Comment by echelon
9 days ago
> I'm sick of everyone trying to come up with a use case to get all my data in everyone's cloud so I have to pay a subscription fee to just make things work.
AI photo and video generation is impractical to run locally.
ComfyUI and Flux exist, but they serve a tiny sliver of the market with very expensive gamer GPUs. And if you wanted to cater to that market, you'd have to support dozens of different SKUs and deal with Python dependency hell. And even then, proficient ComfyUI users are spending hours experimenting and waiting for renders - it's really only a tool for niche artists with extreme patience, such as the ones who build shows for the Las Vegas Sphere. Not your average graphics designers and filmmakers.
I've been wanting local apps and local compute for a long time, but AI at the edge is just so immature and underpowered that we might see the next category of apps only being available via the cloud. And I suspect that these apps will start taking over and dominating much of software, especially if they save time.
Previously I'd only want to edit photos and videos locally, but the cloud offerings are just too powerful. Local cannot seriously compete.
But who said anything about AI? Lots of local-first apps have nor need any AI whatsoever. And by the way, Topaz Labs has good offerings for editing photos and videos with AI that run locally, works great for many use cases (although it's not fully generative like Veo etc, more like upscaling and denoising, which does use generative AI but not like the former).
Most cloud apps have no need for AI either, but companies are pushing it anyway for bullshit marketing reasons, similar to what they did with blockchain a decade ago.
Sure, that's unrelated to my point however, it's a non sequitur.
I suspect that most content will be generated in the future and that generation will dominate the creative fields, white collar work, and most internet usage.
If that's true, it's a substantial upset to the old paradigms of data and computing.
Yes, that is true, but again for apps like a fitness tracker, it is not "content" based. Sure, it might have some AI in the form of chatbots to ask what your diet plan should be based on your current progress, but that's not what you're talking about. In my experience, most local-first apps are like this fitness tracker, utility tools, rather than a means to view content, like TikTok.
2 replies →
> AI photo and video generation is impractical to run locally.
You think it always will be? What can the new iPhone chips do locally?
Regardless of what hardware capabilities exist, the previous post makes it sound like every application needs AI which is just not true.
> You think it always will be? What can the new iPhone chips do locally?
I suspect we're a decade off from being able to generate Veo 3, Seedance, or Kling 2.1 videos directly on our phones.
This is going to require both new compute paradigms and massively more capable hardware. And by that time who knows what we'll be doing in the data center.
Perhaps the demands of generating real time fully explorable worlds will push more investment into local compute for consumers. Robotics will demand tremendous low latency edge compute, and NVidia has already highlighted it as a major growth and investment opportunity.