This reminds me of a funny story a friend of mine told me. He does real time projection mapping (like when they project vines growing up the DisneyLand Castle and stuff like that) and he’s been fighting off a very regular, very boring job.
Basically he’s been hired to project rooms, like kitchens or living rooms on a cylinder someone is standing inside of while they are filmed performing a repetitive series of mundane actions. Like they’ll project a series of different kitchen backgrounds behind someone and they are instructed to pantomime cracking an egg over and over again, or vacuuming, or whatever. The talent does this over and over until they think they’re good and then they move on the next scene.
They’re doing this with multiple cylinders per location, apparently at multiple locations around the country for 8 hour shifts in five week chunks.
He said it is mind numbingly boring for everyone involved PLUS it’s not local so he has to stay at a hotel for the week for 5 weeks at a time. If he has anything else going on or even might have something else going on he turns it down but they keep asking and the money isn’t bad. They told him/didn’t officially tell him it’s for training AI.
We’re really heading for the most boring dystopia ever.
I was thinking the projections would be some kind of robbery prevention for people’s second, third, eighth homes: project a guy cracking eggs in your kitchen while you are away so people think you are home.
This is pure code but it is quite easy to run if you know how to run conda environments…Although I’m pretty sure there will be a comfyui node for this.
Important thing is how vram heavy this is, hopefully they release a fp8 or quantized model in order to run the inference on lower vram usage, otherwise you won’t be able to run this on standar resolutions, you can see there that for a 720p you need 33gb vram. So think of 2k or 4k plates and higher… You’ll need at least 48gb vram gpu.
for those of us who aren’t quite sold on the benefits of AI, I recommend Ed Zitron’s podcast “Better Offline” where he looks at different aspects of the Tech industry and is quite critical of the whole tech-bro economy
So far this is the only decent AI video I’ve seen everything else looks like it was done with a toon shader, but check this out I guess the entire thing was done in AI in 3 weeks https://youtu.be/VqLWWYfCEbI?si=vRVIDxfdhDPceJwK
Looks good. The question is - is this straight up AI material, or did they use AI to generate assets but then did a lot of work afterwards to assemble it into a clean looking ad.
What % of work was AI vs. traditional edit/comp/vfx?
I think what we’ve learned on all uses of AI - it’s great as a companion tool to generate certain elements and can thus speed up the end result. But it sucks in producing delivery ready work.
And it’s that distinction that gets buttered over. I think we could have a much more adult conversation about this, if people would be more clear that AI was used in some capacity.
But execs don’t want to pay for AI unless they can fire everyone else at the same time. So nobody has the right conversation.
Doesn’t look to be a “they” but one dude it seems like.
Check the full video description then his Channel and then his website(which is just links to his social accounts related to his colorist job, projection mapping job, Ai creations, and his photography job/projects.) http://www.laszlogaal.com
Personally I feel the BTS part was far more convincing. That was impressive and crossed many an uncanny valley. Still had some of the camera moves Ai seems to love to use though.
Yo, this dude goes hard. Another podcast I’ve been enjoying is “This Machine Kills.” Jathan Sadowski, very smart, just finished his book “The Mechanic and the Luddite” and it was fantastic
While this is a satire piece, it probably captures the mindset of AI tech bros pretty well, which is why any of our concerns about AI will not register much. There’s a huge divide between our world and theirs.
A problem we’ve seen for some years, now that most tech is mainstream and used by every generation, but mostly built by folks in a small and unique bubble that cannot wrap their head around other people’s circumstances or that people in the 90s and before didn’t have everything as digital assets that you can just upload from your phone to a contemporary intake form. Mind the dropdown and the bay area bubble.
