P: Bring back the ability of Enhance to run in the background. (LrC)

@Incredible_Explorer0544, the sluggishness you're experiencing after denoising a large batch of photos is due to the option Catalog Settings > Metadata > Automatically Write Changes Into XMP. If you uncheck that option, LR should perform much better.

With that option enabled, I've observed in my test catalog that after applying Denoise to a large batch of photos, LR starts saving the metadata to .xmp files. While that is happening (you can see the indication in the upper-left corner of LR), switching between photos in Develop becomes very slow, often taking tens of seconds. You have to wait until the saving completes before performance returns to normal.  And even then, if you apply other changes to Develop settings of the entire batch (e.g. changing Exposure with Sync Settings), the sluggish performance returns until LR finishes saving the changed metadata to the .xmp sidecars.

Normally, LR is very good about updating .xmp sidecars in background without interfering with interactive use. But the addition of the voluminous Denoise data to the sidecars, and most likely the bug with reading that data into the .lrcat database (described previously), is causing the .xmp updates to misbehave. It appears that Adobe didn't test this before releasing 14.4.

The workaround is to disable Automatically Write Changes Into XMP. I very much prefer to have the option enabled (for several reasons), but I'm going to have to keep it turned off until Adobe fixes this bug.

As exUSA said: How about allowing the amount to be set BEFORE it is applied so we don't keep getting the recalculation EVERY TIME? And retain the last setting? The new feature is implemented poorly as-is, it doesn't work as well as it used to.

I agree with this sentiment.  I use denoise at 35%.  Having to watch it calculate 50 and then have to manually fix it is a complete waste of time.  Let me set it first.  And then remember it.  Just as it used to work.

"I've been asked to find out what types of specific tasks you execute while Denoise runs in the background under the previous workflow."

This question being relayed by @Rikk Flohr: Photography demonstrates a clear ignorance by the development team of how photographers use LrC. What type work being impacted by this change is irrelevant. What is important is that users cannot use LrC (also ACR and LrD) while Denoising large numbers of photos; for professionals this cost money.

The new non destructive Denoise is good for a few photos, but blocking users form using LrC while 100s are batch Denoised shows little understanding and is extremely bad design.

The implementation of Denoise Amount adjustment for multiple photos is even worse, triggering the whole Denoise process again using AI Update. Since each photo can have its Denoise Amount adjusted without an AI Update, the current process must be totally unnecessary.

@Incredible_Explorer0544, I did some more investigation of Denoise and better understand what you're observing with regard to size of the Denoise data.

The primary location of the Denoise data is stored in the <catalog>.lrcat-data file (a scalable, high-performance database file). But when you ask LR to save metadata to .xmp sidecars, either manually via Metadata > Save Metadata To File or via Catalog Settings > Metadata > Automatically Write Changes To XMP, LR will also write that Denoise data to the .xmp file, similarly to how it writes all mask data there, including computed AI masks.

But to write the .xmp file, LR first writes a copy of the Denoise data into the main <catalog>.lrcat database file, and from there writes the .xmp file.  This obviously is unnecessary, since the data is also stored in the <catalog>.lrcat-data file, but that's what LR does (probably a consequence of behavior predating LR 14.4's implementation of Denoise). This is why the .lrcat file gets so much larger. I'll file a bug report about storing the Denoise data in the .lrcat file.

So when you Denoise a raw, LR is writing three copies of the Denoise data: the .lrcat-data file, the .lrcat file, and the xmp sidecar, with the second copy unnecessary.

As discussed previously, the Denoise data stored in LR 14.4 is only a fraction of the size of the DNGs used in previous versions. Even when two additional copies are made for .xmp sidecars, the total storage is still less than with DNG files. But as discussed in my next post, a bug in the LR mechanism for writing .xmp sidecars, probably exacerbated by the large size of the Denoise data compared to previous XMP contents, causes sluggish behavior.

Regarding the particular size of the Denoise data for your camera, the Sony a7 R3:  I download 112 sample raws for that camera from dpreview.com. The normal size of the .arw is about 85 MB, but if you enable Sony ARW Compression, which you apparently have done, the size is about 40 - 43 MB.

The average size of the Denoise data in the .lrcat-data file was about 11.6 MB/photo and in the .xmp sidecars about 15 MB/photo (the representation in the sidecars is encoded text, not binary, so a little large). The median .xmp size was 11 MB, but the largest was 56 MB.  The average you reported for your 500 photos was about 27 MB.

The sizes vary because LR must be compressing the Denoise data. I correlated the size of the .xmps with the amount of detail in the photo -- the more the detail, the less you can compress without loss of quality.  The largest .xmp (56 MB) was almost entirely of tiny deciduous tree branches missing leaves. 

In the latest version of Camera Raw (26.8.1), the automatic creation of a DNG file after using the AI Denoise function has been removed. This significantly slows down my workflow, as Photoshop now takes much longer (~10 seconds or more) to open the denoised image.

The previous workflow — where Denoise created a new DNG file and opened it instantly in Photoshop — was much faster, predictable, and allowed better file management and version control. Removing this feature without any option to re-enable it (like the former “Technology Preview” checkbox) is extremely frustrating.

Please consider restoring the option to generate a separate DNG file after Denoise, or at least give users the choice between both methods. Many professionals rely on efficient and controllable workflows — this update feels like a step backwards.

Thank you.

The XMP files's size has explode. Mine were between 10-50 kb > after 14.4 I have some at 20 mb....
I'm now back to. 14.3.1

@zoelarkinphoto: "Now, each image requires manual, individual confirmation, even if we're applying the same settings...it's always the same level of AI Denoising I'm applying - 40% is my sweet spot"

There are two ways in LR 14.4 you can apply Denoise 40 to a batch of photos without manual, individual confirmation:

1. In Develop while editing a photo that has Denoise - 40 applied, define a preset "Denoise 40" with just Detail > Denoise, Raw Details, Super Resolution checked.  Then in Library, select all the desired photos and do Quick Develop > Saved Preset > Denoise 40.

2. In Develop with the current photo, apply Denoise and then change the slider to 40. In the filmstrip, select all the photos. Click the Sync button and select just Detail > Denoise.

You can also use Auto Sync, but that requires you to first check Detail > Denoise, wait for the batch to finish denoising, then remember to change the slider to 40, then toggle Auto Sync off, which is more error prone. (And Auto Sync is dangerous -- too many reports here of it ending in tears when people forget to toggle it off.)

Auto Sync sometimes syncs a change in the slider but more often causes Denoise to be recomputed as well.

@zoelarkinphoto: "It used to take 5 seconds per image with a fast system like mine with 128GB RAM. Now it takes minutes per image."

I compared batch Denoise in LR 14.4 with the old DNG Denoise in LR 14.3 on a sample of 20 raws from Sony, Nikon, Canon, and Fujifilm. LR 14.4 was 18% faster, taking 11.5 secs/photo compared to 14.1 secs/photo in LR 14.3. (This is on my Macbook Pro M2 Max, which has a fast GPU.)

If your LR 14.4 is taking minutes per image, there's some problem specific to your installation.  

@johnrellis 

• The camera is the Sony A7R3.
• The catalog file itself is 13.3 GB and the lrcat-data folder is an additional 13.4 GB, so it's not the entire catalog folder. This is for the catalog only containing the 500 new photos. Please note that the catalog containing the 5000 pre-14.4 photos has just over 1 GB.

In the end, it's not about the actual amount of storage space. What is critical is Lightroom performance. I wouldn't mind the data stored in some XMP file and loaded when accessing a particular image, as long as overall performance would not be affected. 

@Incredible_Explorer0544, the increase in catalog size you reported for Denoise sounds too big by a factor of three, suggesting something anomalous with your installation.  What camera model(s) generated the raws you tested? I can run some tests with that camera to see if I get the same results.

Details

You report Denoise of 500 raws caused a 13 GB net increase in the size of the catalog (presumably the entire catalog folder), or about 26 MB/photo.

I just did a test with 487 photos from my test catalog with raws from 82 camera models from 9 manufacturers. It caused a net increase of only 6.7 MB/photo, which aligns with what others have reported. The average raw file size in my test was about 30 MB/photo, whereas it was a little larger in your test, about 38 MB/photo. 

LR 14.4 stores the Denoise data in the <catalog>.lrcat-data file, which is also used for computed AI masks. In my test, the DNGs computed by LR 14.3 and earlier were much larger than the 14.4 data, about 57 MB/photo. Thus, LR 14.4 used about 88% less bytes for storing Denoise data than previous versions.

You wrote, "Reviewing the XMP files I noticed that they are huge, mostly between 50 and 100 % of the raw file!" 

The .xmp sidecars contain a copy of the Denoise data, just as they contain copies of computed AI masks. In my tests, the .xmp sidecars are about 25% of the raw file size.   But in your tests, the Denoise data per photo was about three times bigger, which accounts for the much larger .xmp sidecar size.