I am not sure where this needs to go. It’s long, but I hope it might be helpful.
I have now been able to have a few “proper” editing sessions with Luminar AI. I certainly have not discovered everything, nor was I trying to be systematic, but I did use it enough on “real” images and projects to form an impression.
I did not have any particular issues with installation or activation, though I did need to set myself a new password for my Skylum account in order to activate. The existing password was working fine for all other purposes (including downloading the software from my account).
My first experience was a crash after half an hour (using it as a photos extension) which lost all that edit. That added to my sense that this software was not ready for release and that it had been very heavily hyped. I have not had any hard crashes using it as a stand alone on my iMac and it has grown on me, though it has a lot of problems and gaps.
Slightly to my surprise, I am not going to be asking for a refund. This software does not live up to the orchestrated hype, has many more problems than it should and I think is never going to be an important part of my workflow because it is not reliable enough and because the opinionated approach Skylum has adopted adds unnecessary friction and obstacles to what I want to do. I am also concerned that it is not robust: there are too many tales of things working for a while, then crashing and burning. I can’t commit important work to this app.
Despite that, like Luminar 4 with its many problems, it can help me be creative with particular images and has a few tools (e.g. AI enhance or sky enhance) which are genuinely excellent. It will have a relatively small role in my image work, but it’s just about worth what I paid for it and so I will keep it.
I’ve often said I have a love-hate relationship with Skylum and Luminar. I have been disappointed much more than I have been pleased and that’s mainly because they loudly offer a single, all-conquering approach to photography, marketed extremely hard, but in my experience, deliver something which is very much less, with many important features missing, undeveloped, not working well or removed as their software is updated. Luminar AI feels like another iteration of that process, I am afraid.
Overall, there are too many things missing for this to be a key part of a semi-pro or enthusiast’s workflow, or even to be the only photography software that anyone needs. It’s too simple, but in many ways too complex at the same time. I frankly don’t believe that the pro photographer affiliates who have been singing its praises for many weeks will be using this as their only or main software. I also don’t think it hits the spot for the iPhone photo influencer either. I don’t know who its intended audience is, or whether it will be a particularly good fit for anyone, which may be part of why it has to be marketed so hard. Brilliant software does not need hype. People who need it quickly realise it meets their needs.
Issues so far:
Thankfully, it’s a lot faster than Luminar 4 on my iMac (2019 and Big Sur) but my main photography programs on the machine (Apple Photos, Pixelmator Pro, Raw Power, Affinity) are all basically instantaneous with no perceptible delay whether I am loading, exporting or editing. Luminar AI is mainly usable, but I can count many delays in seconds, e.g. importing a folder, applying a crop and lots of things feel laggy (e.g. moving a slider and waiting to see what it does).
Exporting is very slow - a basic jpg export of a single image was nearly ten seconds even without re-sizing. This compares with under a second in almost all other photographic software on this computer.
Given the strong position and unfortunate history that Skylum has in relation to photo libraries, DAM and the like, I am not at all sure why they choose to have a catalog at all.
This feels like the worst of all worlds. You have the delays and risks of importing, updating and moving around in a catalog but you have none of the benefits (e.g. ratings, metadata etc). My attempts to sync or copy-paste adjustments failed completely if I had used any kind of local masking and I couldn’t see much evidence of AI selecting and applying adjustments (e.g. skin) even in a series of portraits of the same person, in the same lighting and more or less the same pose.
Use as a plugin or photos extension
I have tried to use it as a photos extension. It works, to a point, but as well as being slow to load the image, more than once it has crashed, losing all my edits and dropping me back into Photos.
There are many other reports of problems installing it, crashes and generally not playing nicely with other apps. Given my comments about the Catalog, I’m most likely to want to use it in this mode, and it’s not stable. It’s OK as a stand-alone, but that adds complexity (exporting from one application and re-importing to round-trip) that’s not welcome.
Adjustments are sometimes unpredictable
Others have mentioned sliders doing nothing. I have not been able to pin this down in detail, but I have certainly moved some sliders over their full range and not been able to see any change in the image while other sliders are so aggressive that I don’t dare even hover the mouse pointer anywhere near them. Some (e.g color harmony) work well.
I don’t know if the AI is not seeing a suitable element in the image, but surely that should grey a slider out.
It will only straighten or apply perspective correction if the image contains massive clues (e.g. a clear horizon or lots of vertical pillars). It sometimes makes things much worse. It sometimes does nothing.
The AI suggested crops have quite often done things like crop out half a hand or foot from a person - an absolute no-no.
The image 3d adjust sliders act very oddly and are usually much too aggressive. I had an image of a dancer with converging verticals (because I was stood below the stage) and I just could not achieve the slight vertical adjustment that I wanted, that took only a moment in other software.
Changing the crop ratio (e.g. to locally print an image onto 7x5 paper) resets the crop completely. It would be much better to attempt to preserve the previous crop as far as possible. this is another illustration of why being able to have more than one digital version of the same image is valuable (so it can be kept in various formats)
The selection brush is laggy and imprecise. Some imprecision is acceptable as the repair needs to blend with the surrounding area anyway, but having a lag between moving the cursor and drawing in the selected area is irritating friction.
The results are variable, to say the least, and unpredictable. Some objects are removed completely and seamlessly, others are not. Sometimes removing a blemish produces a new blemish some way away from the one you removed. Erasing objects at the edge of the image does not work well (e.g. a streetlamp that you did not quite see in your viewfinder). There is almost always something left.
Clone and Stamp opacity
Thankfully, this brush is much more responsive and precise than the repair brush. Unfortunately, even when set to 100% opacity and hardness, it does not always cover completely, allowing what is underneath to show through. Additional brushing usually works.
For an AI based software, it is surprising and disappointing that none of the brushes are even slightly edge or content aware and that there are no automatic selections (e.g. of the subject or background)
It is not a great clone tool, but it works.
Whoever designed this hates histograms. S/he is clearly not a photographer, nor has s/he ever had to produce a photograph for a purpose (e.g. for print, or to go on a web site). It is worse than useless because:
- It is much too small (Even smaller than the much too small one in Luminar 4)
- When shown it covers part of the image
- There is no way to accurately evaluate luminance levels. The left hand and right “edges” of the histogram are indented from the edges of the panel where it is displayed and there is nothing to indicate where 0 and 100% are until you start to crush the whites or blacks and see a peak there.
- The little triangles to show hot and cold pixels are extremely small and fiddly to click and I can’t see any other way to turn that display on.
A perfect histogram would:
- Be of significant size
- Be accurate
- Have guides to show 0%, 50% and 100% luminance and highlight (on the image and in the histogram) the shadows, highlights etc when they are being altered and (in a perfect world) an option to show ten zones with a choice of algorithm (linear or gamma).
- Have an eye-dropper tool so you can see where a point in the image falls in the histogram
That might seem overkill in a program that is meant to do it all by AI and not be technical or complicated, but there is really no better way to pull off an accomplished portrait, for example, than by making sure your skin tones fall in the right luminance levels and there is the right balance between, say, highlights and shadows on the face.
Templates are very hit and miss
Putting it bluntly the AI is not very intelligent. I’m not a great preset user at the best of times, but I had hoped that reasonable suggestions might give me a place to start. That has not really happened so far. I end up going straight to edit.
Template names are not helpful and it is very hard to see what a template will do unless you apply it, and then it is irritating to have to reset all adjustments. The preview is not slick.
Template organisation is poor
Supplied templates are organised into eight categories which cannot be edited at all. There is a separate section “My Collection” which has purchased, legacy and user-saved templates. It’s only in there that you can give templates you have saved a sensible name, you can’t name a template when you save it. There is no folder or category structure.
Templates are limited in their usefulness
I worked across a range of genres and image-types. I was able to find pleasing templates for landscapes and streetscapes, at least as a jumping off point. I was able to find one or two for portraits. There were one or two pleasing templates for “nature” (flower and plant) photography. There was no template I could bear to use at all for any images of performances (concerts and dance), for seascapes and coastal landscapes or for bird and wildlife photographs.
I realise that this is all subjective, but I found myself using templates very much less than marketing suggested. Looking through suggested templates and finding nothing of interest takes time from the process, when it is supposed to save it.
Everyone developing for the Apple ecosystem has known for many months that there is a new direction but Skylum has struggled even to get even basic Big Sur compatibility, but Luminar AI seems to cause problems on older Mac OS’s too. The software does not seem well optimised for Macs (though Windows users might say the same!) and there are multiple design choices which do not sit well with a Mac environment. It’s probably part of why it’s slow.
Compare that with someone like Pixelmator, where they are already using all the features of the new Mac hardware. I’ve seen a demo, and Luminar looks very, very, very slow indeed in comparison.
Please sign in to leave a comment.