p.1 #1 · Using AI in Your Sports Photography Workflow.
On YouTube earlier this week, someone posted a video demonstrating the use of AI culling and editing software commonly used in wedding photography, but applied to sports photography.
Apart from perhaps AI noise reduction, and for non-editorial use, AI generative tools, are you using any AI software or tools in your current workflow? If not, do you expect to at any time in the near future and if so, what?
Recently, I heard an interview with a wedding photographer who commonly comes back from jobs with more than 5000 photos. He claims that before he began to use the new culling and editing software a wedding would take him approximately 18 hours to post process. It now it takes him about three, because he’s been using the technology and it has learned his style and preferences.
I had a strong, visceral, negative reaction to that… I guess that’s primarily unfounded disbelief on my part.But I am stepping back a little, understanding that at some point AI will inevitably be part of my workflow… When and the extent is to be determined.
Curious to hear what others are thinking about this topic …. guess I’m just trying to prompt a discussion on this topic to try to get my head around it.
p.1 #2 · Using AI in Your Sports Photography Workflow.
There are a couple of products out there and I think the paid versions offer a limited free trial. I tried a couple but found that they do not yet do a good job with sports images at least for culling. It's possible that they might do a decent job at editing once you have culled for your keepers. I can see a time, maybe not too far off, when these programs do a nice job culling sports images.
p.1 #3 · Using AI in Your Sports Photography Workflow.
PureMichigan wrote:
On YouTube earlier this week, someone posted a video demonstrating the use of AI culling and editing software commonly used in wedding photography, but applied to sports photography.
Apart from perhaps AI noise reduction, and for non-editorial use, AI generative tools, are you using any AI software or tools in your current workflow? If not, do you expect to at any time in the near future and if so, what?
Recently, I heard an interview with a wedding photographer who commonly comes back from jobs with more than 5000 photos. He claims that before he began to use the new culling and editing software a wedding would take him approximately 18 hours to post process. It now it takes him about three, because he’s been using the technology and it has learned his style and preferences.
I had a strong, visceral, negative reaction to that… I guess that’s primarily unfounded disbelief on my part.But I am stepping back a little, understanding that at some point AI will inevitably be part of my workflow… When and the extent is to be determined.
Curious to hear what others are thinking about this topic …. guess I’m just trying to prompt a discussion on this topic to try to get my head around it.
Not certain how AI would be a benefit for me in culling given that I tag potentials in the camera while shooting on field. Only the tagged photos are downloaded. Now an AI app/plug-in that knows how I like to crop and straighten sports photos would be helpful.
p.1 #4 · Using AI in Your Sports Photography Workflow.
Not an AI tool, but a colleagues husband is building me a tool that reads bib numbers at track meets and auto inputs them into an IPTC field so I can run code replacement in Photomechanic. It is like magic. It took several hours of work and dropped it to about 3 minutes of review time. We're still tweaking the tool, but his plan is to eventually release this to the public. I'll post some more updates as we move forward with the project.
p.1 #5 · Using AI in Your Sports Photography Workflow.
Culing is the worst part of the job. I shot sports for 15 years and not only is the action fast, but the turn around on the images is also. So I rate in camera during time outs, etc. and download those for quick turnaround. Outside of sports photography, I see how nuances in a photo take me time to appreciate.
So I think there is a place for the AI to cull, and a place to reflect a bit more for what I like in a photo. Maybe it comes down to enjoying the "craft" of photography, which culling is a part of, like it or not. It might not be what is the fastest way to do something.
Thinking out loud. Since I'm 70 yo, retired from everything, I see AI everywhere, but I can't find a need for it.
p.1 #6 · Using AI in Your Sports Photography Workflow.
The advancements in AI are so quick that I have no doubt that AI will get an increasing role in sports photography and culling. Time will tell how quick.
Looking at my workflow as an enthusiast amateur photographer mostly shooting local amateur sports, rugby mostly. I do not see AI taking over my culling, as in all honesty I do not have a clear consistent logic myself. Some thoughts with regard to AI:
- I do sometimes publish photos that are out of focus. Reasons to do so are that it has to do with a certain moment in a game, for example a disputable try in rugby. AI could not recognize that most likely and will discard the photo in automatic culling due to it being out-of-focus. Knowing the moments requires having the knowledge of what happened during the game, not the photo itself.
- I try to get photos of most of the players, also of those less talented or less prominent in the actions. Those photos are not great sports photos, but do have value. It might be challenging for AI to understand this, certainly at this point in time. In the future AI may analyse complete sets, recognize players and try to get at least x photos of player y making it through the cull.
- In some cases I select photos for non-sports reasons. Examples would be players returning from an injury, playing a last match or family members playing together of against each other in the same photo. AI most likely has a problem with that, as the selecting is not based on the photo on its own. In the future you might give AI a photo of the players for which there is a specific interest and let it run through the set to for example find all pics showing father and son in the same frame. Technology already exists I think, just not made easily available for this purpose.
- With sports it is more difficult to frame in camera for the perfect composition, due to distance and angle to the action. As a result many sports photos are cropped, certainly in comparison to other genres of photography like wedding, portrait, event and landscape. When culling I try to imagine what I can do with a crop. AI will have the ability to make such assessments if it is properly trained for it. That requires AI to have both the uncropped photo and the photo after cropping. Current AI is mostly trained on published images, but in most cases AI will not know if the published photo was a crop and how heavy a crop it may have been. Training AI for this is possible, but I do not think commercially viable at this moment. In the future we will be able to train/ tune models ourselves, so that might help.
One of the next steps for AI in sports photography I would expect is in tagging/ keywording and categorization. For example with rugby, AI will be able to recognize if the photo contains a ‘scrum’, ‘line-out’ or ‘tackle’ and tag a photo accordingly. Also facial recognition is already available, so AI could help tagging players on each photo in a photo set. Maybe in combination with recognizing jerseys and numbers as well. I just have not seen it marketed in a way that it would suit the sports photographers workflow. But since the basic technologies are available, I see it as just a matter of time to become available to everybody.
Just my 2 cents, as I am by no means an AI expert.
p.1 #7 · Using AI in Your Sports Photography Workflow.
Good coversation ....
I tag because I transmit courtside/pitchside etc... so AI culling is of limited value. Here's what I would be interested in ...
1) AI denoise applied in-camera or a LOT more speedily during the edit process. That can't be too far on the horizon. At some point it would be nice to shoot at, let's say 8000 ISO, and be able to program in-camera, "make these photos look like 100 ISO coming off the card."
RIP ISO as a practical part of the exposure triangle for most forms of photograhy -- I think it's coming.
2) I'm interested in auto editing. While I enjoy the editing process, if AI can "learn" my style (and it does for the wedding folks) and apply it, I'm happy to use it even if it gets me 90% of the way there.
It would be nice if there was a program working in the background of Lightroom that's constantly learning. My guess is that is on the way.
Exciting times .. if you would have told me 10 years ago that my camera would shoot 120 fps, it would seem just as far-fetched as some of the AI stuff. So we'll roll with it!
p.1 #8 · Using AI in Your Sports Photography Workflow.
dugaut wrote:
Culing is the worst part of the job. I shot sports for 15 years and not only is the action fast, but the turn around on the images is also. So I rate in camera during time outs, etc. and download those for quick turnaround. Outside of sports photography, I see how nuances in a photo take me time to appreciate.
So I think there is a place for the AI to cull, and a place to reflect a bit more for what I like in a photo. Maybe it comes down to enjoying the "craft" of photography, which culling is a part of, like it or not. It might not be what is the fastest way to do something.
Thinking out loud. Since I'm 70 yo, retired from everything, I see AI everywhere, but I can't find a need for it....Show more →
100% agree. Mirrorless photography is becoming faster and faster with more and more photos expected to be delivered. Culling is the worst part of the job and the culling workload grows every year. Unfortunately, it's still the part that AI is weakest at in my opinion. I'd much rather have AI edit my photos than cull them currently. It's getting closer though.
p.1 #9 · Using AI in Your Sports Photography Workflow.
cocodrillo wrote:
Not an AI tool, but a colleagues husband is building me a tool that reads bib numbers at track meets and auto inputs them into an IPTC field so I can run code replacement in Photomechanic. It is like magic. It took several hours of work and dropped it to about 3 minutes of review time. We're still tweaking the tool, but his plan is to eventually release this to the public. I'll post some more updates as we move forward with the project.
I saw your comment about the bib identification tool you have been testing and I wanted to see if it was available yet, even in beta form. I'm a motorsport photographer and I'm curious if it would work on race cars. Perhaps I could help your small team test it out.
p.1 #10 · Using AI in Your Sports Photography Workflow.
I agree, I hate culling and it’s the one area of photography where I grow the slowest. I save way too many shots in thinking I may see them differently in 5-10 years, which has happened to me. I’m getting better at trying to keep the keepers and cull everything else instead of grinding through and only deleting the missed shots. This is easier to do when I’m working with shots from someone not in my family. Hard to delete the family ones.
p.1 #11 · Using AI in Your Sports Photography Workflow.
PureMichigan wrote:
Good coversation ....
I tag because I transmit courtside/pitchside etc... so AI culling is of limited value. Here's what I would be interested in ...
1) AI denoise applied in-camera or a LOT more speedily during the edit process. That can't be too far on the horizon. At some point it would be nice to shoot at, let's say 8000 ISO, and be able to program in-camera, "make these photos look like 100 ISO coming off the card."
RIP ISO as a practical part of the exposure triangle for most forms of photograhy -- I think it's coming.
2) I'm interested in auto editing. While I enjoy the editing process, if AI can "learn" my style (and it does for the wedding folks) and apply it, I'm happy to use it even if it gets me 90% of the way there.
It would be nice if there was a program working in the background of Lightroom that's constantly learning. My guess is that is on the way.
Exciting times .. if you would have told me 10 years ago that my camera would shoot 120 fps, it would seem just as far-fetched as some of the AI stuff. So we'll roll with it!
Since this topic got a bump, I came across it again. Reading this post triggered me since I have experimented a bit with "Aftershoot" to see what it could do for my workflow. For my type of culling I did not find it saved me time.
It however also has an editing part, which I have used as well (although the culling was the main reason to try it out). You can feed the program photos you edited to your own likes and create a 'custom' profile, which you thereafter can use to (batch) edit your photos in your style. It can work with RAW-type files and after it made the edits, you can send the set to Lightroom where the adjustments are applied.
I think it does a good job for at least my style of editing. I run them through Aftershoot and then import into Lightroom. In Lightroom I crop my photos (for my style of cropping I have not found an AI solution) and may tweak the basic exposure a little. Never measured it, but I feel it saves me time and the edit is within 90% for most of my photos. With overcast days or very consistent lighting all over the pitch, a preset or sync action in Lightroom works just as good, but as soon as lighting gets less consistent over a match and around the pitch, the AI edit helps me a lot.
I tried out "Aftershoot" and did not dive into other programs/ software, but likely others can do it as well and who is the best will be hard to tell and change over and over with these fast AI developments.
p.1 #12 · Using AI in Your Sports Photography Workflow.
Interesting reads. I have several thoughts based on my own experience using AI and LLMs for my business (printing), which has some overlap with photography in general. I've also used it for photography on its own.
I think one suggestion I might make for those who have doubts is that, rather than avoiding it, give it a real test to see if it makes sense for you. One way I've managed to do that is by looking at previous projects that are already done and should have already had my full attention in the moment. This would allow anyone to compare their finished work vs what an AI tool can do.
A good example is my NBA archive. photos between 10 and 20 plus years old that I edited and transmitted on site. Even back then, I'd go through a folder a week or two later and see that I missed something because I was tagging during a timeout or at the quarter break. I had AI take a look at a few folders, and I looked at some of the high confidence selects it had, and I was impressed. Perfect, no, but it had a different take than I did.
It's AI, perhaps not in the same vein as this thread, but images I took 20 years ago on a 4mp 1D that were too noisy back then, or too small to really do much with? No longer an issue. A quick pass through something like DxO, and some upscaling, and images are restored in a way that was literally not possible a few years ago. I wouldn't have even found them again without a quick AI cull session, because I have, at last count, 250k NBA photos. Total archive is in the 1M range.
Back to something like culling. I think for some of us, we want to feel like it was our eye that chose the photos. But as someone who's had to work with an editor, and sometimes I've had to manage other photographers, essentially becoming their editor, we hand over the culling to someone else. So why not give AI a shot?
If you feel like you need to "do more" for the client, that's fair. I'm not saying have it do all the work, but maybe let AI do the boring stuff, leaving you more time to do the part you enjoy.
Sorry if this is long and tangential. I have way too many other thoughts. Last one: Don't think of AI tools as just being applied to your photography, but maybe the work around your photography. I've had LLMs help me write scripts (bash/python/powershell) that have all worked towards helping me do things like manage my archive, compress photos losslessly to save storage space without losing quality, and even help me improve some of my backup systems.
These are boring things, but they are also necessary. Why not have an AI help with the heavy lifting?
p.1 #13 · Using AI in Your Sports Photography Workflow.
Unfortunately for #2 LLMs are pre-trained, so you may find some way to prompt the model to your liking, but it will never 'learn' new things from you as one would think given the AI marketing.
PureMichigan wrote:
Good coversation ....
I tag because I transmit courtside/pitchside etc... so AI culling is of limited value. Here's what I would be interested in ...
1) AI denoise applied in-camera or a LOT more speedily during the edit process. That can't be too far on the horizon. At some point it would be nice to shoot at, let's say 8000 ISO, and be able to program in-camera, "make these photos look like 100 ISO coming off the card."
RIP ISO as a practical part of the exposure triangle for most forms of photograhy -- I think it's coming.
2) I'm interested in auto editing. While I enjoy the editing process, if AI can "learn" my style (and it does for the wedding folks) and apply it, I'm happy to use it even if it gets me 90% of the way there.
It would be nice if there was a program working in the background of Lightroom that's constantly learning. My guess is that is on the way.
Exciting times .. if you would have told me 10 years ago that my camera would shoot 120 fps, it would seem just as far-fetched as some of the AI stuff. So we'll roll with it!
p.1 #14 · Using AI in Your Sports Photography Workflow.
Tschanrm wrote:
Unfortunately for #2 LLMs are pre-trained, so you may find some way to prompt the model to your liking, but it will never 'learn' new things from you as one would think given the AI marketing.
I do not fully agree on your statement regarding #2. I do believe the learning of AI is more limited than the marketing wants you to believe. But for editing there is some learning.
The software I use currently provides the ability to feed a set of your own (edited) photos and thereof creates a certain editing style. When I let the software edit my photos, I can choose that personal profile and it makes edits according to the style I would have done myself. In this case the edits are based on lightroom adjustments, so if I have done the edit I export to Lightroom and Lightroom sliders are set such that the look and feel of the photo are near what I would do myself. But I do some small tweaks in some of the sliders depending on the photo.
It is more intelligent than just applying a preset since the various sliders in Lightroom are set to different values if there was a significant difference between the photos.
p.1 #15 · Using AI in Your Sports Photography Workflow.
Try not to think of it as 'learning', more like 'predicting'. I'm not sure what tool you are using, but I can certainly dig into it as I've managed a data science team in the past. Without getting too technical, the machine learning model is likely a supervised learning algorithm, trained to make appropriate predictions on new data. For visual data training you need lots of images, like ImageNet which houses 14 million images ( https://www.image-net.org/ ). Once a model has been trained, validated and deployed, as is done in production software, you the user cannot retrain the model. Only the software company can do that. As mentioned above, the production model in the software can take your inputs (images in this case) make predictions on data (future edited images), and generate new data (edited images).
The more data you feed as input, the better the prediction *potentially* becomes. I say potentially because if your edits are wildly divergent, the model will have difficulty in generating an accurate prediction (edited image).
Make sense?
jkaper1977 wrote:
I do not fully agree on your statement regarding #2. I do believe the learning of AI is more limited than the marketing wants you to believe. But for editing there is some learning.
The software I use currently provides the ability to feed a set of your own (edited) photos and thereof creates a certain editing style. When I let the software edit my photos, I can choose that personal profile and it makes edits according to the style I would have done myself. In this case the edits are based on lightroom adjustments, so if I have done the edit I export to Lightroom and Lightroom sliders are set such that the look and feel of the photo are near what I would do myself. But I do some small tweaks in some of the sliders depending on the photo.
It is more intelligent than just applying a preset since the various sliders in Lightroom are set to different values if there was a significant difference between the photos. ...Show more →
p.1 #16 · Using AI in Your Sports Photography Workflow.
Tschanrm wrote:
Try not to think of it as 'learning', more like 'predicting'. I'm not sure what tool you are using, but I can certainly dig into it as I've managed a data science team in the past. Without getting too technical, the machine learning model is likely a supervised learning algorithm, trained to make appropriate predictions on new data. For visual data training you need lots of images, like ImageNet which houses 14 million images ( https://www.image-net.org/ ). Once a model has been trained, validated and deployed, as is done in production software, you the user cannot retrain the model. Only the software company can do that. As mentioned above, the production model in the software can take your inputs (images in this case) make predictions on data (future edited images), and generate new data (edited images).
The more data you feed as input, the better the prediction *potentially* becomes. I say potentially because if your edits are wildly divergent, the model will have difficulty in generating an accurate prediction (edited image).
I did not disagree with what you stated, and what you stated here does make sense. It is however also a matter of definitions, to what one will call or label "AI".
The software predicting the way you probably would edit the photo, based on the photos you fed into it to learn from is also a form of intelligence, but not on the level of the underlying LLM. (The program I use was mentioned earlier in the thread and is called Aftershoot).
