Hey Prusa AI team, why not use data from slicer as a target and camera as input and compare both at some point instead of catching general printing issues? Training could be done every time a printer prints.

Nice format and i also liked no 10 min before show that i need to skip :) Its nice that you bring more ppl from your company. You have so many talented people.

The video looks way better but there is something wrong with audio (I can hear the person talking in other mics too).

Great SHow. Thanks Mikolas. Cheers.

Gnah. Missed it. I love the new format, but please do them on a schedule again. :)

Build model that monitors a print and detects tight/loose belts (and other hardware issues). Partially-clogged nozzles would be another game changer.

Wow, the OpenScad topic was a really interesting take on 3D-AI-generation. And it is also interesting to hear what you harvest from your print data. Nice Podcast Format :)

Very nice discussion 😊

When I was working at Prusa and we were collecting data/pictures for the first machine learning intended to be used on the print farm (not related to this one, it was 5-6 years ago), we actually "broke" 5 printers (loose belt, loosened screws, not working fan...) to collect as much data as possible because, as mentioned, the printers are just too reliable :)

This episode is really interesting. I think that the AI should get the a model for blob detection, since blobs are the worst threat to extruders. Great job Prusa team, looking forward to this feature in the future ❤

That was probably one of the better discussions about AI usage I've seen. I view AI as a tool, though I haven't used it much because I don't like some of the ways I need to use it (I've been a big proponent of "don't make me work around the tool, make the tool work for me" for most of my life). But there are others who use it to substitute their work... code quality isn't as high and style is different, art that grabs the eye because of detail... but seems to lack consistency and falls apart the more you look at it, text that looks fine at a glance but reads weird and doesn't work at all when spoken out loud. They aren't people who are "unskilled", it's people simply looking to do a task faster and expect AI to be at their level but be faster. Then there's the people who don't want to pay... and this is a dramatic group because some simply can't afford (or know where to find) a developer/artist/etc. but the worse are those who can afford the tools, hiring people, have time to learn it themselves, etc. and choose not to. Even going to the point of proudly announcing they didn't need to pay someone to do that work. And that's all before the training debacle that I think is gonna make or break AI. So many things that come out in tech are clunky at first (outside of very specific demo cases), they shoot for the moon... and every single one of them miss. So they start judging success by how close they got. Eventually, some people put time and effort into the tech, work their way through ethics and legal, and become a standard base for others to compete with. We're past the weird tech intro and at the shoot for the moon stage. In 1-2 years there will be laws either calling image/video/text scraping "fair use" (at least for US law...) or it will say "nope, you need to remove it and you'll be charged every day it's still in the model" and you're gonna watch OpenAI and others pull every API and tool from public use and all the repercussions that will come from that. But what comes next will be the important stuff... better interfaces, more options, local models that train on your code and art as you do them, more price options from free to expensive, tools that have permission to use copyright data, and more. That's when it will truly enter mainstream instead of people just talking about it hitting mainstream. And as my old roommate once said "managers think AI can replace workers, but AI is best suited fro replacing managers". We'll find artists making art, maybe not posting it because of risk of scrapers, but they'll be doing it. Coders will be coding. Truck drivers will be driving. Cooks and wait staff will be attending tables. But we may have instructions and guidance coming an AI manager/supervisor. 3D generation and CAD will be there, but I'm really hoping I can say "connect these two pieces with clearance in this area" and it does it.

Analysing what's happening with the whole print is good to get a general idea of whats happening, like whether it has failed or not. But I think you can see many more issues when you look at the nozzle very closely while printing. Especially things like over-/underextrusion or stringing that might not be easy to measure any other way. So maybe putting a tiny macro camera on the extruder and analysing video would give such an AI even more value. I'm imagining a printer that takes sensor data (especially from the load cell) and some image or video data and maybe even the original model into account and tries to adjust settings dynamically to compensate for any deviations between what is actually happening and what the sliced file and/or original model were supposed to look like. Once it finds the optimal settings for the specific machine, filament and environment, it could store that as a profile to use the next time.

I enjoyed the new format the set looks nice!

Great format! Please do an episode about the printer simulator for firmware testing. Sounds super interesting! And maybe try lavalier mics? Sound quality was great, but the occasional phasing and varying distance and therefore loudness levels were irritating.

I'd love to see more blender tutorials from Peter :)

The AI failure detection is super cool and I'm really excited that you guys are working on it! Can't wait for it to roll out to the rest of us. I would also love to see some more options for camera setups, particularly on the Mini/Mk4. With a Mk3 you're already running a Pi for Connect so adding a Pi cam is simple and integrates very well. With the Mini/Mk4, since Connect is already running, the only options are to set up a whole separate device (Pi, smartphone, etc) which also needs to be plugged in, moved with the printer, etc. It would be really nice to have a more integrated solution for these newer machines.

it would be great to detect failed supports and to stop printing that support but continue everything else. I was printing TPU yesterday and one of the supports fell over (Never use snug towers with TPU, organic is the way to go) The printer made a little mess but was able to finish. It would have been flawless if it abandoned that support.

Any consideration to adding scarf joints to PrusaSlicer?

Love the content, but miss the aspect of the old "Live" format that allowed engagement between the community and Prusa.

Oh jeez you made a MK4 sound at the beginning of the video and I ran over to the print farm thinking there was a problem with one of my prints...😡

Prusa printers are by far the best on the market and wont burn your house down because they check everything before releasing to the public they do have some small issues but are not a safety issue and top notch hardware i just wish they were more affordable so more people would buy instead of the others that give you headache after headache.

how have i never heard of this podcast lol

AI finds use in various fields and for making 3D printing more user friendly and less baby sitting, it will surely find way more interactive features in the future

Great podcast, but next time please check sound and make it better ballanced.

If we're talking AI can we fix the fill bed feature. Most times it can't figure out how to do this and add the most amount of a part. I can do better adding and then auto arranging (even this isn't very good and puts parts outside the build plate).

It would great if one part of many is detaching it abort only this part and prints the others to the end.

YAY prusa

I like this

The beep at the beginning and end kind of freaks me out. I'm printing with brittle filament that sometimes snaps. So the printer beeps a lot. That's the same beep! (of course it is since you guys programmed it). So I think maybe my printer needs another re-load.

Water marking - I had an idea to hide it inside the model at layer 3 or 4 you should not see it underneath the model however it will slice, maybe someone smart will make a nightshade for 3d models.

Fascinating - always feel free to collect my data/images :).

wait jo has an @allenpan shirt lol

Prusa connect can be better if the user can change the finised print statick. Like printfailers or just i stoped because i dont need for this...

i hope this year can afford prusa mark 4 amieeen

Another class should be blobs of doom. When the print gets caught by the nozzle and gets pushed around, so the extrusion starts to engulf it.

Prusa printers are not making Spaghetti, at least not mine, which I have since June 2019 and printed hundreds of parts !

for anyone who watched any update on the mmu for the mk4 besides the articles?

Palec dolu za češtinu. Bohuzel marketing je vetsi do světa než u nás. Anglicky každopádně rozumim vše. 😊

Too bad Prusa printers don't come with cameras to take advantage of this new AI detection.

Not to be a downer, but why are we worried about AI Image Print Detection if you have no cameras on your printers? I love my MK3s, MK4, and Mini but the fact that you are still selling printers without built in cameras is odd for the price of these printers and the fact that all your competitors have them for cheaper machines is surprising. Should be standard in 2024 at the price point of your hobby grade/small business 3d printers. Also, the integration for adding cameras is in "beta" stages at best for Prusa Connect. Shouldn't have to hack my Wyze Cameras to enable RTSP or make RTSP cameras from Pi Zero Ws I luckily had around the house. Then you have to have a PC running OBS 24/7 to stream the RTSP to a virtual camera for a Prusa Connect Camera browser window that is always open. Also you have to multiply that by each machine you want to have a camera on, it is just waayyy to much hassle to go through to just monitor my prints remotely. This is coming from a fairly technical person. Imagine someone that doesn't have a computer/technical background.

nice

E

AI for slicing.