The GPT moment for AI interactive games has arrived! Google Genie 3 first test! So awesome!

Google released a world model called Genie 3 last summer.

It supports real-time generation of 24 frames of 720P interactive video content.

Moreover, it can maintain consistency for over several minutes.

The demo video is really impressive, but it has not been made available for use due to high costs.

Just now, Google announced that Ultra users in the U.S. aged 18 and above can directly try Genie 3.

The master quickly went to try it out while there were fewer people.

The result is, wow, it's amazing!

I feel that this year will definitely be a big year for AI interactive games and interactive videos!

To summarize the features of Genie 3:

• Extremely low control latency
• Ability to control the direction of characters and camera angles.
• Image clarity is quite usable
• Physical interaction between characters and the environment is very realistic.
• Can continuously control characters for over a minute

Words are not enough, let's take a look at some tests conducted by the master.

First, I created a prehistoric forest world, and I chose a dinosaur as the character.

It first generates a first-frame image, and once you confirm there are no issues, you can start generating the world.

You can see that even with high latency connecting to the U.S. server, the perspective control and character movement control are very responsive.

It's almost like playing a U.S. game with high latency.

Moreover, the dinosaur's movements are very smooth and biologically accurate, and the character's consistency is maintained very well, with almost no changes.

I deliberately made some large movements, and the environmental consistency was also fine, with no sudden changes or jitters.

At this point, I realized something was off; this thing seems really amazing!

Then I selected a felt-style snail world from the preset worlds.

I remixed it and changed the snail's blue shell to a red shell.

Then I noticed that it mentioned in the prompts that jumping was possible, so I focused on testing this.

First, you can see that its stylistic consistency is also very well maintained, continuously reflecting the felt changes.

Then, after I got used to the latency and controls, the snail became much more responsive.

I tried jumping, and indeed it could jump, and the height of each jump was consistent, with no instances of getting stuck in the ground.

At this point, the physical performance was starting to show.

Later, I found that the snail got blocked when it reached the stone part and couldn't move.

Then I started jumping frequently; you could even jump up steps and then jump down.

Moreover, throughout my random jumping and changing perspectives, the previous world scene remained stable, which is incredible

Finally, I want to say that this time I tested its physical performance even more rigorously.

I created an optimized version in Istanbul, along with a tabby cat.

At this point, I had already gotten used to the delay and could manipulate it very freely.

I started off by taking an unconventional route: I didn't follow the usual path but jumped directly onto the box by the roadside.

As a result, I found that it could really simulate a cat's jumping form, rather than just moving stiffly like a zombie.

It actually jumps like a cat, extending its legs first.

Then it easily jumped onto the box by the roadside. I could even jump from this box to another box!

This is amazing, friends! Especially if you've played AI videos before.

Sometimes the video models can't even guarantee that it won't sink, let alone the common issue of clipping in previous games, which is not a problem here.

Then I kept jumping between the boxes by the roadside and arching over pedestrians.

I found that there was no clipping, and it could perform movements that matched a cat's actions, colliding with the walls and boxes by the roadside.

After playing three worlds, I could no longer generate worlds normally (I guess everyone realized it and rushed in to play).

These are my current three tests.

Really, after playing, I found: the new era has arrived, friends!

A brief introduction to its operation process:

First, when you enter, you will see many prefabricated worlds made up of "bubbles."

If you click in, you can remix it, which means changing certain elements of that world. You can change the style, the content of the world, or the content of the character.

When remixing, you just input the prompts you want to change directly, just like using Lana to modify images.

After making changes: first click the button to the left of "Create World" to change the image, and then click "Create World."

In the middle is a creation button, and you can use this button to create a world completely from scratch.

The information you need to fill out is divided into the following parts:

1. Left: Fill in the prompt for the world (describe what this world is like)
2. Right: Fill in the prompt for the character (describe what this character looks like)
3. Left supplement: You can also write style prompts (such as various felt styles, etc.)
4. Below: You can upload other images you generated, or even upload yourself.
5. Mode selection: You can choose to switch between first-person perspective and third-person perspective.

I tried this, and sometimes the perspective switch doesn't work.

After entering the game, the controls are as follows:

1. WASD: Control the character's movement direction.
2. Right arrow: Control the camera direction (similar to common 3D game controls).
3. Additionally, the character can also jump.

I feel that for my friends in the U.S., if the latency is low, it would be even more enjoyable.

720P is already fully playable; we have truly entered a new era.

Imagine!

You can build worlds using large language models, create NPCs, add various random events, and include some audio.

You can go wherever you want, and everyone can create their desired game world and storyline.

Those 3D live-action games that recently exploded on Steam now seem very accessible; anyone can make them.

The imagination behind this is immense. I hope the costs can come down a bit by the end of this year, and the time can be extended.

Gemini 3 changes the world (i.e., the ability to add random events and influence the world), but this time it wasn't included due to computing power issues. I hope Google can update it later.

If you are an Ultra user, you can try it here (labs.google/fx/projectgenie).

But I hope Google's servers can be more reliable; it's currently impossible to generate normally.

Alright, this is the first test of Genie 3 brought to you by Master Cang.

Risk warning and disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this is at your own risk.