Key 🗝 🦊✅
@keytryer.net
He/him. Fox gamedev irl. Machine pervert. Furry stuff here and opinions about technology, games, movies, AI, VR. Pfp + Banner 🎨 by @Perpleon.bsky.social
https://vrchat.com/home/user/usr_d9ee62da-234d-445a-8acc-720e0d33f8f0
https://vrchat.com/home/user/usr_d9ee62da-234d-445a-8acc-720e0d33f8f0
Reposted by Key 🗝 🦊✅
Like there was that Reddit comment on one of the early Stable Diffusion discussions like 4 years ago where someone said that basically this would happen and they were like "Nuh huh maybe your grandchildren will see it."
January 11, 2026 at 7:47 PM
Like there was that Reddit comment on one of the early Stable Diffusion discussions like 4 years ago where someone said that basically this would happen and they were like "Nuh huh maybe your grandchildren will see it."
January 11, 2026 at 8:09 PM
Reposted by Key 🗝 🦊✅
When James Cameron or the Paramount people say "We're using AI the right way, to iterate and enhance human work" I think they're talking about the next 2 or 3 years or so, but not about the next 5 or 15.
January 11, 2026 at 7:48 PM
When James Cameron or the Paramount people say "We're using AI the right way, to iterate and enhance human work" I think they're talking about the next 2 or 3 years or so, but not about the next 5 or 15.
Reposted by Key 🗝 🦊✅
Some concrete predictions for 2026 are:
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
January 11, 2026 at 7:47 PM
Some concrete predictions for 2026 are:
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
And today, there's even a local free option and it's not even by Google, but by another lab researching the same things.
January 11, 2026 at 7:49 PM
And today, there's even a local free option and it's not even by Google, but by another lab researching the same things.
I'm sincerely still shocked that we got native video-audio models in 2025. I legitimately thought that was purely in research stage around mid-2025, when Google reported papers about it in late 2023, and would be years before a usable product.
January 11, 2026 at 7:49 PM
I'm sincerely still shocked that we got native video-audio models in 2025. I legitimately thought that was purely in research stage around mid-2025, when Google reported papers about it in late 2023, and would be years before a usable product.
Honestly, there are probably few things in the world more inevitable than AI in mass media, but I still see lots of cope about it.
bsky.app/profile/keyt...
bsky.app/profile/keyt...
I think the freakouts at Hollywood as an industry are mostly justified, and in fact it is a huge threat to it, using their own data against them to create very cheap things. But I think that's fine, too.
I do think that AI filmmaking will be a thing, but I'm more convinced that ultimately what's important about AI is more the algorithm rather than the data. That is to say that, sure, they have to ingest lots of films, but I think data will end working out more as "reference"
January 11, 2026 at 7:49 PM
Honestly, there are probably few things in the world more inevitable than AI in mass media, but I still see lots of cope about it.
bsky.app/profile/keyt...
bsky.app/profile/keyt...
This will ultimately destroy the mass media industry if it doesn't change to fit these sorts of things. I think Disney are probably the only ones who have shown a good adaptation plan by adopting it early.
January 11, 2026 at 7:48 PM
This will ultimately destroy the mass media industry if it doesn't change to fit these sorts of things. I think Disney are probably the only ones who have shown a good adaptation plan by adopting it early.
When James Cameron or the Paramount people say "We're using AI the right way, to iterate and enhance human work" I think they're talking about the next 2 or 3 years or so, but not about the next 5 or 15.
January 11, 2026 at 7:48 PM
When James Cameron or the Paramount people say "We're using AI the right way, to iterate and enhance human work" I think they're talking about the next 2 or 3 years or so, but not about the next 5 or 15.
It'll also be very different from current systems in that prompting specific words won't be as important, and the way to control will involve "real" controls like currently using Blender or even weirder interfaces like full body manipulation.
January 11, 2026 at 7:47 PM
It'll also be very different from current systems in that prompting specific words won't be as important, and the way to control will involve "real" controls like currently using Blender or even weirder interfaces like full body manipulation.
I expect that in 15 years the experience will be a lot more different than anyone thinks. Nothing like 3D animation or programming.
Literally interfacing with virtual characters through a computer (or in VR) and guiding them like actors or manipulating things in an environment.
Literally interfacing with virtual characters through a computer (or in VR) and guiding them like actors or manipulating things in an environment.
January 11, 2026 at 7:47 PM
I expect that in 15 years the experience will be a lot more different than anyone thinks. Nothing like 3D animation or programming.
Literally interfacing with virtual characters through a computer (or in VR) and guiding them like actors or manipulating things in an environment.
Literally interfacing with virtual characters through a computer (or in VR) and guiding them like actors or manipulating things in an environment.
for very powerful algorithms that more or less will be able to understand how any reference is supposed to work.
January 11, 2026 at 7:47 PM
for very powerful algorithms that more or less will be able to understand how any reference is supposed to work.
I do think that AI filmmaking will be a thing, but I'm more convinced that ultimately what's important about AI is more the algorithm rather than the data. That is to say that, sure, they have to ingest lots of films, but I think data will end working out more as "reference"
January 11, 2026 at 7:47 PM
I do think that AI filmmaking will be a thing, but I'm more convinced that ultimately what's important about AI is more the algorithm rather than the data. That is to say that, sure, they have to ingest lots of films, but I think data will end working out more as "reference"
Some concrete predictions for 2026 are:
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
January 11, 2026 at 7:47 PM
Some concrete predictions for 2026 are:
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
- 90% video models longer than 45 seconds
- 40% video models longer than 2 minutes
- 70% a "decent" world model with coherence of 1 minute is usable for free or at a reasonable price.
I think two things will happen after this. One, they'll get more coherent and solve a lot of issues they currently have. Two, they'll get a lot longer and a lot more efficient.
January 11, 2026 at 7:47 PM
I think two things will happen after this. One, they'll get more coherent and solve a lot of issues they currently have. Two, they'll get a lot longer and a lot more efficient.
Back then it really was like "It's one thing to generate one image, but 24 images a second is a whole different deal", though they did end up solving that issues by predicting and denoising spatial temporal vectors rather than diffusing whole pictures.
January 11, 2026 at 7:47 PM
Back then it really was like "It's one thing to generate one image, but 24 images a second is a whole different deal", though they did end up solving that issues by predicting and denoising spatial temporal vectors rather than diffusing whole pictures.
Like there was that Reddit comment on one of the early Stable Diffusion discussions like 4 years ago where someone said that basically this would happen and they were like "Nuh huh maybe your grandchildren will see it."
January 11, 2026 at 7:47 PM
Like there was that Reddit comment on one of the early Stable Diffusion discussions like 4 years ago where someone said that basically this would happen and they were like "Nuh huh maybe your grandchildren will see it."
Reposted by Key 🗝 🦊✅
anyway i just realized i forgot to actually include a link, lmfao.
carrion.chat
please join us, have fun, and your reskeets and support are greatly appreciated. we can take the internet back from corporations but we can't just sit there and wait for it. we have to fight.
carrion.chat
please join us, have fun, and your reskeets and support are greatly appreciated. we can take the internet back from corporations but we can't just sit there and wait for it. we have to fight.
carrion - Privacy-Focused Adult Roleplay
Adult roleplay platform with zero server-side message storage. Messages route through XMPP in real-time and persist only in your browser.
carrion.chat
January 11, 2026 at 1:37 AM
anyway i just realized i forgot to actually include a link, lmfao.
carrion.chat
please join us, have fun, and your reskeets and support are greatly appreciated. we can take the internet back from corporations but we can't just sit there and wait for it. we have to fight.
carrion.chat
please join us, have fun, and your reskeets and support are greatly appreciated. we can take the internet back from corporations but we can't just sit there and wait for it. we have to fight.
Reposted by Key 🗝 🦊✅
Journalists are still making fun of Dario Amodei for predicting 90% of code written by AI, but this is exactly the sort of thing I'd expect to start happening if they were approaching that capability.
January 11, 2026 at 2:35 AM
Journalists are still making fun of Dario Amodei for predicting 90% of code written by AI, but this is exactly the sort of thing I'd expect to start happening if they were approaching that capability.