Artificial Intelligence on PhotoLab: what do you expect for?

[Wlodek]

Current public AI is a marketing term. AFAIK, technically it’s all some type of ML (Machine Learning). Probable trend (e.g. for Adobe) would be to use image recognition and identification techniques, which are already used by military and police. This would allow photo editing software to decompose the image into easily manageable objects, help it to invent correct details, make DAM easy, make your hair look better, and so on.

I would like also software to learn certain photographer’s editing style, and apply it automatically to my photos, down to the pixel level. Not sure if joking – maybe something like that is already there. Initially it will be good for nothing, but you have to take first step. Not sure about electricity bills, though.

What will still remain outside “AI” reach, are your memories.

[uncoy (Alec Kinnear)]

@Wlodek

I would like also software to learn certain photographer’s editing style, and apply it automatically to my photos, down to the pixel level

Sadly, this is coming. I’m a pretty good photographer and Midjourney (I’m just a novice) can outbat me.

Realistic

Stylised

I could take the first shot, and do that processing if I could get there with models that good but the second one is out of my reach. And this is just the beginning.

Like you I think image creation will soon become providing an image of a face and perhaps a background and then working on an ideal version. The instagram software is partly there already. I’ve been shocked at how some of these social media sensations look when the auto-filters are turned off. Completely different people.

We are slowly moving towards living in a dystopian cyberverse. The darkly amusing part is that both men and women who date online often end up just walking away from real world dates these days. The men advertise themselves as 5’11 and fit, when they are short and out of shape. The women don’t look anything like their touched-up photos.

[Fixx]

Sky replacement is not generally needed but sky enhancement is welcome. Fix burnt out areas, darken blue, tune more dramatic clouds etc.

[Kitzii (kontakt@akitzberger.de)]

AI Sky Mask, AI AWB and AI Repaire Tool would would be my favorit.

[JoPoV]

AI make my shot I would like to be a photographer is coming anyway …

[Fixx]

@florisvaneck

My take: DxO is not a DAM application

If it is not then it should supply some expansion path to an app that is DAM. I think though DxO currently IS a basic DAM.

[Stenis (Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))]

Digital Asset Mangement is a general form and I have always interpreted that as a system that is able to handle not just pictures and maybe videos but in fact all sorts of data.

Corporate/Enterprise DAM systems can handle any type of asset, pictures, video, program code, Office-documents, PDF-files and really any other file type really and it does so just by adding XMP-sidecar-files like Photolab today does with RAW-pictures.

An Excel-dokument called Calc123.XLS will get a sidecar named Calc123.XMP. It is not more complicated than that. Then depending on design different filetypes will have different XMP-schemas and forms or templates but normally there will always use the Dublin Core namespace as a common glue and that makes it possibe to search through all kinds of documents and get answers including both pictures and any other filetype. Common namespaces are Dublin Core, EXIF, IPTC, Photolab …but you are also free to make your own that suits every organisation or company.

Today many converters still relies on IPTC in their interfaces and example of this is Photolab, PhotoMechanic and Capture One and even Lightroom. Neither of these software are fully based exclusively on XMP. In Photolab I work in IPTC that in fact almost is a fossile system since decades, even if that corpse recently has start to move again to meet the needs of Adobes CAI initiative and C2PA. PhotoMechanic, Photolab as well as Capture One are all “forking” that IPTC-metadata even to XMP.

The only other software I have come across that to most extents lives up to being a real DAM among softwares dedicated to photographers is iMatch, that I have been testing last weeks in order to see if it is possible for me to migrate to it from PhotoMechanic in the first place and I have found that it is. BUT, for my needs I feel reluctant to do that transition if I don´t have too (I still have a perpetuallicence for PM Plus 6 that might work still for many years). Sad to say (because there is a lot to like with iMatch) but PM is just so much faster and effective to work with - at least for me. Big batch updates are a real pain in iMatch. I have never got that feeling with PhotoMechanic. The difference is not marginal but decisive.

If I shall be more philosofical I need to go to the old Russian/American Nobel Price-winner in litterature Joseph Brodsky. He was asked in an interview what the difference was between writing in Russian and American.
He answered: When I write in Russian I think of if it sounds good but when I write in American, I just think “If it makes sence”.

iMatch is a very complex and competent software with a substantial learning curve due to it´s purely technical design. Even PhotoMechanic has a learning curve but it is a lot more straight forward and simple to understand and configure and it is far more effective as I see it. iMatch is developed in Germany and even in my country Sweden I can see that we too often make many things a little bit too complicated. We are very much like the Germans here.

When I was young I worked in the industry with all sorts of machinery from aeroplanes, engines to printing press machinery (I was a flight engineer for a few years too when I was young working for the Swedish domestic airline Linjeflyg ) and I remember attending a certificate course for an Ameican plane and a little later a French plane and it struck me already then how pragmatic the american solutions often were and how sofisticated and often overly complicated the french were - not so far from Brodskys reflections between the American and the Europea ways really.

[gserim]

A language or text input similar to generative tools would also be useful. As an example, I have taken Joanna’s suggestion for black and white point editing:

Start with SOOC image
Find White Point and Black Point
Adjust Tone Curve based on these points
Make suggestion for contrast adjustment
Bring out more cloud detail

Each instruction requires the corresponding intelligent processing.

These instructions could also serve as a list of requirements from the point of view of Photolab users. This would have the advantage that we could formulate what we expect without having to worry about how it could be implemented.

[brendxn (Fujifilm X-T5 + Leica Q3)]

AI masking is probably the most relevant thing for Photolab. In my opinion most others would be distractions relative to other areas where the product could be improved.

Face tagging (local, not in the cloud) would also be useful to me, but I’m not sure how much others care about it. Before spending time on features like that though, I think some “basics” like smart albums, the ability to edit smartphone photos, viewing files in subfolders, etc are much more critical than most of the AI “features”.

There are plenty of AI photo editors out there, and it seems to me that DxO has positioned the product very differently, more for “purists” (and I very much appreciate that).

[Stenis (Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))]

There is Face Tagging in iMatch that I just tested a week ago and that works pretty well I have to say. There is even AI tagging of keywords and AI tagging of Descriptions is in Beta. I tried those too and was not at all impressed with the AI Tagging as it works now. The keywords it suggested did not convince me at all.

[gserim]

… and no hallucinations please …

[Stenis (Sten-Åke Sändh (Sony, Win 11, PL 6, CO 16, PM Plus 6, XnView))]

Yes, The AI-tagging in the 2023-version is absolutely hallusinating now. I´m using the AI built in in the iMatch-version. There is also a possibility to use external AI-sources BUT that costs extra and takes a special code from the provider to activate.

[NaturalThought]

AI subject masking, please. If you’re reading this DXO that would be so great.

[noir.fonce (Noir Foncé)]

Autumn 2024 update:

Of all the proposed AI applications, let’s take stock:

#1 AI mask : now in Capture 1 and Lightroom!

A lot of people would like this feature, even if personnaly I find it is not a priority: I found the Control lines and the U-points of PhotoLab enough performing and easy to use.

#2 AI highlight recovery : still pending, and I think (hope) a huge improvement is possible on this point.

#3 AI auto-fill : now in Photoshop! It seems that AI generative features are not in the spirit of PhotoLab…

#4 AI noise reduction : improved again with DeepPRIME XD2s: it remains the best in class. But is it sufficient to convince users of Capture One, Lightroom, or others to shift to PhotoLab ?

#5 AI DAM assistance : content aware adding keywords/tags to the photo library, so that photos of the same types of subjects can be searched and catalogued easily.

#6 AI auto-clean up of sensor dust : this is really what I’m looking forward to the most.

#7 AI-help for editing (automatic multi-criteria scoring/notation aid): I find the idea more and more useful when I see the new max frame rates or the pre-capture function. How to deal with such an insane amount of pictures?

#8 AI object detection and replacement : why not starting with wires, people, cars and their replacement by their background? Simple, limited, but efficient.

#9 AI luminosity, automatic white balance on a series of pictures .

#10 AI super resolution upscaler . Already available in Topaz and now AIarty Image Ehancer! Combined with DeepPRIME, this feature would give a new youth to old pictures, from 6-10 Mpix cameras. A smart way to merge old pictures with new ones.

#11 AI Auto HDR , from one picture or a series. Is there any supporter here?

#12 AI blur-removal : already existing in Topaz Sharpen-AI, but it could be a nice improvement for PhotoLab.

#13 AI improvements for Auto-horizon, Auto-perspective . Really, actual tools are very poor in automatic mode in PhotoLab. They deserve to be improved.

#14 AI lens blur simulation , that I discovered with Lightroom. AI generates a 3D-map of the image to create an artificial bokeh with realistic blur transitions + predefined bokeh styles (circles, rings, cat eyes, and so on). Interesting concept.

#15 AI-based database reliability management: finding duplicates, looking for moved files, deleted files, ignoring non-RAW files, exploring sub-folders, and so on.

#16 AI batch processing in natural language, such as: “Balance the lighting of [this set of photos] like [this one], adjust the horizon of [this set of photos], etc.

My vote is quite the same as last year:

• First, compatibility with dng of smartphones. It is not AI, but I really miss it.
• I would add this year a more precise setting for output sharpness. The two existing levels are really poor: you always hesitate between too little and too much.
• Second: #6 AI auto-clean up of sensor dust.
• Third: #2 AI highlight recovery.
• Fourth: #12 AI blur removal and #10 AI super resolution, ex aequo.

[pierre5018 (PIERRE LABRECHE)]

@noir.fonce

First, compatibility with dng of smartphones

Mee too. This was a feature abandoned by DxO. It should be revived.

[bekesizl (Win11, DPL 6, FP, VP, Nik4, Canon M50)]

For #7 try Radiant Photo.
New version is to be released soon.
As a hobby photographer with limited time and post processing knowledge it does wonders with my photos, without Generative AI.
I export the photos first from Photolab, then run them through Radiant.
It is aimed to photographer to make a first run like Auto Enhance, but with a much better quality.

[gserim]

#xx Interaction

I would like to be able to discuss with the AI system why the decisions were made the way they were and make my own changes to the joint solution. In case of doubt, I would also like to be referred to further information on the topic.

[noir.fonce (Noir Foncé)]

Do you want DxO AI to participate in the forum and discuss with us? It would be fun after all! I vote for it!

[gserim]

Whether it would be good for the forum is certainly worth discussing. I think it would be more than interesting for providing personal assistance with image processing. In any case, it should be fun .

[George]

I still wonder what AI is. One essential part of it is self learning, to me. If not it’s just a sophisticated program.

George