I work professionally on game engines and also have my own custom engine. Watching the space for 15+ years, I would say that nearly _every_ custom game engine exists as a demonstration of graphics capability, not anything to do with improving the actual experience of game development. This is in part to do with the fact that working on 3D at all is a slippery slope towards continuing to work on graphics stuff.
Almost certainly not if things remain as they are. The reason there's been little traction is the quality gap between diffusion and autoregressive models is pretty stark. I mean just look at the benchmarks here. Large dropoffs, with the hardest benchmarks seeing the largest drops. On top of that, almost all the speed benefits of diffusion models become negated at scale. So this is only attractive for local model development and almost everyone training local models still care about pound for pound quality and inference efficiency at scale.
It's fast enough that "ask it twice and pick the best" should still come out ahead performance-wise. I don't know how much that would close the quality gap by, but it's worth a play.
The thing is, diffusion models perform somewhat worse than autoregressive on text. So you lose some performance.
Speed is the big advantage. Autoregressive when doing local inference is mostly memory bound; you're doing one token at a time, for each token you need to load all weights. MTP helps a bit by allowing you to draft tokens in a smaller model and then verify them in parallel with the larger model, allowing you to do a few computations for every memory load, but because you're still doing tokens sequentially and need to discard invalid drafted tokens, you can only get so much speedup.
For hosted models, however, you can batch many token generations together, fully utilizing all of the compute while no longer being bottlenecked on memory bandwidth. So they are already operating at close to max efficiency.
So, diffusion kind of loses its beneifit in hosted models. Sure, maybe you could pay more to have slightly lower latency responses by doing diffusion for one user at a time instead of autoregressive for many in parallel. But given that it also reduces accuracy, it's hard to see where you'd really want that. Unless they're able to bring it up to par with autoregressive, it seems like it's a bit of a dead out outside of local models where you're generally just doing one thing at a time.
I'm particularly curious to know how this plays out, and I seriously hope that more labs focus on diffusion models for text usage.
My immediate thought - this performs slightly worse than the autoregressive gemma equivalent, but it may also let me functionally run better models in diffusion variants.
Ex - I can run 70b-120b autoregressive models locally right now, but I get ~5-15t/s, which just isn't fast enough for serious work.
Which caps me down in the 20-36b models (ex - gemma4) where I can get 100+t/s on the same hardware.
So the question becomes - does the quality drop from a diffusion model outweigh the quality bump from using a larger model?
Because if not... sounds like diffusion models have a lot of space to thrive.
---
Sadly - if they can't be hosted profitably, I question whether this space will actually be explored.
This has been around for a long time and I've always been so surprised it has had seemingly so little traction outside of the author's own projects. The love and care and thoughtfulness of every library has always been so great to explore.
I love karsten and have been following his work for a long long time. Used/Use his old toxiclibs library extensively both in processing and java. Have ported parts of it to golang for personal use. I want to say he also had some librarys for openframeworks/c++ and I probably used those as well (it was perhaps someone elses rough port of toxic to c++).
Originally I bounced off this because of clojure, it just wasn't worth the time to learn. Now it's typescript.. another language I'm just not interested in learning. I'm happy to port whatever bits of the library I want to use to a language I prefer when I need it.
But this is a absolutely remarkable set of libraries covering all kinds of nooks and cranies. It's worth putting on everyones list.
The audio-driven animation stuff here is so nice. A year ago I went on a journey to produce a video podcast waveform based off the audio track, and the process was incredibly painful for no obvious reason. My hope here is that I can now just do this all within Fusion and not need to render this in an external tool.
Also nice is built in loop (ping pong) animations! No more duplicating keyframes!
Not only this but hermetic checks on local machines for spot testing new models is becoming increasingly difficult, if not impossible.
- We have 0 visibility into what Anthropic does with our own prompts server side (do they return cached results from similar queries? Do we develop our own hot paths?).
- Local memory files are written independent of project directory and are acted on by the new models, even if old models wrote them
- CLAUDE.md files have varying degrees of efficiency and different models (and effort) treat them differently
- Our own git history "supports" newer models - ie if you have a larger body of work in git when you adopt a new model (like 4.8) than when you started from scratch with 4.6 or something, 4.8 may "appear" smarter when in fact you just have more evidence and signal about what you intend for a model to do.
currently there's a huge risk Marathon might get shut down like high guard or concord because of player numbers. If they shut it down, nobody could play it anymore.
If there's a system in place for us to be able to host our own servers, or... I don't know? OWN the game we bought and play it because we OWN it because we PAID for it. Then yeah, that's good and it's a good movement and I fully support it.
I genuinely don't understand the other side of this argument because it just feels like no for the sake of no.
what's the risk? "Oh no we might have to provide the product we sold!" Lmao. That's the right kind of risk to force on sellers in a market. Sellers risk getting penalized for not providing what they have paid for. Great. Fanastic. Sounds healthy for any market and may even increase average customer confidence enough for a surge in sales, one that can't be created by one company alone.
Same boat — looking at both the product page and a lot of the comments here, people seem to miss how great C1 is (and how much better it has been than lightroom for years). So much of photo editing as well isnt just color touchups but media management, and I think C1's workflow is incredible and fast and doesn't really leave me wanting anything else.
I love (video) Resolve, but I dont see anything here where it has some of the great C1 features like "group by similarity" and other media management options.
"I make AI output lots of stuff" is not an intrinsically valuable thing. I can run the same thing on Claude in research mode and get a report with cited sources in a more digestable format on my phone. What's the eval here on if any of this is good? Is it even possible to test (ie, you cant really AB test startup ideas)?
Great question. The core of Spine is coordinating multiple specialized agents across multiple models, using the canvas to store and pass context selectively so each agent works with exactly what it needs.
reply