Broccoli Man, Remastered
mbleigh.dev
mbleigh.dev
>In no world would I ever have put together a real cast and crew to remake a 15 year old inside joke video for Googlers, but I was able to make it with AI.
BUT IT DID !! and part of the charm is that this involved real people talking, mutual understanding and a shared culture. That world existed it can still exist unless we surrender to the depravity of conformity and comfortability.
In fact, the overtly robotic voices added to the humor of the original, IMO. It was lost in this translation.
Any assertion that the world in which Broccoli Man was created was fundamentally different (in terms of "relying on someone else's framework to do a low-effort meme") from today is nostaglia. In fact, I suspect mbleigh spent more time on making the recreation work than was spent on the original Broccoli Man video.
AI has raised the bar, in terms of making it more difficult to create the trust necessary for people to be willing to open themselves up to that connection.
3 people, the phone with the best camera across those 3 people, and some costumes that aren't THAT had to recreate? Honestly could even skip the costumes and just print out two A4 sheets with orange bamboo shoots and broccoli and tape them to your shirt and I would honestly get a better kick out of that. It's fun to make stuff with friends, even if it's a bit crap!
Saying that, I don't think anyone making something with AI discourages anyone else from doing something with the joke, so I think all this has really replaced is a video the author would not have made otherwise... but they should try anyway! They're missing out on what would be a fun afternoon with friends
“Later. We’ve got a lot of work to do”
Now that I think of it, that skit was incredibly prescient. You can literally ask Gemini for more hat wobble now.
The bureaucracy is way worse. BCID requirements. MDBs that take 4+ hours to roll out. The Byzantine array of different MDB group types, often with two party control. GDPR company compliance. Tagging proto fields with provenance. Documents are private by default now. Most support happens in chat groups, and is if you're not at deepmind your help requests are getting ignored. Spending a month filing GUTS tickets to get your intern access to basic tools. XManager idle pruning. GCP automated boq setup/teardown tools usually fail and you have to fall back to getting support from a contractor 12 time zones away.
If a company publishes loads of articles about how they have technical controls for privacy and security, through encryption and compartmentalization and code review and build provenance and so forth, and all the people who work/worked at said company are always whining about how onerous those processes are, then what gives you reason to doubt it?
Everyone who wants to work at a startup knows where to find the rest of Silicon Valley (and Austin and etc.). I wish them the best and I look forward to reading their data-breach disclosures if they get popular enough for anyone to care about what they're doing.
Now I'm at google, and onboarded on to the version of the infra that already went through that, and I can take it all at face value. It is a PAIN still, but this is the reality of a system that interfaces with O(10^8) users, O(10^2) governments.
I think AI slop is decidedly different, because it just doesn't have the charm. I don't know if I can yet decompose exactly why that is.
After all, the creator didn't want to be an professinal full-time animator, he just wanted to animate three minutes.
The original author chose those assets and that background, other people made those assets on the first place and had to take a ton of tiny creative choices that changed the final thing and help transmit ideas and feelings (of uncanniness, vulgarity, surrealism, whatever).
Anyone can tell the difference between one and the other.
Maybe I'm wrong but I would say that the slop feeling of the original version was a deliberate decision and part of its charm. That kind of video and voices were a meme back then, IIRC.
None of that is what an Xtranormal auto-puppet show is about.
Pretty much exactly what you see in that video.
“No it was always slop!!!! Nyeh!”
I don't think it is, actually? "Friend-slop" is a super-popular genre of video game right now (for example, indie smash-hit PEAK), typically characterised by low-friction co-op experiences with a low-fi and/or kit-bashed aesthetic
It’s requires a good base group of friends that you will have fun with already, and the design for the game is “slop” / cheap / lazy in that regard. But not necessarily bad.
You'd think AI would make it easier to splice together individual clips, but I haven't find a tool that does that well yet. Opus seems tailored for doing fine tuning on long form content like podcasts.
I wouldn't and don't consider this to be AI slop. The author's own reflection captures perfectly my own feelings on the matter... intent does matter.
I think what gets lost in a lot of the AI media discussion is that intent matters. I find auto-generated slop-farm TikToks just as dystopic as the most fervent doomer, but I also remember when I was a 10-year-old kid making movies with my parents’ VHS camcorder and how much fun I would have had learning how to make things if I’d had tools like this.https://help.kagi.com/kagi/features/slopstop.html#what-is-co...
We are marking domains as slop if they contain mostly AI generated content. It is hard to say if a single piece of AI generated content is slop or not (there is useful AI generated content too). But if all the content on a domain is AI generated, it is likely something that people would not like in their results.
Personal opinion, using modern AI slop generation to recreate AI slop from a previous generation of the technology is pretty good art.
"... unstoppable." - What? I'm genuinely curious. Do you think this type of video is enjoyable? Maybe one, or even two like this is palatable - but I don't see anything 1) interesting or new 2) that would make me want to watch this by choice.
And when everyone is swimming in a sea of this type of work I don't think most people will enjoy it, either. It's also too bad we aren't more cognizant of how we got here, either. All of the artists work that was stolen to generate those images and then the watt hours burned. I'm not saying these types of offerings are completely for naught, but I do think some of these services should be forced to put a few pre or post pended frames that show (like nutrition facts) the environmental impact including model inference and that copyright was violated to make it.
Most of the creativity is clearly from the human driving the AI, but I suspect it would have been borderline impossible to make something like that as an individual without AI.
I use LLMs a lot and I like older weird AI generated stuff, but outside the uncanny valley... no.
What I find so interesting about this is how much is NOT new. It is a highly faithful reproduction of the beloved original.
The difference is the voices and visuals are much higher quality, and there are some added effects and cutaways that are nice touches.
I would still share the original video because it is a useful cultural artifact and teaching tool (as ridiculous as it is), and I would be happy to share this version at the same time, just because it is higher quality.