alt Hacker NewsGenie 4
Comments
Surprising how Tencent's HY-World 2.0 open weights world model didn't get as much attention:
> Genie 4 has advanced audio capabilities
> Prompt asks for saxophone
Looks inside:
> Sound of a clarinet
Maybe Genie 5 will get it.
It looks like they had a hard time with the magic wand, peacock-to-cat example. Despite using a very long prompt that repeatedly stresses that the video should look photorealistic and NOT like a video game, the output video looks like a video game. The model is clearly capable of generating photorealistic humans and animals, so I wonder why it struggled to do so with that particular prompt.
Reminds me of a post[1] by Sam Altman in 2024:
> movies are going to become video games and video games are going to become something unimaginably better
Sometimes I really wanna skip 10 years into the future and see where all this eventually leads.
Cool demos, but does anybody else find it troubling that companies keep claiming this tech can train robots? It’s so obviously not representative of the real world.
A big chunk of the right side of castle disappeared, including catapult, when camera turned back while flying on dragon. See 1:25 to 1:29. I was not expecting this to happen in a "world model".
> Genie 4 is natively multimodal, enabling audio-like music, ambient noises, and interactive dialogue
What on Earth is audio-like music? Do they mean music-like audio?
Alert me when these "world models" can make a video of a skateboarding line that fools a skateboarder.
And no, this didn't: https://www.youtube.com/watch?v=V2K_OLiZ_RM
Looks cool, but it's very hard to get me excited about yet another proprietary model for which absolutely zero technical details are released (not even the number of parameters), which is safetymaxxed to the limit, and whose usage quota will (most likely) be so limited that it'll be effectively useless for anyone who doesn't work at Google or doesn't have infinite money.
If only there was some way we could create some sort of arrangement of 3D models and place a virtual camera within the arrangement. We could introduce some physics simulation and make the position of the virtual camera move and render what it sees to the user.
If such a system were possible, it seems like it would be far better at consistency and realism, since it could be fine tuned and hand crafted to best represent the desired world. It could be used to make interactive experiences, games, perhaps even feature films!
Alas.
In the clip with a marble bust dropped on a watermelon, at this point https://imgur.com/a/IHLnkUs the bust should've toppled over, if the physics were properly modeled.
In the man walking through the dressing room scene, the camera starts behind the man. And you can see that the left half of his body is wearing a grey suit... but the right half is wearing a grey dress. :O
That stained glass landscape video is pretty cool. I wonder if it would be possible to implement that effect in a real-time game engine.
Gotta love the kraken destroying the city in the background of the Hong Kong lovers motorcycle scene despite nothing in the prompt about monsters. Filmmakers are still safe.
A few years from now, Lydia in Skyrim will have a fully consistent personality with dynamic dialogue and full contextual memories of all our adventures together in Skyrim.
A visual makeover is great but what happens when millions of people fall in love with virtual characters who seem to deserve that love?
Looks like we are inching closer to on the fly open world game creation soon. While the updates excite me as a senior data scientist but as a human they all just look scary. Just me?
If the checker clip is meant to be used to develop a robot to play checkers, then the physics is not remotely good enough. When the robot takes the first piece, the piece looks like it's moving up from the board, then suddenly doesn't. The last move also isn't valid checkers, unless I'm missing something -- the robot offers its piece to be eaten and the human doesn't take it.
With Anthropic going public, I expect a lot of model releases from the competition signalling to the market. "We are still around, please leave some money for us too"
In the first comparison of Genie 3 to Genie 4 it seems that:
- in Genie 4 the character's suit stopped making sense in the back where it now has a weird fold?
- the controls don't control the character even anymore as now its controlling some kid-height character in first person instead?
- Everything looks like Unreal Engine.
This does not look good.
guys they are building digital cages so they can rob you blind while you are in your nice digital world, sorry we will ignore this stuff into oblivion
Can't wait for the hyper-realistic customer service bots and deepfake propaganda videos.
The prompts build impressive clips.
I am more interested in how to make consistent "participants" and "environment" with multiple clips so they look like one longer clip.
Will that require same dev/coding harness and spec-kit-like engineering flow?
Like create a spec on how a movie will look like, feed contexts, create tasks so that multiple clips look congruent?
I don't know. With each new release of these models, they compare side-by-side how they are better than the previous ones, but at the same time I feel less impressed by each new release. Somehow they manage to be a seemingly big step forward but at the same time I feel I have seen so much AI that I just look at the common failure modes and they are all still there.
It's easy for them to show demos of abstract things because they don't necessarily have to follow real-world rules, but anything modeled after real world falls short of expectations.
I don't know if I'm missing the point, but Genie 4 almost looks worse than Genie 3 in terms of player control? It doesn't even seem to respond to inputs, it just looks like they took some AI generated videos and put the key widgets on top.
Why do we need this? No thank you. We don't need flying cars and more vendor lock in. We need to serve the current google within constraints. That should be driving direction. Being a better ecosystem member. Another fascist product from google.
Honestly MiniMax H3 looks better than this at most of the things I've been able to generate locally. These look super AAA game like - movement is uncanny and feels like it was trained on GTA3 or something like that - is this just me?
I would like to see how it breaks, don't spend all your focus on how nice it looks with specially curated shots. At least give a few examples of what happens when it breaks from running too long. They mention that it can fail under "Understanding current limitations" but don't show any examples. Especially since they mention the cases of safety and responsible use, it seems quite relevant to help prepare people by showing them some examples of how it can break and so you an learn to spot it.
Q: Is there a easy way for average people to recognize these live super-deep fakes?
I know I could run the content through a fingerprint detector. However, I'm a professional software engineer...not a normal user.
What's the ethical impact of this technology??
Again, very impressive from a technical point. But language is so flawed to do anything creative. It just turns out all looking the same and kind of boring
what good in the world could possibly come from this being available? did anyone working on this ever think about that?
I highly suspect that Genie is being made primarily for internal use, and access is more to gather feedback/training data and a bit of PR.
The future of movies will be promising.
[1]: https://www.minid.net/2026/8/2/the-actor-was-the-expensive-p...
I guess they used Genie 4 to make a video about people using Genie 4 in different contexts. This may have some implications.
WERE NOT IN A SIMULATION! YAY!
What is the point of this vs cheap robotic arm simulations?
The only place I see this applicable is for visual artists prototyping
This should not exist.
Can someone ELI5 how this "Leads to AGI"? I can kind of understand some sort of "theory of mind" going on here, but how would having a complex world model enable, say, an AGI that could file my taxes for me?
Amazing!
Question I have is when will they be able to render something more then 30 seconds.
And how much will it cost? There has to be a HUGE amount of computing power behind this. As a video game developer I live in fear and anticipation.
>audio-like music
My favorite kind of music.
> "A REAL, ultra-photorealistic, high-end 35mm cinematic first-person perspective..."
Real nice prompting there, Google. It also still looks like a video game rendering.
>coming soon to Google AI Ultra subscribers in Project Genie
Would be fun to muck around with world models. Are there any that I can try for under £200 p/m without being part of a special project?
A nit, but given that this is from a major corporation, and LLMs are supposed to be em-dash-happy, it's surprising (disappointing?) that the announcement uses:
> audio-like music
That's a hyphen. What's "audio-like music" -- music that is almost sound but not quite? Music made from raw audio? :-)
Wow, wow! What a treat! Such a fascinating time to be alive. Talk about fast feedback loops! This is pretty quick loops for trying out ideas!
I'm very impressed!
Surprised that "do you really think we're going to make it?" scene made the cut of the first video, her expression and tone is totally out of sync with the scene and demonstrates the problem with almost every AI gen video I've come across when it comes to capturing humanity.
The painting->interactive scene looks astonishing though
The playable element is cool PR but the news is that endless synthetic training data for embodied AI models is basically here or close.
It's hard for me to understand Google's redefinition of "world models" (basically: a txt2videogame modality for LLMs) as anything other than an attempt to bury Yann Lecun's definition of "world models", which includes an implicit bear case against LLMs and thus Google's stock price.
For several of these examples the details change if something goes out of frame and returns. A second castle is added in the dragon one. The people walking away from the camera change in the train platform one. The Hong Kong moped rider doesn't have a helmet and then he has one.
It's very good at maintaining consistency for things that stay in frame, but I don't know that this points to it developing a strong "world simulation" like they claim.