Background information

What you can do with Adobe’s new AI features in Lightroom

Samuel Buchmann
19/4/2023
Translation: Patrik Stainbrook

As of yesterday, Adobe Lightroom can reduce noise with AI-powered filters. The program can also automatically detect individual features and create masks from them. I tried out the new features.

First off, the header image above is a joke, of course. I didn’t age 30 years overnight, nor did I discover Botox or turn into a vampire. Sorry for the clickbait. The fact that I was able to deface the image to a horrifying extent within two minutes is all thanks to new AI features in Lightroom. Here’s the original photo.

In real life, my jacket is grey, my skin isn’t made of wax, my hair is brown, and my eyes are normal.
In real life, my jacket is grey, my skin isn’t made of wax, my hair is brown, and my eyes are normal.
Source: Samuel Buchmann

With the 12.3 update for Lightroom and 15.3 for Camera Raw, Adobe is beefing up artificial intelligence (AI) in its raw converters. Two new features are meant to save you a lot of drudgery, namely AI-assisted noise reduction and intelligent masks for separate individuals. In addition, you can now make local tone value corrections via the gradation curve. How well does it all work? And how is it useful?

AI noise reduction

Clicking on Denoise activates the new AI-based noise reduction. I can customise how aggressive the filter should be. Then Lightroom creates a new raw image in the DNG format. On my MacBook with its M1 Max chip, this takes around 30 seconds for a 50-megapixel photo. The process seems to put a heavy load on the graphics card in particular. Here, a 100 per cent crop of a photo with ISO 25,600 and +1 exposure compensation.

Left: original file, without noise filters. Right: denoise at 60.
Left: original file, without noise filters. Right: denoise at 60.
Source: Samuel Buchmann

In fact, this method removes image noise better than most other filters. While previous ones do reduce fine grain too, they also blur images and thus destroy structures. Turning up the Details slider also leads to artefacts. With AI noise reduction, both problems occur much less.

Left: normal noise filters (luminance 75, details 50, contrast 20, colour 20). Right: AI filter denoising.
Left: normal noise filters (luminance 75, details 50, contrast 20, colour 20). Right: AI filter denoising.
Source: Samuel Buchmann

The AI isn’t perfect either, as you can see compared to a «real» noise-free image that I shot at ISO 100. Using the AI filter, artefacts do show up, especially with text. Nevertheless, the overall performance of the new noise reduction feature is remarkable. In future, I won’t have to worry as much about cranking up ISO while shooting.

Control image with ISO 100 (left), denoising AI filter (centre), normal noise filters (right).
Control image with ISO 100 (left), denoising AI filter (centre), normal noise filters (right).
Source: Samuel Buchmann

Intelligent masks

Lightroom has been automatically recognising entire subjects – people, the sky, or objects, for example – for some time.

What we’re familiar with: automatic mask selection for different subjects.
What we’re familiar with: automatic mask selection for different subjects.
Source: Samuel Buchmann

The AI now also disassembles people into individual items. I can mask the following body parts separately or in any combination.

  • Facial skin
  • Skin
  • Eyebrows
  • Eye sclera (the white of your eye)
  • Iris and pupil
  • Lips
  • Teeth
  • Hair
  • Facial hair
  • Clothing

The automatic selection works surprisingly well in my tests. Only rarely does the AI miss its mark and fail to draw the line between clothing and the background, for example. In such cases, however, I can intervene manually and correct or refine the mask using a brush.

AI selection offers more control. I can mask separate body parts. It isn’t perfect, but it’s very good.
AI selection offers more control. I can mask separate body parts. It isn’t perfect, but it’s very good.
Source: Samuel Buchmann

The new feature saves a lot of work when retouching portraits. With just a few clicks I can whiten yellow teeth, soften skin or brighten eyes a bit. This was possible before the new update, but I had to do most of the masks by hand. For quick editing, the new feature is a blessing and saves me the time-consuming switch to Photoshop.

Local gradation curves

This is also in part thanks to the new local gradation curves. Until now I could only adjust the tonal values of masked areas with sliders: exposure, contrast, highlights, depths, white and black. The gradation curve was only available globally for the whole image.

Local tone value corrections via the gradation curve – here, for example, I want to lighten dark areas.
Local tone value corrections via the gradation curve – here, for example, I want to lighten dark areas.
Source: Samuel Buchmann

With the new update, this all changes. Now I can finally control individual tonal values in a masked part of the image using a curve. This feature will also save me an extra round in Photoshop on a regular basis.

Verdict: time for creativity instead of busywork

The new AI-based features in Lightroom show where image editing is headed. More automation, more possibilities. Not a bad development, as long as I can manually intervene if I don’t like the AI’s suggestions.

Unfortunately, AI selection hasn’t yet arrived in Photoshop. That would be great. After all, Lightroom still only possesses limited tools for structural retouching or selective colour correction. Adobe, in local correction, why can I only set white balance and global saturation – but not the hue, luminance and saturation of individual colours?

Despite the fast and capable selection masks, keep your hands off Soften Skin. Please.
Despite the fast and capable selection masks, keep your hands off Soften Skin. Please.
Source: Samuel Buchmann

Therefore, the tools can’t yet replace professional image processing. Really good skin retouching can’t be done using the clarity slider in Lightroom, no matter how accurately the program detects correct areas in an image. In certain cases, however, the AI masks take a lot of extra work off my hands. Unlike the purely mathematical selection tools of the past, they actually analyse the content of an image. In portraits, I get a mostly flawless mask of skin or hair in two clicks.

Sooner or later, artificial intelligence will take over more and more repetitive and boring tasks. Similar to photography, the definition of what I need to be able to do for good image processing is changing. Technical know-how is becoming less and less important, creative decisions are taking centre stage.

55 people like this article


User Avatar
User Avatar

My fingerprint often changes so drastically that my MacBook doesn't recognise it anymore. The reason? If I'm not clinging to a monitor or camera, I'm probably clinging to a rockface by the tips of my fingers.

These articles might also interest you

  • Background information

    What makes a photo «real»?

    by Samuel Buchmann

  • Background information

    Why I’ll be taking holiday photos on film from now on

    by Samuel Buchmann

  • Background information

    I tried to develop better images than NASA – and failed

    by Samuel Buchmann

Comments

Avatar