed history to the import step and applied Denoise. It too was instantaneous. Meaning no new calculations. I would expect when starting with the import step in history that Lightroom should recalculate Reflections and Denoise when reapplied instead of using prior calculations from discarded history steps. It does not.

Yes it's weird how these things are being implemented. Adobe seem to want to keep a memory of AI actions at a deeper level to what is shown in History. No doubt, this is to save time later on. We really should know how this is being done so we can incorporate it into our workflows. It reminds me of critical path scheduling. It's like you build a house, you don't like the walls and remove them, then wonder why the roof has fallen in.

 

[Swingman]

If you search the forum, you can find an earlier post of mine that said I was "disappointed" in the reflection results (v14 IIRC) my images did not come close to the advertised results. Since then I have not tried to remove any reflections with one exception. That exception was a scanned image where the scanner created reflections off of wrinkles in the original print . There the reflection tool did a marvelous job of cleaning up the print image. With high hopes I tried your image with denoise and without. My v15.0.1 results appear to match yours. And I think I am beginning to understand some reasons. When I first applied denoise, the process took about 9sec. I then applied reflection removal and the process took another 15 sec.

Next I reverted to the initial history step (Import) applied the reflection removal and it was instantaneous. Meaning that no new removal calculations were taken. I reversed history to the import step and applied Denoise. It too was instantaneous. Meaning no new calculations. I would expect when starting with the import step in history that Lightroom should recalculate Reflections and Denoise when reapplied instead of using prior calculations from discarded history steps. It does not.

In applying reflection removal at +100, I can see the process does a pretty good job of removing glare but objects on either side of the glass confuse the process My best results include the people on both sides of the glass. And no matter what I do in the (+) direction, the reflected hillside houses are always enhanced.

May be you are right. AI is using either discarded info or similar in the current. Let me explain what happened to landscape photoes I took in Japan. I applied adoptive landscape preset to these as starting points. There were very similar photos taken adaject or almost at the same location. Just before I could edit or during edit process the AI update symbol appeared on my IPad version of Lightroom. So I obliged. When I edited the next photo (very similar to the one just edited), there no AI update symbol appeared. Is it extracting the results from the previous photo edit?
This is not an isolated incident. It happened few times.

 

[John R Ellis]

I believe that LR / Camera Raw caches the previous results of applying some of the AI settings, including Denoise, Lens Blur, Landscape, People, and Reflections. So you can disable the AI setting, adjust other settings, go to another photo, come back, and apply the setting again, and it will use the cached result. For at least some of these AI settings, the cache is in-memory only -- if you restart LR, the cache is cleared. the size of the in-memory cache may be limited to a few or even just one entry.

And some of the cached settings are discarded if you apply other edits that affect the setting -- e.g. if you apply Landscape, delete the Landscape, then apply Remove, the cached Landscape is discarded, because it depends on the pixels affected by the Remove.

The cache entries are specific to each photo. Some cached settings, such as the results of Generative AI Remove, are shared among masters and virtual copies.

As with all such caching, this is for speeding up LR only -- you shouldn't see any difference in the computed results regardless of whether LR uses a cached setting.

 

[clee01l]

I believe that LR / Camera Raw caches the previous results of applying some of the AI settings, including Denoise, Lens Blur, Landscape, People, and Reflections. So you can disable the AI setting, adjust other settings, go to another photo, come back, and apply the setting again, and it will use the cached result. For at least some of these AI settings, the cache is in-memory only -- if you restart LR, the cache is cleared. the size of the in-memory cache may be limited to a few or even just one entry.

And some of the cached settings are discarded if you apply other edits that affect the setting -- e.g. if you apply Landscape, delete the Landscape, then apply Remove, the cached Landscape is discarded, because it depends on the pixels affected by the Remove.

The cache entries are specific to each photo. Some cached settings, such as the results of Generative AI Remove, are shared among masters and virtual copies.

As with all such caching, this is for speeding up LR only -- you shouldn't see any difference in the computed results regardless of whether LR uses a cached setting.

I discovered this Cache problem trying to run reflection removal before and after running DeNoise. When you reset the develop settings, the expectation is that LrC will reset ALL of the settings I see this as a bug.


Sent from my iPad using Tapatalk

 

[John R Ellis]

I discovered this Cache problem trying to run reflection removal before and after running DeNoise. When you reset the develop settings, the expectation is that LrC will reset ALL of the settings I see this as a bug.

I don't see the caching producing different results in my LR 15.0.1. with UliP's DNG. These steps all produce the same end result:

1. Denoise, Reflections
2. Reset, Denoise, Reflections
3. Reset, Restart, Denoise, Reflections
4. Reset, Reflections, Denoise, Update AI Settings

I think the issue UliP called out, that Reflections produces much different results after applying Denoise, is some issue other than caching.

 

[clee01l]

I don't see the caching producing different results in my LR 15.0.1. with UliP's DNG. These steps all produce the same end result:

Yes, Caching always produces the same result. The reason seems to be that Reflection removal uses the Cached image results from before so that you can not get a different result of Reflection removal with and without deNoise on the same image If I have an image that I run reflection removal on AFTER deNoise has been applied I would like to run reflection removal with out Denoise being applied and NOT use the Cached result from before but instead completely recalculate the Reflection removal.

 

[John R Ellis]

If I have an image that I run reflection removal on AFTER deNoise has been applied I would like to run reflection removal with out Denoise being applied and NOT use the Cached result from before but instead completely recalculate the Reflection removal.

Hmm, that's what I'm observing with these steps on the sample DNG:

1. Reset
2. Restart
3. Denoise -- computed in about 6 secs
4. Reflections -- computed in about 7 secs
5. Uncheck Denoise. The AI Edit Status button turns yellow
6. Update -- recomputes Reflections in about 6 secs.

The result from step 6 looks different than step 4, so it's definitely not using the cached value.

 

[UliP.]

Hello,
unfortunately, I am only now getting around to responding. Thank you for the extensive discussion.
Regardless of the fact that the results with LrC 15 are worse than with older versions, I don't understand why reflection removal only produces visible results when AI Denoise is involved. On the one hand, Denoise is not necessary for my images and, on the other hand, using AI Denoise even with a value of 1 unnecessarily inflates the catalogue and helper files.
Here is my summary again:
Visible results from reflection removal in LrC 15 only with the following workflow:
A.) Import > AI Denoise > reflection removal
Or
B.) Import > reflection removal >AI Denoise > Update AI

Should I report a bug for this process?

Thanks again for your support!

 

[clee01l]

Should I report a bug for this process?

If you can document a repeatable procedure and compare Reflection Removal in v15 against prior versions, then submit a bug report. If you report the link to the bug report back to this thread, then those of us here can comment and add support. A documented repeatable procedure that the Adobe developers can run for them selves. as well as multiple user support will be more likely give the bug reporat a higher priority.

 

[John R Ellis]

I don't understand why reflection removal only produces visible results when AI Denoise is involved. On the one hand, Denoise is not necessary for my images ... Visible results from reflection removal in LrC 15 only with the following workflow:
A.) Import > AI Denoise > reflection removal
Or
B.) Import > reflection removal >AI Denoise > Update AI

Should I report a bug for this process?

I think it's worth reporting, especially if you have more than one example, and regardless of whether the problem occurs in a prior version of LR. If you do report a bug, be sure to provide links to the problem photos -- otherwise, Adobe surely won't pay attention.
https://community.adobe.com/t5/lightroom-classic-bugs/p-how-do-i-write-a-bug-report/idi-p/12386373

I examined about a 20 of my Iphone 11 Pro DNGs and found two in which the Reflections result changed dramatically if Denoise was also applied. After you file a bug report, post the link here and I'll add my two examples.

I have two hypotheses about what's going wrong:

1. There is some issue with the way the Iphone has encoded these DNGs. There have been other bugs specific to DNGs containing raw images, and maybe this is one.

2. Some of LR / Camera Raw's image settings algorithms are mathematically "unstable", in which seemingly very small changes to the input produce huge changes in the output. That's what might be happening with these problem images -- Denoise changes the image in visually minor ways, causing Reflections to output a much different result. Similarly, exporting the DNG as a TIFF makes a visually version small change to the image visually but a large change in the output of Reflections.

If the problem is caused by some format decoding issue, I think Adobe will be more likely to fix it. But if it's a result of unstable AI models, Adobe may not have much practical control over what the AI model does on a small percentage of photos. AI algorithms are never perfect, and Adobe may just throw up their hands say, "Oh well".