AI Keywording – a new contender, Cyberstock

Keywording is important!

I’ve said this many times. We love taking photos and videos, and we love earning money from those assets, but we hate the middle bit. Describing, keywording, uploading, submitting to the many agencies that can take our work. For many years, I used the low-tech approach. I used IMS Keyworder where you enter a few key words of your own, multiple matching images appear, you choose the ones that best match your image and get a combined set of keywords to paste into your image metadata. I almost always write my own titles and descriptions. Surely the magic of AI can do this far better? I have written before about my experiences, but I’ve recently been trying a new contender, Cyberstock.

Stock photo and video upload process

This bit is important as well in this discussion, because your approach will dictate which tools work for you. I keep all my photos in one Lightroom catalog (145,000 images now) and I keyword and describe them in that catalog. Only when they are completely ready to go do I export the jpegs used for the stock agencies and so the jpegs have embedded metadata in the files themselves and could be uploaded to agencies one by one in that form if you wanted to. However, there are two issues with this – one is a technical one in that Lightroom has never allowed anything other than alphabetic order for keywords and some sites – Alamy and Adobe Stock – want the most important keywords first in the list. They prioritize searches against the first 10 keywords or so. I always thought it a bit funny that Adobe wants prioritized keywords and yet does not support that in their main image management system, but there you are!

It is a valid question as to why I keyword in Lightroom rather than in various systems that take in the exported jpegs and allow keywording in their upload systems. My Lightroom catalog contains all the original image – mainly raw files and processed photoshop files – and I always want to be able to find the original if I want to upload it to a print on demand site, or reprocess it using newer technology, or resize it for a big print. The keywords embedded against those master files allows me to find them very quickly. I don’t reimport my exported jpegs back into this catalog which I guess would potentially give me this capability as that would clutter my system with duplicate images. All this means is that keywording systems that expect you to export a jpeg, upload it and have it keyworded and then it gets to the stock agencies just does not work for me.

Incidentally, I do use a bit of a work-around to solve the key keywords issue in Lightroom. Every time I come across a new thing or place, I capitalize the word. So, if I am going to Costa Rica, I add “Costa Rica” as the keyword. If I photograph a bird called a brown jay, I use “Brown Jay” as the keyword. Why? The answer will appear shortly!

The second issue is that each agency has a different approach to submission and for many of them it is a two-step process – upload the images, then visit the site to submit them. iStock takes it one big step further in requiring that your keywords match their controlled vocabulary, which is actually quite a big step!

As a result, I use StockSubmitter (or its more modern version, Microstock Plus) to handle all those issues. It allows me to look at all the jpegs ready for upload and it automatically adds categories to them. Then, in the keyword editor, I can select a whole bunch of similar images and move some common important keywords to the front of the list. It also automatically puts capitalized keywords at the head of the list, so my photos of Costa Rica will have that keyword first in the list of keywords even though alphabetically it would have been lower in the order. But critically, I can select 40 or so photos from Costa Rica and assigns the correct iStock managed vocabulary keyword to each of my entries. The end result is that I have the entire set of jpegs with at least some of the important keywords early in the list, the iStock work done, and the app then uploads and submits them without any further involvement from me. Stock photography doesn’t pay well at the best of times, and so the time saved in not worrying about uploads and submission processes is very important for me.

What about AI Keywording and its promises?

You might be thinking – what has all that information above got to do with the subject of this post? The developers of most of the systems start with the requirement that the AI system “sees” the image and so it needs to be uploaded. So naturally, their thinking is that the actual jpeg that is ready for upload gets into their system, they add the keywords using AI and provide an updated jpeg with the metadata embedded. Or they provide a csv file that can uploaded with the images. The one I have been using for the past six months or so, Phototag.ai, (affiliate link), also has a Lightroom plugin so that you can work in Lightroom directly with their system. I personally don’t like that as I like to be in control of my Lightroom database, but it does work.

The new contender, Cyberstock, takes this a lot further; by allowing the upload of both photos and videos in batches and then continuing the process by storing your ftp login details in the system and it will upload the keyworded files to the agencies for you. It promised much more besides, including the combination of an AI system with data received by external systems such as Semrush to add intelligence about the search frequency for certain words to try to come up with keywords, titles and descriptions that are tuned to include words that have a higher chance of being searched for to increase the chance of the image being found in a search.

And to add to that, they recognize that each agency has different requirements for titles, descriptions and keywords. Anyone looking for the ultimate customization would tailor their upload to match what Adobe wants, say. Adobe has a limit of 200 characters in the one Title field. So, it would make sense to pack as many keywords into that title as you can and use the full 200 characters to get the maximum exposure for your important words in that title.

Dreamstime has a limit of 130 characters in the title field and like many other agencies, also allows a description field with 200 characters (I think). However, Alamy only allows a caption and that has a limit of 150 characters. You can see that a jpeg optimized for Adobe will not be accepted by Alamy, for instance.

As a result, Cyberstock can process each image multiple times, once for each agency, and then upload the jpeg with the right metadata to the right agency. And it can do that in a hands-off batch fashion. I believe that this process is upload only. You would still have to visit each agency to submit the files using whatever steps they have designed. I haven’t tested any of this, because it just does not fit my process.

Does any of this matter in the real world?

Good question. When I first started looking at Cyberstock, the titles generated were all capitalized (the first letter of each word was capitalized. That looked pretty strange to me, and I was told this was optimized for good results on the agencies. When I looked at some best sellers on Adobe, I didn’t see capitalized titles and so I asked for a feature to remove that. I also know that apps like StockSubmitter get round some of these different standards by allowing you to upload your description rather than title to Adobe, for instance. However, to use my approach of using the same file for all agencies does require some compromise. I need to keep my title less than 130 characters for Dreamstime and I need to keep the description less than 150 characters for Alamy. So even with the use of my descriptions for the title for Adobe Stock, I still end up with a sub-optimal title in theory.

I don’t know how much this matters, to be honest. We don’t know how Adobe Stock balances the top 10 keywords with the remaining keywords with key words in the title, and how they then choose popular images versus new images in the search order. Is it worth the cost of running multiple AI metadata processes on the same image to get that optimal title? I’m betting that the answer is that it is not worthwhile, but at least Cyberstock gives you the option of making your own choice.

How good are the AI optimized titles and keywords?

This is the real test of a system. How good are the results? With my current AI keyworder, Phototag, I usually look for 30 keywords from the AI system, remove a few that I think are not appropriate (it always sees wildlife in any landscape image for instance) and then I add more words of my own.

I have to say that I found Cyberstock to be initially very unstable. I asked for the non-capitalized titles and sometimes that worked and sometimes it doesn’t. When it does work, it puts proper nouns like Costa Rica in lower case, which might not be important in a search, but a human might take it into account. There was an option for what they call Long Tail keywords, or I call compound keywords. For instance, Costa Rica should be the keyword, not costa, rica. However this option didn’t work. They worked on that, and it started work, then it put “wildanimal” as a keyword and left others like central, america as two separate words. Their system is complex to handle all these agency rules, but I think the development is not 100% there. I’m sure they will get there, but it did frustrate me a bit. On the positive side, I suggested that also creating a much more complex result that could be used for a Print on Demand site to entice a buyer into buying a print. They have that on their development plans and I’m going to work with them to help define what that might be. They have been very responsive to my questions and ideas.

Comparison with Phototag.ai

I was working on some Costa Rica pictures and thought I would use those as test cases. Not destined to be big sellers, I’m sure, but they also were more difficult to keyword than perhaps more obvious photos of well-known places. Incidentally, I use 1000-pixel images to speed the upload process and both systems worked quickly, in a couple of seconds, to give me the results.

Let me say straight away that I could adjust the number of characters that Phototag.ai puts in the description. This whole exercise has helped me to understand the different constraints that agencies have, and I would now set the title to 130 characters and description to 150 characters. This also shows a related issue with Cyberstock in that there is no agency that has those specific requirements and so generating an answer for my “generic agency” was not an initial option. They did work on a solution for me by specifying a generic agency as well as the obvious ones.

However, both of these are well matched, with Cyberstock having the more “keyword” filled titles and descriptions, but they did understand what this image was and came up with good results.

Once again, Cyberstock has many more keywords in the text fields, but missed the fact it was raining heavily outside. As a human, I would have added concepts like shelter, retirement, restful, to the keywords. Cyberstock struggled with those compound keywords again, putting straw, hat instead of “straw hat”. I think the range of compound words in the Phototag keywords is better and they had a better feel for the image.

I thought I would see how well they recognized a plant. I had checked this with Google images, and it was Beehive Ginger. Cyberstock recognized that but separated the words in the keywords. So now I’m supposed to see a beehive in the image. I had chosen iStock for the agency and the title reverted to this strange capitalization again. I was also interested to see if choosing iStock would automatically come up with the controlled vocabulary that Getty uses. It doesn’t look like it and so you would have to do that on the iStock site after upload.

How about my famous cat?

I used Alamy as the agency for Cyberstock. Title is now the strange (to me) capitalized version, but I must admit it is full of important words. I do think I prefer the compound keywords that Phototag chose, such as cream color, cute cat, cat staring and so on. Those are words that I can imagine a buyer using to get this sort of image.

Conclusions

I think Cyberstock has some interesting ideas and I am sure they can improve the consistency of the results I was seeing. I like the detailed titles and descriptions and can see myself using those. The lack of meaningful compound keywords is an issue for me. I would really like to see a new field in the future for the description that I would add on Etsy, say, to really get across the background to taking a particular image so that a buyer could feel more emotionally attached to it. Creating an Etsy relevant title would be interesting as well. The reason I’m pushing this, is that my best stock photos often end up on my print sites and so it would give me an all-in-one solution for smooth uploading to all the sites that need the images.

So, there is no clear winner, but I will continue to work with Cyberstock (although see the Update below) as they evolve the system. If you are interested in trying Phototag, is my affiliate link:

Phototag AI keywording with Lightroom plugin

Update

The guys from Cyberstock reviewed this article and sent me a long email telling me basically that I just did not understand their system, I was comparing an airplane to a bicycle and that their integration of all sorts of other sources of data was making their system so much more powerful. They also said that that paying for the Stock Submitter submission system in 2026 was nonsense. I will put their email to me and my response in two comments below in case you are interested. I was told to rethink and rewrite this article and they would help me do that. I refused, clearly, and now I see they have cancelled my subscription to the service. So, I guess I don’t think I will recommend Cyberstock!