Lightroom’s AI Workflow — Why Order Matters MORE Than You Think

[Only RF]

Brian Matiash has very good and interesting video about this.

- YouTube

Enjoy the videos and music you love, upload original content, and share it all with friends, family, and the world on YouTube.

www.youtube.com

The suggested order shown at 2:28. It seems like a bit of a pain but using icons themselves as tools helps. A few other things first. I don't use HDR so that is not included in my workflow. Since Lens Corrections are applied at import I decided to leave them enabled as I have not see any issues. Since LrC is a parametric developer some may find better performance while editing with Lens Corrections disabled but I have not experienced that. I've also excluded anything that does not apply to LrC so my list looks easier

This is my short list but I will use the list from the video for reference

1. Basically I will either apply Denoise, Raw Detaill or Super Resolution so that is always first. After Denoise runs I will tweak at 100 or 200% view.
2. For the next edits the Remove tool icon is a mental road mark. If I'm going to use any AI that is next. By the way if you do step 7 before 3 and 4 you will get the update warning.
3. Apply Lens Blur if desired.
4. Crop and Transform.
5. I apply Adaptive Color profile.
6. Then apply Global edits followed by Masking and I'll tweak both as desired. I'm not too worried about updates at this point.

 

[RedCobra]

I am assuming the youtuber means all this applies to LrC as well as LR (which I don't use)

My import preset has camera neutral profile, lens correction off, sharpening off.
For me Denoise in PR5 is first. PR5 does the lens correction, sends back a DNG.
(The Cr3 file stays in the Library as well as the DNG)
Occasionally use Adaptive Color on the returned DNG.
edit edit edit mask mask mask etc.
For Blur if necessary, I use the Neural filter in PS. If GExpand and/or GRemove is necessary I use PS (Beta)
If resize is necessary I use Gigapixel as it is better and more flexible than the LrC or Camera Raw.
Sharpening is last and done with Nik Output Sharpener.

I use PS (Beta) instead of doing this stuff in LrC as LrC uses the CPU for many things, whereas PS uses the GPU more, and since I upgraded to a RTX 5060 Ti with 16GB of VRAM it's faster and seems to do a better job.
It would be interesting to see if LR, as opposed to LrC, uses the CPU or GPU more for these tasks.(I have Task manager with processing uses shown on a side bar as I edit) This all on a Wintel desktop.

Just my $0.02

 

[Dean Wilson]

1. Basically I will either apply Denoise, Raw Detaill or Super Resolution so that is always first. After Denoise runs I will tweak at 100 or 200% view.
2. For the next edits the Remove tool icon is a mental road mark. If I'm going to use any AI that is next. By the way if you do step 7 before 3 and 4 you will get the update warning.
3. Apply Lens Blur if desired.
4. Crop and Transform.
5. I apply Adaptive Color profile.
6. Then apply Global edits followed by Masking and I'll tweak both as desired. I'm not too worried about updates at this point.

Am I missing something here...like maybe Step 7?

 

[Only RF]

Am I missing something here...like maybe Step 7?

That is my refined workflow that has Lens Corrections enabled. I used the steps in the video as a reference for my explanation. I combined the AI tools available under the Remove Icon in step two - 2, 3 and 7 of my process to make to make the list shorter.

Here it is at 2:28. Since 5&6 are not available in LrC and I don't PS I didn't include them in my workflow. I probably should have said I don't use 5&6 in my first post.

 

[Only RF]

I am assuming the youtuber means all this applies to LrC as well as LR (which I don't use)

My import preset has camera neutral profile, lens correction off, sharpening off.
For me Denoise in PR5 is first. PR5 does the lens correction, sends back a DNG.
(The Cr3 file stays in the Library as well as the DNG)
Occasionally use Adaptive Color on the returned DNG.
edit edit edit mask mask mask etc.
For Blur if necessary, I use the Neural filter in PS. If GExpand and/or GRemove is necessary I use PS (Beta)
If resize is necessary I use Gigapixel as it is better and more flexible than the LrC or Camera Raw.
Sharpening is last and done with Nik Output Sharpener.

I use PS (Beta) instead of doing this stuff in LrC as LrC uses the CPU for many things, whereas PS uses the GPU more, and since I upgraded to a RTX 5060 Ti with 16GB of VRAM it's faster and seems to do a better job.
It would be interesting to see if LR, as opposed to LrC, uses the CPU or GPU more for these tasks.(I have Task manager with processing uses shown on a side bar as I edit) This all on a Wintel desktop.

Just my $0.02

Yes. This video has nothing do to with any other apps except for LrC, Lr and PS. He created it specifically to help people apply AI tools in the correct order to prevent possible issues. It is kinda like using layers. Below is a response from the Adobe Community forums. Since you don't LrC, Lr or PS (I'm assuming) this is more geared towards the other members who do. I myself did use them out of order and just pressed the reset (re-build) button but after watching it I decided to create a new workflow.

"You are free to edit in whatever order you like. Realize that there will be a price in rendering accuracy and time spent rebuilding out-of-order edits".

It takes very little time to re-build an out of order edit for a single file. If you are doing events, etc then time can be an issue. The presenter has become an Adobe employee within the last year and the list he shows is the same one on the Adobe website. He has a lot of excellent videos pre Adobe employment. He presented a few simple masking hacks he said that even Adobe didn't know about. Probably why they hired him.

 

[Only RF]

Yes. This video has nothing do to with any other apps except for LrC, Lr and PS. He created it specifically to help people apply AI tools in the correct order to prevent possible issues. It is kinda like using layers. Below is a response from the Adobe Community forums. Since you don't LrC, Lr or PS (I'm assuming) this is more geared towards the other members who do. I myself did use them out of order and just pressed the reset (re-build) button but after watching it I decided to create a new workflow.

"You are free to edit in whatever order you like. Realize that there will be a price in rendering accuracy and time spent rebuilding out-of-order edits".

It takes very little time to re-build an out of order edit for a single file. If you are doing events, etc then time can be an issue. The presenter has become an Adobe employee within the last year and the list he shows is the same one on the Adobe website. He has a lot of excellent videos pre Adobe employment. He presented a few simple masking hacks he said that even Adobe didn't know about. Probably why they hired him.

I forgot to say. For me rebuilding edits is no big deal. The comment about "rendering accuracy" caught my attention because you don't really know what can happen and it may not discernible by the human eye. Since this new workflow is very close to my previous one, a few changes were pretty easy to adjust to. Again this is more geared towards Adobe users.

 

[RedCobra]

Yes. This video has nothing do to with any other apps except for LrC, Lr and PS. He created it specifically to help people apply AI tools in the correct order to prevent possible issues. It is kinda like using layers. Below is a response from the Adobe Community forums. Since you don't LrC, Lr or PS (I'm assuming) this is more geared towards the other members who do.

I do use LrC. After processing in PR5 the DNG gets sent back to LrC from PR5. Then I do the rest of the editing/masking there unless there is something to do that requires PS. (or upsizing or sharpening)
Color profile gets set when importing into LrC (before PR5). Lens profile and Pre-sharpening gets set in PR5.
So.
1. Import into LrC. (with color profile set)
2. Send from LrC to PR5 (comes back as DNG into LrC)
3. Edit DNG in LrC. (Tone/Color Grading/Transform/Effects/and Calibration, but no detail/sharpening/Lens Correction/or Lens Blur)
4. Any AI or masking. (sensor dust/removal/heal/clone etc.)
4a. Resize if necessary in Gigapixel. (then back to LrC)
4. Send to Nik Output sharpener, then back to LrC.
5 Export or Print from LrC.

 

[jcass]

I have a habit in my workflow of cropping and masking (local edits) first. Looks like I should change that.

Which should come first in LrC, cropping/transform or masking? I've read conflicting advice on these.

 

[Only RF]

I have a habit in my workflow of cropping and masking (local edits) first. Looks like I should change that.

Which should come first in LrC, cropping/transform or masking? I've read conflicting advice on these.

It says crop/transform first but try both ways and see if you get an update warning.

 

[Bryan Conner]

I have a habit in my workflow of cropping and masking (local edits) first. Looks like I should change that.

Which should come first in LrC, cropping/transform or masking? I've read conflicting advice on these.

There are different ways of approaching this for sure. The philosphy that I agree with most is to do the following:

1. Crop and Transform: This defines the final composition and corrects any perspective issues. This is the base image.

2. Denoise: Now you have a clean, noise-free foundation for subsequent edits and avoids processing pixels that have already been discarded. (saves time maybe)

3. Adaptive Profiles: If you use them.

4. Generative AI Tools: Generative Remove or Generative Add etc.

5. Global Adjustments: Exposure, Contrast, Highlights, Shadows etc. (but not texture and clarity)

6. Manual Sharpening or Noise Reduction: if needed for fine tuning.

7. Texture and Clarity: Global or selectively using masks

8. Masking: precise targeted edits.

I am not presenting this workflow believing that it is definitely the way anyone should do it, it is just the way which seems logical to me based on what I have learned and understood (or at least I think I understand it....haha) In the end, if you like the result, it doesn't matter how you got there.

 

[Only RF]

I see your reasoning. I did seesaw between doing cropping earlier or later as suggested. Several days ago I had it near the top of the list and then decided to move to to the suggested sequence. I'm going to do a visual assessment and not apply Remove tool edits to areas that I'll crop out later. I'll see that goes. Not that it is a big deal but it was kinda nice not to have that orange reset warning illuminate several times.

I'm old dog so new tricks are always easy these days Doing Denoise first like I always did followed by the using the Remove tool icon a part of a mental road map has helped a lot with that. With my shortlist (which I'd still opening) I'm half way there.

 

[Dean Wilson]

It seems that Crop should be the first thing. Try this, mentally, apply a heavy Vignette and then crop 50%. Defeats the vignette.

 

[Only RF]

Since vignette was not specifically mentioned then I would put it in the Global adjustments which is after cropping. Again those are suggestions and nothing has to be followed.

 

[Bryan Conner]

In my opinion, it depends on how you are adding/subtracting a vignetter. If you are using a Radial Gradient Mask then it would be in the same step as other masks. If you are using the Post Crop Vignetting in the Effects Panel, I would do it as the final step, after all other adjustments have been made. It is 100% a finishing step.

If you do a Post Crop Vignetting before masking, then Lightroom/ACR will have to include the vignette in its processing. This could potentially result in a longer AI processing time as well as run the risk of introducing some type of anomaly etc. in the vignette, which I really doubt would happen. But including the vignette in AI processing such as a mask, for example, seems to be extra, unnecessary, work for the AI. A vignette is like a watermark, it goes on top of the image so to speak. I definitely would not use a watermark....and if I did, I definitely would not want AI monkeying around with it.


Once again, the above is only my opinion and if you do things in a different order and you like the result, then that is what is best for you...in my opinion...

 

[Only RF]

As Rikk Flhor said do whatever you to. I’m finding the new flow pretty easy to adapt to.

 

[RedCobra]

There are different ways of approaching this for sure. The philosphy that I agree with most is to do the following:

1. Crop and Transform: This defines the final composition and corrects any perspective issues. This is the base image.

2. Denoise: Now you have a clean, noise-free foundation for subsequent edits and avoids processing pixels that have already been discarded. (saves time maybe)

3. Adaptive Profiles: If you use them.

4. Generative AI Tools: Generative Remove or Generative Add etc.

5. Global Adjustments: Exposure, Contrast, Highlights, Shadows etc. (but not texture and clarity)

6. Manual Sharpening or Noise Reduction: if needed for fine tuning.

7. Texture and Clarity: Global or selectively using masks

8. Masking: precise targeted edits.

I am not presenting this workflow believing that it is definitely the way anyone should do it, it is just the way which seems logical to me based on what I have learned and understood (or at least I think I understand it....haha) In the end, if you like the result, it doesn't matter how you got there.

Don't know about LR, but PR5 denoises the whole frame even it you crop it in LrC then send to PR5. The DNG comes back uncropped.

 

[Bryan Conner]

Don't know about LR, but PR5 denoises the whole frame even it you crop it in LrC then send to PR5. The DNG comes back uncropped.

Lightroom Classic and Adobe Camera Raw no longer produce a DNG file when using AI Denoise. Apparently, by doing the work internally without the need for an extra DNG file is part of what streamlined the process and was at least partly responsible for this discussion. I don't know for sure, but apparently Adobe Camera Raw (which is also the raw processor in Lightroom Classic) only processes the part of the image selected in the crop. This is one of the reasons it is more efficient to crop before running AI Denoise.
Once again, this is my understanding on the topic, and I might very well be misunderstanding. So, always take what people like me on the internet say with a grain of salt and always verify for yourself if possible.

 

[Only RF]

Lightroom Classic and Adobe Camera Raw no longer produce a DNG file when using AI Denoise. Apparently, by doing the work internally without the need for an extra DNG file is part of what streamlined the process and was at least partly responsible for this discussion. I don't know for sure, but apparently Adobe Camera Raw (which is also the raw processor in Lightroom Classic) only processes the part of the image selected in the crop. This is one of the reasons it is more efficient to crop before running AI Denoise.
Once again, this is my understanding on the topic, and I might very well be misunderstanding. So, always take what people like me on the internet say with a grain of salt and always verify for yourself if possible.

I recall that happening when I used PureRaw 2.

Personally with LrC/ACR I found no time difference running a cropped file or full sized. I just tried it again for a 6000 x 4000 sized file and one cropped to 4033 x 2689. It was from a burst sequence so identical files. Identical processing times. I know the with the old DNG Denoise process it stripped your edits and put them back after Denoise did its thing. It may do (did) that with cropping as well and I will investigate it.

I'm surprised how many people complained about the new no DNG Denoise process at the Adobe Community forums. Apply Denoise and when done view the full sized file (no window) or at any size you like and tweak it as much as you like. I waited for months for that. Previously there were complaints about being only able to view it at around 240% in the Denoise dialogue window. Can't make everyone happy.

With the 14.5 release they fixed a bug that caused Denoise to re-calculate each time you tweak it thus not adding more data to the lrcat-data folder.

 

[Bryan Conner]

I'm surprised how many people complained about the new no DNG Denoise process at the Adobe Community forums. Apply Denoise and when done view the full sized file (no window) or at any size you like and tweak it as much as you like. I waited for months for that. Previously there were complaints about being only able to view it at around 240% in the Denoise dialogue window. Can't make everyone happy.

I believe that some people will complain about not having anything to complain about.

 

[Dean Wilson]

I believe that some people will complain about not having anything to complain about.

I'm not sure you fully underestimate the powerful ability of bant and blather to minimize the non-worthiness of wasting time.

 

[RedCobra]

"I'm surprised how many people complained about the new no DNG Denoise process at the Adobe Community forums. "

I like having the original CR3 and the DNG since the DNG is processed and the CR3 is not. (For archival purposes)

 

[Only RF]

"I'm surprised how many people complained about the new no DNG Denoise process at the Adobe Community forums. "

I like having the original CR3 and the DNG since the DNG is processed and the CR3 is not. (For archival purposes)

That’s one method. Since LrC is non destructive the original untouched RAW is always still there. There is also a virtual copy which is use at times. Even though storage is cheap I’ve never found the need to keep DNG’s.