Wyze AI has become sentient! Did Wyze merge with Cyberdyne?

I received a concerning notification that a person was detected in my master bedroom.

Naturally, I was worried because no one is home but when I checked out the video I realized it is much worse than I thought!

The “person” was none other than my Roomba Smart Vac!!

This is terrifying. The Wyze AI has apparently become sentient and now recognizes a fellow AI device as an actual living Person!


Shhhh they’re internet connected. They’re reading this as you type! :joy:


:joy: As funny as that is, chances are that the robot vacuum was just what triggered the motion event. The way the AI currently works is that it doesn’t just analyze what object caused motion, but it will analyze everything in detection area and if anything resembles a person shape, it will trigger a person detection even if that thing wasn’t moving. So it’s more likely the Vacuum was just the event trigger and the actual person object is something else.

Another way to test this and see that this is the case is, the next time this happens, right after that, while everything is still in the same lighting conditions and placement, remove the vacuum and have something else like a laser pointer trigger events again and even though the vacuum is gone, you are likely to still sometimes get false person detections, because it’s probably fixated on some other object seeming to be person-shaped. Once you confirm that, you can start messing with things in the room one at a time to see if you can mess up the detection. Once you do, you’ll know the cause and can use some counter-measures to resolve that so it doesn’t happen again.

Since it is your bedroom, it is understandable if you’d rather not share any screenshots, but often we will have other people show us a screenshot of what the camera views and we can often tell people what is likely being fixated on by the AI as a person when it isn’t. Still, you can figure it out with a little testing yourself as described above.

Having said all that, I like your story better, it’s much more entertaining! :joy:


Fascinating! :nerd_face:

Or just turn on your detection settings, that little green box thing, and it’ll show you where it is picking up the motion :woman_shrugging:t3:

Good thinking. Unfortunately, the Wyze AI isn’t currently setup to only analyze MOVING objects for object detection. If there is ANY movement anywhere onscreen, it will actually analyze the ENTIRE camera view, including everything that is stationary. Then, even if a stationary object is humanoid shape in some way, it will then label the event, and whole screen as containing an object somewhere.

I have actually tested and proved this because during an AMA I got a Wyze Dev to publicly offer to add one of my V3 cameras to a program that would highlight in blue what the AI detects as an object, and the green square would still highlight motion. They are not always the same highlight.

This is a common false assumption people make. The green highlight is only showing motion at a fixed sensitivity level, and has nothing to do with what the cloud AI has determined was a detected object (person, pet, package, vehicle). Having said that, Wyze has reportedly said they DO now have a prototype that will only analyze moving objects, but it is in testing and has not been launched publicly. We don’t know when or if it will.

This is one of the things I love about Scrypted (over Frigate, much less cloud ai of any kind). Here’s the behind-the-scenes debug of a notification I got today, the picture of the bird in the middle is the Thumbnail of the notification I received on my watch and phone. It’s easy to get confused about the order of operations here: One could assume that motion was detected, it analyzed everything, found a bird, zoom-cropped it, and sent that.

But it’s the opposite. Motion was detected, the bird caused the motion, and ONLY a small area around that motion is analyzed (defined by “pixel dilation” settings and pixel “blur radius”, aka, how many pixels need to move, and how many pixels around the movement do we want to look at, etc).

The compute power to analyze this entire 4k image, every pixel, would be enormous. But Scrypted is smart, it uses resources wisely (no pun intended), and it only uses processing power analyzing the thing that moved, This is what allows it to run state-of-the-art object detections locally, on a very tiny computer that doesn’t even have a fan.

Wyze has reportedly said they DO now have a prototype that will only analyze moving objects, but it is in testing and has not been launched publicly.

This is open source, Wyze, here’s the code: scrypted/plugins/objectdetector/src at main · koush/scrypted · GitHub

ffmpeg / tensorflow / python / typescript

But, the thing is, if being open source meant that “anybody can just do it”, then everyone would do it. You can have the code, but you have to have the talent to integrate it. And wyze doesn’t have that talent. They pour every bit of their talent into hardware, releasing more and more and more models, more things no one asked for, more “upgrades” that are bare updates, more product launches, more shiny things, anything to show we’re advancing… And not giving any talent and resources to make any of that hardware do stuff well.

90% of the things people complain about on this forum, are commercially usable for free on GitHub.

I love that! Thanks for sharing. I believe I’ll be setting up my server within the month. It’s probably not as fancy as yours:
Beelink SER5 MAX Mini PC, AMD Ryzen 7 5800H(up to 4.4GHz) 8C/16T, Mini Computer 16GB DDR4 RAM 1TB NVMe SSD

But I am still fairly excited to set it up and start loading some fun dockers. :slight_smile:

Whoa that’s way fancier than mine (I have some on a Mac but most are on a Beelink, which is honestly just as good) - HOWEVER - please look at the Buyer’s Guide Server Hardware | Scrypted Docs and get on the Scrypted Discord Scrypted ASAP while you have time to return that thing. For that much money get a used M1 Mac Mini on ebay. You will get far BETTER performance for 3x less getting the beelink s12, or anything with the Intel N100 chip. I don’t know enough, or anything, or CPU architecture to understand this, but the developer of scrypted is a CPU architecture savant, and things have been VERY optimized for the N100 (and apple silicon). The N100 because it’s cheap yet apparently crazy powerful when you use it in the right way, and apple silicon , cause obviously nothing has that much AI on a SoC. But honestly, there’s been so much focus on keeping things affordable for people (that’s really important for the developer) that I think the N100 (with ubuntu/docker, or I hear ProxMox is even better), might be top tier. Now, I know you have 9k cameras, so, maybe go used M1 Mac Mini? Either way, beelink s12 will work far, far better than what you’ve got there. I don’t know why, I’m not that smart, go ask the man himself on Discord.

Also, while you’re asking, everyone will say “well, if you’re using ONLY 2.4Ghz, wifi cams, no set up, no matter how much you spend, will work super well”. And they’re right… There’s a reason no “real” commercial security outfit will add a wifi cam to any home security system, at least not a 2.4Ghz wifi cam, but generally none. So get a few cameras that are PoE or at least connect to ethernet, and start your self-hosted journey there. You can still have the zillion Wyze cameras on your existing setup. But you will be gravely disappointed with that AMD either way.

Glad I stayed up late enough to catch this!! Also, the buyers guide is slightly outdated in one sense – the N100 integration is now so good that a Coral TPU could honestly slow you down, unless you have 10+ PoE cams. The beelink s12 is honestly that good. That’s what the live cams use is app.scrypted.com/demo and obviously the video scrubbing there is almost perfect.

1 Like

@carverofchoice and just emphasize, read ALL of the docs and recommendations super carefully first, before going into Discord. You will literally get the best support ever on the Scrypted Discord, but, understandably, it’s little bit frustrating when people come in without reading everything first. You can’t compare the whole thing to a company, it’s not a company like Wyze, you have the founder helping people literally 18 hours a day, and a few devs helping, but it’s mostly him. And he gets thousands of legit questions a day, so when there’s a question where the answer was in the docs… It’s obviously not going to the top of the list lol. The buyers guide is not so much “recommendations”, it is, but really it’s “if you want this to run the way I wrote it, you have to have to use the stuff it was written for”.

Too late for that. I bought it on a big Black Friday discount (so it was still pricey, but not as much as it’s showing now), just haven’t had time to set it up because of end of the year/start of the year work crunches (I own my own business and contract for some others).

Still, I will keep that in mind for the future if I consider expanding. I almost never consider Apple since I don’t have anything else in their ecosystem, but I should they make amazing chips, even when the chip is used in a different system (I almost bought an ODroid for Home assistant that was using some kind of Apple chip that was outperforming a lot of other mini-PCs). I just decided to go with this for my first one, and then I might get other cheaper dedicated units for some other things later. I’ll consider an M1 Mac Mini in the future.

Good be true. I looked at the B12. I went with the SER5 Max because it had twice the cores and 4 times the threads, 4.4GHz vs 2.4GHz, more ports Including USB-C), better Bluetooth, etc. I figure if I want one dedicated to fewer dockers then I can get a cheaper B12 later (or maybe a Mac M1 Mini now)

I will do that…though since I’m stuck with the one I have now anyway, I’m going to make the most of it. :slight_smile:

Quite possible. I’m not too concerned. Worst case scenario, I just get another one for other stuff. I’m just starting with home lab stuff and I figure I’ll make some mistakes, financially and otherwise, but after a couple of years of stalling debating what device I should get, I decided I’m going to just started with SOMETHING and then I’ll learn a little bit and adjust for the long term. :slight_smile: It’s better than still being stalled out completely for another 2 years. I gave myself a deadline of Black Friday and just pulled the trigger so I’d stop procrastinating.

No worries, totally understandable. That’s part of what takes me so long, I usually search the docs for my questions first, then ask Multiple AI’s :joy: AI has been a life-send. I use it constantly with all tech stuff.

Thanks for the suggestions. I am very excited to get started when work stuff calms down a bit. I might do Scrypted as one of the later dockers on my list just to make sure I have enough spare time.

Priority 1 will be learning Proxmox. Then I will move my Audiobookshelf docker over there. as well as a few other dockers I already have set up on my Synology. The main thing is to make it so no dockers are on the NAS, and only use the NAS for storage. Then I’ll start a few new dockers like Scrypted. :slight_smile:

I am excited to turn a bunch of my Wyze cameras into personalized detections and be able to make them do more complex automations/routines. I am sure I will also eventually do some PoE cams, etc. But I don’t look forward to wiring and might have to negotiate that with the wife…so it may not be in the cards until we get a new house. :wink:

I appreciate all the advice!

turn a bunch of my Wyze cameras into personalized detections and be able to make them do more complex automations/routines

Oh, if you just want to do this, primarily, then that thing will work just fine! When you set up object detection zones in scrypted, there is a setting called “observe” that you can just use for automations, basically.

Priority 1 will be learning Proxmox.

If you’re using Proxmox, you won’t use docker (at least for Scrypted, and probably most things). You’d run it in a LXC. It’s already a virtual container. Which is the point of docker, to create a virtual, walled, container within a larger environment doing a bunch of other things. On Proxmox, literally everything your run is already a walled off virtual container, so it’s redundant. I can’t speak for every platform, but for Scrypted, the install script is written for LXC, not docker :slight_smile:

1 Like