Finally, true next-gen video generation and video game graphics may just be around the corner (see details)

I came across this YouTube video just now and it presented two recently announced technologies that are genuinely game changing next-level leaps forward I figured the community would be interested in learning about.

There isn't much more info available on them at the moment aside from their presentation pages and research papers, with no announcement if they will be open source or when they will release but I think there is significant value in seeing what is around the corner and how it could impact the evolving AI generative landscape because of precisely what these technologies encompass.

First is Seaweed APT 2:

This one allows for real time interactive video generation, on powerful enough hardware of course (maybe weaker with some optimizations one day?). Further, it can theoretically generate an infinite length, but in practicality begins to degrade heavily at around 1 minute or less, but this is a far leap forward from 5 seconds and the fact it handles it in an interactive context has immense potential. Yes, you read that right, you can modify the scene on the fly. I found the camera control section, particularly impressive. The core issue is it begins to have context fail and thus forgets as the video generation goes on, hence this does not last forever in practice. The quality output is also quite impressive.

Note that it clearly has flaws such as merging fish, weird behavior with cars in some situations, and other examples indicating clearly there is still room to progress further, aside from duration, but what it does accomplish is already highly impressive.

The next one is PlayerOne:

To be honest, I'm not sure if this one is real because even compared to Seaweed APT 2 it would be on another level, entirely. It has the potential to imminently revolutionize the video game, VR, and movie/TV industries with full body motion controlled input via strictly camera recording and context aware scenes like a character knowing how to react to you based on what you do. This is all done in real-time per their research paper and all you do is present the starting image, or frame, in essence.

We're not talking about merely improving over existing graphical techniques in games, but completely imminently replacing rasterization, ray tracing, and other concepts and the entirety of the traditional rendering pipeline. In fact, the implications this has for AI and physics (or essentially world simulation), as you will see from the examples, are perhaps even more dumbfounding.

I have no doubt if this technology is real it has limitations such as only keeping local context in memory so there will need to be solutions to retain or manipulate the rest of the world, too.

Again, the reality is the implications go far beyond just video games and can revolutionize movies, TV series, VR, robotics, and so much more.

Honestly speaking though, I don't actually think this is legit. I don't strictly believe it is impossible, just that the advancement is so extreme, with too limited information, for what it accomplishes that I think it is far more likely it is not real than odds of it being legitimate. However, hopefully the coming months will prove us wrong.

Check the following video (not mine) for the details:

Seaweed APT 2 - Timestamp @ 13:56

PlayerOne - Timestamp @ 26:13

**https://www.youtube.com/watch?v=stdVncVDQyA**

Anyways, figured I would just share this. Enjoy.

https://redd.it/1lcc5p9
@rStableDiffusion

Depth Anything V2 Giant
https://redd.it/1lchv89
@rStableDiffusion

Revenant accidentally killed his ally while healing with a great hammer
https://redd.it/1lcoa0i
@rStableDiffusion

how to start with a mediocre laptop?

I need to use Stable Diffusion to make eBook covers. I've never used it before, but I looked it into a year ago and my laptop isn't powerful enough to run it locally.

Is there any other ways? On their website, I see they have different tiers. What's the difference between "max" and running it locally?

Also, how long much time should I invest into learning it? So far I've paid artists on fiverr to generate the photos for me.

https://redd.it/1lco242
@rStableDiffusion

Random realism from FLUX

https://redd.it/1lcp8dy
@rStableDiffusion

June 2025 : is there any serious competitor to Flux?

I've heard of illustrious, Playground 2.5 and some other models made by Chinese companies but it never used it.
Is there any interesting model that can be close to Flux quality theses days?
I hoped SD 3.5 large can be but the results are pretty disappointing. I didn't try other models than the SDXL based one and Flux dev.
Is there anything new in 2025 that runs on RTX 3090 and can be really good?

https://redd.it/1lco4i8
@rStableDiffusion

Self-Forcing WAN 2.1 in ComfyUI | Perfect First-to-Last Frame Video AI
https://youtu.be/lTVZ3jfMkug

https://redd.it/1lcsgqw
@rStableDiffusion

What is 1=2?

I've been seeing "1=2" a lot lately on different prompts. I have no idea what this is for, and when applying it myself I can't really tell what the difference is. Does anyone know?

https://redd.it/1lct2su
@rStableDiffusion