iMatch - Autotagger - AI API will change my workflow now completely

[Stenis]

Despite IPTC and XMP has been is use for decades, I feel it is still not all that mainstream that webb-applications are designed to actually use these metadata in the pictures themselves when pictures are displayed in these applications. Mainstream has been to use an SLQ relation-database as the data source with linkes to the pictures or to add these data manually through some user interface.

Of course, indexed XMP-metadata might also be stored in relations-databases but in this case I´m talking about the XMP-metadata stored for example in JPEG-pictures.

I have many years been a member of "Fotosidan", which is the biggest photo community in Sweden with around 250 000 members and more users than so. It is a commercial site financed by advertisements for photo equipment beside member fees. It covers most peoples needs  think.

I have my photo-blog hosted these where I have around 40 illustrated stories published. I have also some portfolios stored there with pictures I use also for my blog-stories. The metadata have been important for me to reach out beyond the scope of Fotosidan itself. It really makes a world of difference.

To the point: Fotosidan has supported the use of embedded-IPTC-metadata for a long time but I have never before felt it was any idea to use it, since this workflow wasn´t all that streamlined before I started to use iMatch, AI and Autotagger. I had normally to write all Descriptions manually and fill in a lot of other fields too anyway like Headlines, Copyright notices together with a few other tags too, BUT now this is going to change.



On this side I had to write this metadata manually before but now I just check a checkbox when I start a new Portfolio to tell Fotosidan to use the IPTC-data in the pictures insted.

Example of the  "Portfolio": https://sten-ake-sandh.fotosidan.se/viewpf.htm?pfID=393543

My new workflow:

iMatch

- I add static metadata by the iMatch Metadata form to all pictures I will keep.
- I then confirms all faces in the pictures that I want to get Autotagger to inklude in the text.
- Then I add Descriptions and Keywords using Autotagger and Open AI GPT 4.1 Mini.
- After that I use the Map-form to set the coordinates and Reverse Geocode with the help of Google Map API and the OpenStreetMap-map.
Since I use DXO Photolab I have to start with adding the metadata because if I don´t and make a Virtual Copy of some picture without metadata the copy will never get any metadata on it (if I don´t add it later on manually).

In Fotosidan

- I create a portfolio and configure it to use IPTC-metadata in the pictures
- I open the Portfolio-maintenance form and batch import all the pictures i want in the Portfolio.

DONE



An example of how it looks in the Portfolio after the pictures are uploaded

There is also a picture viewer were we can look at the flow of new pictures and there we even can see the geodata. It looks like below. (You have to scroll to see it)

https://www.fotosidan.se/gallery/viewpic/3978216.htm?set=lp


This improved workflow of mine makes me very happy even if there are things I still want to improve in my iMatch-workflow. One thing makes me pretty irritated and it is that the integration with Capture One seems to be impossible to get right between CO and iMatch. Photolab works fine with iMatch but not with CO. It has earlier been the same with PhotoMechanic and CO BUT CameraBits fixed that recently in the ever last updare of version 6 of PhotoMechanic (new version now). Even if my main converter is Photolab I have to use even CO for tethering when digitizing old color slides and for mor demanding jobs where Phtolab just isn´t advanced enough.

[Mario]

And so time is freed for more important things. Perfect use of DAM.

[Stenis]

That is exactly what this is about.

I think Fotosidan is an interesting example just because it has this design and ability to use the metadata embedded in the pictures. Most webbsites do not make use of IPTC/XMP. When I worked with FotoWare DAM-systems I learned that it was possible to have a different approach to that and actually use the IPTC/XMP-metadata directly in the webb-applications. One of the four-five parts of Fotoware is Fotoweb and it is using the indexed Fotoware XMP-metadata databases (you can have many active simultaneously like in PhotoMechanic). (Foto is by the way is not a misspelling of English and FotoWare is just sort of a strange Norwegian mixture of Norwegian and English - Norwish)

Is this the explanation why even you have developed your own IPTC/XMP-driven webb-application with iMatch Anywhere? I think I will have a look on it by myself. Is that software aimed just at smaller companies or is there a possibility to use it openly with a license for unlimited users (Internet-version?). I´m talking about publishing content to the public on a totally open site on the Internet where licenses can´t be used. I can´t figure out that by the info I got here:

https://www.photools.com/imatch-anywhere-editions/

I have tested to have several databases mounted even in iMatch when I first tested iMatch. Is it possible to query more than one database at the time even in iMatch and publish the results to a public audience??

[Stenis]

I have tested a little bit now and read some more info in the Imatch Anywhere. I can see that this application seems oriented towards "in house" use of companies or organisations and that is fine but a did not see ny opening to publish a viewing environment for general use on the Internet for the public. Is that a plausible description Mario?

Do you know any other products that will be able to read the metadata from the JPEG-pictures and use it like Fotosidan can?

[sybersitizen]

https://sten-ake-sandh.fotosidan.se/viewpf.htm?pfID=393543

Very nice.

I will just point out that the English description for one of the photos I looked at has some problems (I don't know if it was AI generated):

"Former Kingdom of Ladakh Kashmir India 1976, A ladakhi man is leading a couple of yaks. The India army camp in the back ground. There was an army alert when we were there so we met a lot of army lorries with fuel and war material on the narrow and quite dangerous."

This wording would be better:

"Former Kingdom of Ladakh Kashmir India 1976, A ladakhi man is leading a couple of yaks. The India army camp is seen in the background. There was an army alert when we were there so we met a lot of army lorries with fuel and war material on the narrow and quite dangerous road."

[Stenis]

Yes you miight be right, all the first pictures from Kirtipur are all AI. The second half with pictures from Ladakh might be partly AI or me. In fact OpenAI writes clealy a better English than I do - I'm Swedish which abslotely is a handicap here even if I have worked with American software companies for many years in my job , so that is definitely a reason to use AI instead of writing it myself.

That Portfolio upload is just an JPEG IPTC-test in order to test the workflow itself. With that I mean to verify that these IPTC-texts really are read-used-displayed in a proper matter by Fotosidan Portfolio. Since they are not in Fotosidan Blogs I wrote their developers yesterday and asked them to look into that problem.

One thing now for me is that I have taken a decision to reprocess most of my Description-texts since most of them are just common descriptions. There are some sets of pictures thoguh where AI just won't be sufficient because it can never understand the deeper meaning of what these pictures really means and shows and the significance of the them. That goes example for some of my picrures taken 1978 at the Saur-Revolution in Afghanistan.

In that case AI would never be able to really explain what is going on and OpenAI is strictly unpersonal and can't even understand a simple relation between people on the pictures and that makes rather simple to distinguish between personally written texts and ones written by AI-systems. I have no reall problem with that when it concerns pure descriptive texts and if it gets wrong, the tone wring or we feel we have to intervene it is very easy tovdo so but so far I havn't felt especially many times that I have too.

The reason to is that I have worked pretty much in the first months with optimizing me static prompting at the prompts for Descriptions and Keywords. My trick is also always to add a mandatory first row with place and year where a picture is taken and mark with a hyphen where I leave it to AI. Like:Mandatory text: (just to make AI to always start with the following text on the first row):  Bridgetown Barbados 2016 - The Friday Fish Festival - and then comes the AI generated text.

Since we are living in a very changing imaging world I also have seen I have to reprocess even a lot of these old digitized color slide pictures since their poor technical quality disturbes me a bit. They are once processed with DXO Photolab  which can't either denoise or sharpen them properly because these tolls just works properly for RAW but luckily I have found Topaz Photo AI will be able to sort that in a much better way.