Lightroom adds Feather and Edge adjustments for AI masks – but how useful are they?
One of the issues with Lightroom’s AI masks is that although they create very precise and accurate masks, they also tend to be very hard-edged. If your masking adjustments are fairly mild, you probably won’t notice, but if you push them to extremes, these hard mask edges become obvious. In particular you start to see strong edge artefacts and outlines. So can the new Edge and Feather adjustments fix this?
The first thing I have to say is that I think these are a really good addition and long overdue. The second is that I had to look very hard to find an image that needed local adjustments strong enough for these edge artefacts to become visible. I found on, in the form of this backlit Norwegian church against a strong sky, so here’s how my adjustments went.
Step 01: Subject mask
In the original shot the church is essentially a silhouette. In fact there is detail in the shadows, you just can’t see it. The first job, then, is to create a new subject mask. As usual, Lightroom recognises the main subject – it’s almost infallible! – and creates what looks like a perfectly precise mask.
Step 02: Subject adjustments
I have to carry out some pretty extreme exposure and curve adjustments to bring out the detail in the building, but luckily my RAW file has plenty of dynamic range (it’s from an OM-5 II, if you’re interested). At first glance, the result looks OK, maybe a little artificial, though, and there’s something just not quite right about the slate roofline against the sky.
Step 03: Edge outlines
If I zoom in to 200% I can see there is a distinct white outline around the roofline and the small spire on the left. This is so common with Lightroom AI masks that really it’s inevitable to some degree. Now this does look like exactly the problem that the Edge adjustment should be able to fix, right? But if I move the mask inwards, it just moves the problem. Now I’ve got a black inner outline instead. It’s next to impossible to find a middle ground, and if you do, it often doesn’t work for another section of the mask.
Step 04: Small feathering adjustments
So is the solution to shift. the mask inwards and then soften it with a Feather adjustment? Not really, because if you’ve applied some heavy local adjustments, you’re not going to be able to feather them in over such a small distances. Also, if you apply a Feather you quickly create a ‘glow’ effect that escapes the boundaries of your subject and gives you worse edge effects than you started with. By all means experiment with this on your own images. I couldn’t make it work to my satisfaction.
Step 05: Little feather or big feather?
This is always the problem with feathering adjustments. A small adjustment can quickly product ‘ghosting’ outlines long before it’s done any real good. The solution? Just increase the feathering until that glow effect is dispersed You lose all the precision in the masking and the adjustments spill over into surrounding areas, but this can often look the most natural outcome, even if it’s not the one you were hoping for. I’ve increased the Feather setting to 65 here, and while it does look a little less artificial, there’s still a bit of an HDR ‘glow’ about it.
Step 06: Maximum feathering
So what happens if I push the Feather setting right up to 100%. I do get a much more natural-looking result, but the top of the main spire is now darkened. Or, to be more precise, the mask is now so soft that the local adjustments are really weak at the edge of the subject. Luckily, there’s something else I can do.
Step 07: Back to manual masking
I’ve really gone back to basics here, using the Add button in the Mask panel and selecting the masking brush tool to brush in those areas that need to be added, notably the main spire. I’ve set the brush to auto masking and a medium flow rate so that I can build the effect carefully.
So what’s the verdict on Lightroom’s new Edge and Feather tools?
I’ve not yet found a convincing use for the Edge tool but the Feather tool is very useful, both for small amounts of edge masking if your local adjustments are fairly light or large adjustments if you have to blend in major adjustments. I think both are worth having, even overdue, but they’re not the game-changers that users might hope. As ever, successful masking means using the right tools in the right way, and there’s no single masking tool that can do everything you want. Not even with AI 🙂
Presumably auto and semi-auto masking functionality in every software application uses machine learning in some form and AI, characterised by much greater sophistication and autonomy, is essentially a development of that. Or am I missing something else really significant? For those who have not used the new processes you describe, what were your expectations based on whatever Lightroom technical/marketing documentation has provided? Do I detect a subsequent hint of disappointment therefore? That is, though speed and ease of use on challenging images are evident benefits, they are quickly eroded when the user is as scrupulous as you are?
Ha – a hint of disappointment, yes! I don’t think I’m terribly scrupulous, it’s just that once you notice an issue with specific tools you start to see it all the time. I do think AI masking starts from the wrong premise, or at least a limited one. It assumes we want to isolate objects and areas, when very often we’re trying to emphasis lighting or change the composition or tonal balance in a photo with what we would have called dodging and burning back in the day. With regard to Lightroom’s new Edge and Feather tools, I think they are definitely worth having and help you improve AI masks, but they are a particular fix for a particular tool, not a new way to approach masking.
After reading your comment I mede another version that I was much happier with using regular masking tools (two radial masks and a bit of erasing, all subtasks of a single main mask). Sometimes it’s easier to do a thing yourself than explain to a machine how you want it done! Radial masks are my favourite. With a little skill in shaping and positioning a radial mask and adjusting the feathering, you can disguise it remarkably well. It also addresses the way the light falls in an image in a way that AI masking doesn’t.