Opinion: AI Subscriptions are coming, and it shouldn’t surprise us

[justaCanonuser]

Regarding power consumption: it is sn often overlooked fact that cooling is more than 50% of the total power consumption of a data center, simply because it's not 100% efficient.

I read yesterday that there are rumors that Microsoft is planning to build a data center that need power from no less than 3 nuclear power plants
Yes, for A.I.

Yepp, as long as neuromporphic computer architectures still are scifi, the growing use of AI will cause a maybe exponential explosion of power consumption. The bottleneck is the classical silicon hardware that is on a miniaturization path yet but isn't really made smarter.

 

[LDS]

I really don’t see this working with current 5G technology. The latency is too high.

While 5G has some low-latency features they are not thought as general use ones - they are reserved for specific "critical" tasks. Moreover, unless Canon builds a lot of sites worldwide to put processing power closer to the user, there's still the whole internet latency to take into account, once the data leaves the 5G network. If there is one when on a safari or near the Poles, of course... and maybe combat drones will start to target specific 5G signals (using their AIs to spot them, of course).

But I agree with you that running using models doesn't require the processing power to create them - and there now are specific "AI processors" to run models on end-user devices. So probably subscriptions will download models (or activate them) locally - what they are used for is the real question.

I don't think generative AIs have much use in a camera used to create images form their lenses - unless someone wants to replace the Moon with a better looking one like Samsung already does, or their partner with a better looking one - but there are other kind of AIs that could be useful - for focusing, controlling DoF, stacking, HDR, IS, T/S etc. - for Canon the real question is what is in the default camera configuration and what should go into a subscription - and how willingly are users to pay for such features. It they get the wrong balance, it can seriously hamper sales.

 

[john1970]

I guess if Sony can charge $149.00 for custom grid lines that Canon and Nikon will eventually follow suit. As long as it is not a monthly subscription service I will be okay. In my new car I pay a $10 monthly fee so it can communicate with my phone. I do pay it, but I feel that when you buy a new car, services like this should be lifetime or at least for the first 3 yrs like the bumper-to-bumper warranty. Such as life....

 

[Dreysi]

I have no issues paying a modest fee for additional post-processing AI features or major firmware updates, but if I have to pay a subscription to Canon to have a functional camera I will have to research pitchfork suppliers.

I'm sorry your pitchfork subscription has expired, how would you like to pay to unlock it's features?

 

[SnowMiku]

I'm not interested in subscribing to have my camera connect to 5G or Wifi to use AI features. If a picture is enhanced by using large model AI - the type that uses many other pictures to enhance yours, is it even photography anymore?

I agree that we need a system where we could easily check if a photo uses any large model AI or not.

Paying for firmware updates, I can understand why some would be happy to pay for extra features if needed, but the R5 got many firmware updates and extra features for free. Keeping the firmware updates for free will keep the camera competitive with other brands for longer.

 

[chik0240]

it's a lot for sure. a company that i work for has patents on tackling the issue. However, it's one thing to invent it.. it's another to convince Microsoft and others that this is a viable technology.

the waste heat from AI is going to be absolutely bonkers.

this is very counter intiuitive for the age of environmental protection, and most AI end products are pretty much useless, just for fun type things... a good idea for the next generatino

 

[LDS]

In my new car I pay a $10 monthly fee so it can communicate with my phone.

So you're paying monthly for something that was previously part of the car price. The problem is exactly that buyers is accepting it, even if in some cases it backfired. I don't know what Canon has i mind, but if some features that always had been part of the camera set are becoming subscription based just because it's easier to extract money from customer that way, well, the future doesn't look good. Reminds me about the fees you had to pay in medieval times for most thing people were doing.

 

[John Wilde]

That sounds like a CR0 to me.

 

[john1970]

So you're paying monthly for something that was previously part of the car price. The problem is exactly that buyers is accepting it, even if in some cases it backfired. I don't know what Canon has i mind, but if some features that always had been part of the camera set are becoming subscription based just because it's easier to extract money from customer that way, well, the future doesn't look good. Reminds me about the fees you had to pay in medieval times for most thing people were doing.

I agree, but what else can I do. I do like the ability to lock/unlock and start my car using my phone. I once locked my keys in the car and it was a major pain in the neck; luckily I was locally so I called a neighbor to pick me up so I could get another set of keys. I was very very lucky.

 

[LDS]

I agree, but what else can I do. I do like the ability to lock/unlock and start my car using my phone. I once locked my keys in the car and it was a major pain in the neck; luckily I was locally so I called a neighbor to pick me up so I could get another set of keys. I was very very lucky.

If it is a service that relies on an external infrastructure - which needs to be maintained - a subscription makes sense (even if 10$/month for such simple connection is on the "very expensive" side). If it doesn't, it's just gouging. Still any car that lets you lock the key inside is ill-designed (and you can still lock both the phone and the key inside). If Canon sells rents models that are kept up-to-date a subscription can make sense - although while in the beginning users could see big improvement when the technology is new, soon enough they usually becomes smaller and smaller.

The problem might be what happens when the camera is no longer supported, or you wish to stop paying. In what conditions does the camera keep on working? In the software world some software just loses updates, and you are "locked" into one of the versions you paid for, other like Adobe neuter the software that loses most functionalities.

Until we see what Canon really does it's too early to light the torches. I'm sure inside Canon there are people who would like to see nice stable revenues flowing in, but they are also worried about finding the right balance between new profits and unit sold.

 

[fiendstudios]

Something that I would like to have and could pay to have done is automatic culling images that I take so when I come home I have images that are the "best" ones already picked out by AI. Where people smile, don't blink, are in focus etc.
Could this be solved by AI in Lightroom etc? Yes, but even better if its already done when I get home or to show a client at the back of the camera etc.

 

[P-visie]

Something that I would like to have and could pay to have done is automatic culling images that I take so when I come home I have images that are the "best" ones already picked out by AI. Where people smile, don't blink, are in focus etc.
Could this be solved by AI in Lightroom etc? Yes, but even better if its already done when I get home or to show a client at the back of the camera etc.

Canon launched an AI powered IOS app for photo culling (link) in 2021 and later provided a Lightroom plugin. The subscription was $2.99 per month or $15 per year.
I guess both were culled by Canon since the app and plugin are no longer availabe.

 

[Uneternal]

There is no way people would have a camera that needs to be connected to the internet to get all features. Won't happen in a world where people even went mad over the touch bar of an EOS R.

And it's also not "a matter of time" - anybody who will try this BS will miserably fail, because its not practical or viable for the customer and therefore there won't be a demand. Canon is currently testing it with DPP and they probably already found out.

A.I. will certainly come, but by adding faster, A.I. specialized processors and more system memory to the camera. In return we get less battery again, like we also got when we went mirrorless with its advanced tracking.

 

[mxwphoto]

Not sure what the plan is, but current camera hardware specs certainly cannot support the types of AI processing being discussed. If it is on board, then you need modern cell phone processors (snapdragon 8 gen 3 or better) and if it is via wifi then a very solid and speedy symmetrical connection ala fiber. Either way the battery drain would be immense and the heat generated would make current camera overheating problems seem like a spring breeze.

It would make much more sense if Canon expanded on their DPP with AI to actually be competitive with lightroom and charge for that than have it on camera.

 

[David - Sydney]

it's a lot for sure. a company that i work for has patents on tackling the issue. However, it's one thing to invent it.. it's another to convince Microsoft and others that this is a viable technology.

the waste heat from AI is going to be absolutely bonkers.

And from bitcoin mining previously/now. One reason to have data centers in cold climates and use natural cooling.
Let's not forget about electricity generation waste heat that needs lakes to cool down. A lot of inefficiency in the total ecosystem.

 

[David - Sydney]

I could see off-camera processing being a thing in 5-10 years, I really don’t see this working with current 5G technology. The latency is too high. Think about how oftentimes events at large venues result in crowded networks and slow internet. The last thing any of us want is shutter lag due to network conditions. I’m sure they’d build a fallback to built-in autofocus, but if its significantly worse than the subscription version I’d be livid.

In theory, 5G technology offers an extremely low latency rate ie the delay between the sending and receiving information. One "big" selling point of 5G was the reduced latency enabling over the air self-driving car/video games etc with appropriate edge compute/storage (which hasn't necessarily been implemented).

From 200 milliseconds for 4G, we go down to 1 ms with 5G but typically <50ms.
1ms is still 1ms but much faster than the average human reaction time of 250ms.
The issue you see is speed and not latency.
https://www.ericsson.com/en/blog/2022/8/who-cares-about-latency-in-5g

Yes, it takes a lot of money to build and train models, but it doesn’t necessarily take as much power as folks would think to RUN models. Various image recognitions models run quite well on economy hardware like Raspberry Pis. I’ve also got an LLM running locally on my MacBook Pro, and it performs quite well without consuming too much battery power.

Completely agree. LLM generation - compute, power and data input are the big issues.
On-device AI compute is much smaller as Apple's neural processors seek to optimise as part of their system-on-chip architecture. Apple (TSMC) can use the latest low power chip manufacturing which Canon couldn't possibly access or embed.
They could use a discreet chip (eg Qualcomm) though they will be hard to come by and expensive but their only real option for low-power in-body imlementation.

 

[David - Sydney]

However, if they tried to charge for say that pixel shift 400mp thing, that would not be cool. I don't think it was very well done, but there could have obviously been technical limitations with the hardware that prevented it from being better. If they go this route, they will have to develop hardware with the future in mind.
It could also help Canon become better at software, there have been *cough* a few misses in the past.
I will add.. no monthly BS... I think a lot of us are tired of that model.

One-time paid upgrade for specific feature = okay but Canon will need to manage an "app" store of some description.
Recurring subscription (SaaS) = not okay
Bug fixes are mandatory of course but we can't assume new features added into existing devices unless announced as part of a new body release.

Sony tried the app model many years ago but ultimately removed it. Easier to bring in a new hardware device and include the feature there as the incentive to upgrade.
Apple/Samsung etc have a much harder time now to get people to upgrade their phone as the slightly previous models are still good enough for most people. I will upgrade mine this year after 4 years. They use camera improvements as some of the biggest improvements which is slightly ironic for our discussion
The app store is a massive revenue contributor to Apple without significant input cost. It changed (starting with iTunes) from being an almost pure hardware revenue company to a hardware and SW company. Canon doesn't have that experience compared to Sony for instance.

 

[David - Sydney]

So you're paying monthly for something that was previously part of the car price. The problem is exactly that buyers is accepting it, even if in some cases it backfired. I don't know what Canon has i mind, but if some features that always had been part of the camera set are becoming subscription based just because it's easier to extract money from customer that way, well, the future doesn't look good. Reminds me about the fees you had to pay in medieval times for most thing people were doing.

Tesla is a good example. If you want autonomous driving (I don't) then you pay for it (one time).
Previously, for certain models, you could unlock additional acceleration performance for a one time fee (definitely cr1ppled).
Tesla are also adding new features and improving existing features via OTA firmware upgrades. 2024.14.3 improved the layout of the screen for instance amongst a raft of other features. Most of which are great.

A big change in thinking from the "get used to it" firmware of previous cars I have bought. I didn't buy a Ioniq 6 mostly because you can't turn off the overspeed chiming except doing every time you start the car. I have no problem if it worked but it doesn't locally where we have school zone that only operate during certain day/hours and only during school terms.

 

[David - Sydney]

I agree, but what else can I do. I do like the ability to lock/unlock and start my car using my phone. I once locked my keys in the car and it was a major pain in the neck; luckily I was locally so I called a neighbor to pick me up so I could get another set of keys. I was very very lucky.

Yep, using your phone is excellent for my Tesla.
I also have a key card in my wallet that I use when going diving as I leave the phone in the locked car with bluetooth turned off. You can also buy a ring for your hand which acts like a key.
One time, I didn't lock the car with the car (phone with bluetooth turned off) and my wife got the notification and locked it remotely using her phone.
Worst case, my wife could unlock the car remotely for me or temporarily authorise another phone to drive the car.

 

[David - Sydney]

my AUD0.02
Subject recognition = awesome use of machine learning AI
but I am struggling to find a use-case for onboard genAI. I get the usage on a phone where post-processing is common from filters to bunny ears to sky replacement etc but that doesn't happen on camera bodies and unlikely to in the future.

It looks to me that a lot of the improved phone images taken are based on taking video and building images using algorithms for low light/HDR and stitching panos. That is a huge step on from pressing a shutter button but now common for phone users.

Options in DDP for AI denoise/upscale/panorama merging/focus stacking algorithms etc make sense but cloud (requiring connectivity) processing doesn't seem okay even with improving coverage and decreasing roaming costs etc.

Clearly users are getting more used to paying for SaaS and Adobe is a big winner here. The decreased overall cost of LR if upgrading a license every 2 years and including PS is a no-brainer even with some minor price increases since the beginning. Being able to access your images in LR even if you don't pay the subscription is comforting.
I have been convinced to pay for certain SW now eg Topaz AI de-noise and PTgui as it improves my workflow and significantly reduces the need to upgrade bodies or lenses.