In my experience so far, separately from privacy concerns there's a specific use case where cheap local shines--when unlimited shots on /goal with a dumber model is better than limited time with a smarter model. This looks less like the assisted-coding scenario that's commonly brought up as a good local scenario, because if you're searching for a fast and accurate solution to a single blocking problem, the bare minimum for a model that can do that is a 24GB dGPU or a 64GB Mac. Instead, this looks more like a Hermes agent on an Raspberry Pi driving OpenCode on an old gaming computer with just enough RAM and VRAM to handle an MoE, churning out something overnight that would quickly exhaust the subscription plans, like a knowledge graph for a large document corpus.
It's also New Wave science fiction, and once you get past the conlang and the genre conventions, it's still incredibly violent! Kubrick iirc even once said in an interview that he removed some book scenes from his (initially X-rated) film adaptation, because he thought the invented language helped make them less disturbing to read, but audiences watching them dramatized would just give up and walk out. Anyway I sure do see some of the trends described in the article with my younger students and relatives, but such a strange unforced error makes me suspicious of the strength of the research.
In my experience, yes to hardware-accelerated video from CLI--running in EGL mode, with no window manager, that RPi model works very well. (A C++ app that uses video as a texture can be surprisingly performant, too.) But no to playing video in a full desktop environment and browser, not smoothly--it's just too much overhead.
An imo fascinating wrinkle that the article doesn't get into, Black Forest's main use case for their Flux models is "analytical" (modifying the user's pre-existing material, storyboards in this case) rather than "generative" (modifying stored material from the model's training corpus). In my experience with these tools so far, the analytical approach is more filmmaker-friendly, with image models fitting comfortably into well-established rendering and compositing roles. Meanwhile my current guess is that creative applications of the generative approach are going to end up looking a lot more like gamedev than filmmaking.
> “…modifying the user's pre-existing material, storyboards in this case…”
Let me understand this. You’re saying this Flux model takes as input your own images and drawings and then modifies them only?
If I understand this correctly, then it would get past the authorship safeguards public llms put up every time you ask for an illustration in the “style of”. Yes?
imo it's not setting up the site once that's the problem, it's keeping it maintained indefinitely without making mistakes, because hostile automated systems will keep on rattling the doorknob like Jurassic Park velociraptors. (And I agree, for sophisticated users keeping it off your home network goes a long way towards preventing worst-case outcomes.)
Afaik the basis of the Bluetooth thing is that out of all the different kinds of spread-spectrum radio we use now, Bluetooth happens to be the closest to her original design. So it's a drastic simplification, but not wrong.
The point is that Bluetooth descends from an independent intellectual heritage. No one was aware of her patent. The original experiments with Frequency hopping happened in the early 1900s with Marconi and Tesla. FSK was used for military radio in the 1930s and by the 40s had trickled out into widespread knowledge. Here's an example of a trade magazine from 1948 describing FSK [0] citing military work in the late 1930s, which also says
It may appear to the reader that this recent advance in telegraph technique is a rather obvious one
Going on to say that it's little studied because there simply isn't a good method to synchronize the transmissions yet. That's what Hedy and others were trying to solve, but the widespread solution would come from the independent invention of Barker codes in 1953. Many decades later when the people at Ericsson were designing what became Bluetooth, they looked at the FCC requirements that had been written to accept FSK radioteletypes in widespread use after the war and common best practices for radio system design.
Forgive me if you already know this, but there are reasonable arguments to describe either the Apple I or the Apple ][ as the first real personal computer.
Although Woz engineered both, without his partnership with Jobs, they wouldn't have been consumer products (which even the Apple I was, if barely!).
The reason I used 'Steve Jobs invented the personal computer' as a comparison, isn't that I think it is dead wrong, but that, to use your words, it's a drastic simplification.
Informative-Drawings already has monocular depth estimation built in--that's why its line results are so beautifully consistent. But without this extra step combining results from multiple camera positions, you get 2.5D geometry, not 3D.
Afaik it's a misconception that MRT necessarily involves the destruction of an embryo. The spindle transfer method transplants the mother's egg's DNA into the donor's egg before fertilization, so only one embryo is ever created. The UK trials exclusively used the older pronuclear transfer method, where two embryos are created and the donor's is destroyed, because the journey to full regulatory approval took about a decade and embryos are currently safer to freeze and thaw than eggs. In a hypothetical scenario where MRT became as widely available as IVF, this would not need to be the case for new patients.
Afaik they do respond to DMCA requests for the individual uploads, and they do more active curation for the collections where only specific user accounts have upload access.
Yeah it's a common misconception; I think it's it's because After Effects has far outlived the source of its original UI metaphor, the Oxberry (rostrum) animation camera, which at this point few people still have firsthand experience with. Nuke and most newer node-based compositing software offers a more scalable approach for a broader range of tasks, but it can't adequately address that very specific AE niche of under-the-camera-style animation.
