I suspect that the built in AI is just a replacement for the external CPAI system, as it was only just released and there is no documentation any thoughts would be nothing more than speculation!
Version 6
- Thread starter Blue Iris Update Bot
- Start date
Vettester
Known around here
- Feb 5, 2017
- 1,076
- 1,075
Exactly. I have not seen where AI will do anything particularly awesome for me and my 9 cameras around my house. I am certainly open to hearing what the benefits are if someone can enlighten me.
I just looked as far as the change log and there was no entry
So before I try to write a novel on the subject, what are you trying to detect and do you get a lot of false alarms!
I started with cpai and it did a decent job of filtering out the triggers from BI (that was set to trigger on evereything). It did eliminate most false triggers (rain, snow, wind, small animals, sun/shade changes) and alert me to people, cars, and animals (I fight the deer problem!)
Then when new firmware arrived (thanks Andy) that let IVS do the same work for the same cars, people, and animals, I stopped CPAI and let the built in IVS and Onvif handle alerts, and it does a great job, at way less of a lift than cpai. So now with BI v6 offering built in AI, I'm wondering if that will be something useful for my situation or not. I'm ok with the extra setup if it does more than the IVS does (not sure what that would be though!)
Don't fix what ain't broke lol.
IAmATeaf
Known around here
I would only ever contemplate using AI if my cams don’t support something like IVS. For me with my Dahua camera IVS just works come rain or shine so I’ll carry on using it.
That was what I was thinking. (and I dd have it enabled yesterday just to see, and it alerted ok during the day...but missed all alerts at night (car driving up driveway). Not sure I had it set up right, but still, as Wittaj said, "If it ain't broke..." So IVS it is.
I'm not sure if i am the only one, but i had BI 5.9 running without any issues on a clean windows 11 out of the box install. After updating to 6.0 i had some cameras loose their connection. After installing LAN drivers from Intel the camera's worked fine again. So if you run into something similar, fresh lan drivers might fix the issue.
I have alot of wildlife where I live and enjoy watching motion events from the critters. One of the things I love about my 54IR-ZE-S3 cams is the built in animal detection. This is not available on my five 4K T cams. So, I am currently experimenting with the new built in AI from the latest Blue Iris update to see if I can include animal detection accuratly on my 4k T cams. If it works well that would be one benefit........at least for me.
I thought of another benefit of using the Blue Iris built in AI. There's always comments on how these cameras are under powered to run higher FPS and that 15 FPS is the suggested speed. Why not take even more stress off the camera and have Blue Iris AI do the detecting? I recently built a new Blue Iris PC and felt like I was hardly tapping into the new hardware. I'm happy to let Blue Iris AI do the detecing on some of my cameras. I want to use some of that 32GB of RAM!!
Last edited:
Vettester
Known around here
- Feb 5, 2017
- 1,076
- 1,075
Yep that's what I'm going to try. I'm currently testing a few AI things out with the camera here inside my office to get a feel for things.
My 4K-T's now detect animals. 
Cool! Now THAT would be a use for AI for me! I have a few different animals come through my yard regularly that I pick up with simple motion detection. Of course, that means rain and snow storms trigger continuously at night on those cams. How hard of a hit is the built-in AI going to do to CPU utilization? OR-- how beefy of a graphics card would I need to be efficient?
I'm using the integrated Intel UHD Graphics on my i5 14600K CPU. I don't even have a dedicated graphics card in my BI system. My CPU usage is the same as before. My RAM has gone up slightly but so far I only have two cameras switched over to AI.
Seems to me that there's an advantage to having the AI on the camera, where it can focus in on movement. Off camera AI has to scan full frames all of the time.
biggen
Known around here
- May 6, 2018
- 2,861
- 3,274
Of course. But on-cam AI is limited to what the camera manufacturer gives you. At least with doing AI externally, you have more options with more targets (e.g. Blue Onyx, CPAI, Frigate, etc..)
Todays update (12/22/25) lowered RAM usage for me with six cameras using onboard AI.
EDIT: Scratch that......RAM usage back to where it was before. Probably just the rebot during the upgrade.
EDIT: Scratch that......RAM usage back to where it was before. Probably just the rebot during the upgrade.
Last edited:
Has anyone tried the Coral M.2 accelerator with the Blue Iris on board AI? How has it worked with the Coral accelerator? My frustration with the Coral accelerators was that it used the full Coco name set. Thus, I was getting airplanes, my garage was a train, my mailbox was a toilet, I got hundreds of persons in my lawn, etc. So if you can limit the objects detected by hyphenating them, it is probably time for me to upgrade to BI ver 6.
Tim_Ettinger
n3wb
I'm still watching this thread and waiting. I'm not going to switch over to BI 6 before Q1 2026.
My application is not quite mission critical, but I have to answer a lot of questions several times a day if things aren't working right. I have 31 cameras, and 15 fps isn't good enough. My hardware is OK, dual x5690 (old high end Xeons), 288gb RAM, Quadro RTX 4000 8gb, SAS, two GBE networks passing video traffic. I think I would be ok to make the switch and have a couple days to monitor and tweak. However, due to security concerns, I'm going to be adding at least 15 more cameras. My network won't choke on it, but the struggle for storage is real. There are some cameras I can justify not continuously recording, but others are 24/7. I have basic motion detection to trigger some of the cameras now, but several of the new ones are going to be looking outside at traffic, parking lots, and monitoring a tree line. I really would like the integrated AI to work well for me because of financial and bitrate budgets. I'm not going to get a lot of money for top tier cameras, and I'm not getting a higher level LTO storage system. Cameras that watch nature eat more bandwidth and generally stay pegged at CVBR limits.
I know I've asked for advice here recently before, and was a little resistant to some of the suggestions. The constraints I have aren't just me being stubborn. A few months ago someone dropped dead, and there wasn't a camera watching that spot directly. The police were good with all the footage they could see, but there were a lot of questions and suggestions of inadequacy on my part after the fact. More recently there is a transient people problem, and it exists just beyond the trees, and I need to have adequate eyes on them.
The basic question is, is Version 6 totally ready, will take less than a day to get fully up and ready, and will it eat up too many resources for me to handle as I add more cameras? I have done a bit of math and with Version 5 I should be able to handle 64 cameras, although I may need to connect up to a 10gb ethernet port. I'm sure someone out there can relate in hardware, and with greater experience.
My application is not quite mission critical, but I have to answer a lot of questions several times a day if things aren't working right. I have 31 cameras, and 15 fps isn't good enough. My hardware is OK, dual x5690 (old high end Xeons), 288gb RAM, Quadro RTX 4000 8gb, SAS, two GBE networks passing video traffic. I think I would be ok to make the switch and have a couple days to monitor and tweak. However, due to security concerns, I'm going to be adding at least 15 more cameras. My network won't choke on it, but the struggle for storage is real. There are some cameras I can justify not continuously recording, but others are 24/7. I have basic motion detection to trigger some of the cameras now, but several of the new ones are going to be looking outside at traffic, parking lots, and monitoring a tree line. I really would like the integrated AI to work well for me because of financial and bitrate budgets. I'm not going to get a lot of money for top tier cameras, and I'm not getting a higher level LTO storage system. Cameras that watch nature eat more bandwidth and generally stay pegged at CVBR limits.
I know I've asked for advice here recently before, and was a little resistant to some of the suggestions. The constraints I have aren't just me being stubborn. A few months ago someone dropped dead, and there wasn't a camera watching that spot directly. The police were good with all the footage they could see, but there were a lot of questions and suggestions of inadequacy on my part after the fact. More recently there is a transient people problem, and it exists just beyond the trees, and I need to have adequate eyes on them.
The basic question is, is Version 6 totally ready, will take less than a day to get fully up and ready, and will it eat up too many resources for me to handle as I add more cameras? I have done a bit of math and with Version 5 I should be able to handle 64 cameras, although I may need to connect up to a 10gb ethernet port. I'm sure someone out there can relate in hardware, and with greater experience.