World Models 101
Contents
08

Are Video Models World Simulators?

By Nilushanan Kulasingham

PDF

Add an action input to a video model and it is steerable in principle. From Genie to Genie 3: how to tell conditioning from control, what scaling bought, and why fitting physics is not the same as having the rule.

If you have watched any generated video lately, you'll have seen somebody walk through a scene with the arrow keys. Until 2024 a video model was something you watched. It chose its own future and you had no say. One extra input changed that.

Take a model that predicts the next frame. Now tell it, along with the frames, which key is being held. From the same opening it can give you one continuation for left and another for right. Have a go with the figure below, and watch who picks each frame.

the frames beforenothingthe next framegivengiven
next frame picked by

Press Next frame. You have no say in what arrives.

frames so far

2 of 8

chosen by you

0 of 0

the extra input

none

Fig. 8.1Two ways of getting the next frame. Leave it on watch and press Next frame: the model picks, and starting again gives you the same clip back. Switch to steer and the arrow you press goes in alongside the frames, so the frame you get is the one you asked for. Watch the tally underneath for who chose each frame. The clip is illustrative.

Jake Bruce and colleagues at Google DeepMind made that move with Genie in 2024. Their paper was called Genie: Generative Interactive Environments. Genie was trained on footage of 2D platform games, Mario and its descendants, and it turned them into worlds you could steer. Nobody had labelled the keypresses.

Instead, the model had to explain each change between frames with one of a handful of codes. The codes it settled on behaved like a controller. Genie also married two families that had grown up apart: world models built for agents (chapters 5 and 6) and video models built for audiences.

frame pairs, before and afterno keypresses in the dataeight codescode 1code 2code 3code 4code 5code 6code 7code 8a fresh frame

Twelve changes and no labels. The model has to file each one under a code.

pairs

12

codes in use

0 of 8

keypress labels given

0

Fig. 8.2This is that trick in miniature. The model sees pairs of frames and no keypresses, and has to file each change under one of eight codes. Press Sort the changes and watch the piles: one code collects every step left, another every jump, and nobody told it which was which. Then press a code and see what it does to a fresh frame. The frames and the codes are illustrative.

A video model with an action input is somewhere you can be, and that is new. Whether it is a simulator, something that obeys your actions and holds the rules of the world, has to be measured, and the announcements tend to skip that.

An action input is not control

A model can receive an action and still ignore it, or respond only weakly. It can also let the action change how things look without changing the path underneath. Think of a steering wheel that turns without being connected to anything.

So control has to be measured. Hold the start fixed, vary only the command, and check whether the futures split apart as asked. I call it the steering-wheel test, and the figure below is that test as a picture.

command left−53command straight0command right+53same starting framerealised motion

Action input

present in every row

Left-right separation

107 px

Control test

The command changes the video, but weakly.

Fig. 8.3Every row gets a different command from the same start. Conditioning is the name for giving the model the action as an input, and action fidelity is how far the model obeys it. Drag fidelity towards zero and you should see the paths fold together while the input stays present.

As you can see, the input can be there the whole time and mean very little. That failure is the one I'd expect. A model trained on footage is rewarded for frames that look right. If the footage mostly goes straight, going straight is a good bet whatever the key says.

Good video hides weak control. If three beautiful continuations all go straight, you have been shown video. A planner that asked for three counterfactuals, three what-ifs from one start, has been handed one of them three times.

From watching to walking in

Scaling did more than most people expected, including most of the people building it. Let's take the two systems that mark the change.

GameNGen came from Dani Valevski and colleagues at Google Research in 2024. Their paper was called Diffusion Models Are Real-Time Game Engines. A diffusion model makes a picture by starting from noise and cleaning it up step by step. GameNGen used one to run DOOM, the first-person shooter from the early nineties.

the frames beforeforwardkey heldconditioningthe frame being made

Start from noise. Each step cleans it a little, guided by the frames before and the key.

steps

0 of 12

time per frame

72 ms

frames a second

14

playable

about 20 a second, the playable line, illustrative

Fig. 8.4This is how a diffusion model makes one frame of a game. The frames before it and the key being held sit on the left as the conditioning. Press Clean up and watch the noise turn into a frame one step at a time. Then drag the number of steps down and read the frames a second: fewer steps is faster and rougher, and somewhere along that slider watching turns into playing. The picture and the timings are illustrative.

It made each frame from the earlier frames and the player's inputs, fast enough to play. People shown short clips struggled to tell it from the real game, its authors reported.

Genie 3, from Google DeepMind a year later, is where the numbers stopped looking like a demo. It is reported to make interactive video at 720p and 24 frames a second.

It is said to hold a scene together for several minutes and to take instructions part-way through: change the weather, add a flock of birds. Walk away from something, walk back, and it is roughly as you left it. Have a look at the clip below and watch for exactly that.

Google DeepMindGenie 3: Creating dynamic worlds that you can navigate in real-time
Fig. 8.5DeepMind's own clip of Genie 3. Watch for the moments the camera turns away and comes back, and ask yourself what you were shown. The scene is roughly as it was, and nothing in the interface lets you open up whatever held it there. Everything in the clip is the lab's report, including the numbers I quoted above.

The public interface offers no stored 3D layout of the scene, and no list of its objects, that you could inspect. Yet DeepMind says Genie 3 can look back over its own path and keep visual information for about a minute. Whatever state supports that stays hidden inside the generator.

In under two years the field went from steering a blurry platformer to walking around a photoreal scene that remembered where you had been. Scaling did not give you anything you could open up and read.

Emergent physics, and where the footage runs out

You will see the claim made about all of them. OpenAI's 2024 report, Video generation models as world simulators, put emergent physics into common use. It was the write-up for Sora, OpenAI's own video generator. The title is the part that spread, and the claim in it bundles an observation with a conclusion.

The observation is real. Train on enough video and the outputs start to respect regularities nobody put in. Dropped objects speed up as they fall, water pools rather than piling up, and hidden objects come back into view when the thing in front moves. Nobody supplied an equation; those regularities are in the training footage, and predicting it well means copying them.

training footagethe gaps grow frame by framea new dropgiventhe model continues

No equation anywhere. Only footage, and a drop it has not seen.

equations supplied

0

gap between frames

····

matches the footage

not yet

Fig. 8.6On the left are the training clips, each one a thing being dropped, and each one speeds up on the way down. On the right, press Continue and the model finishes a drop it has never seen. It speeds up too, and nowhere in the figure is there an equation. Now switch the footage to the made-up kind, where things fall at a steady speed, and press Continue again. The regularity follows the footage. Illustrative.

The conclusion is where the trouble starts. "It learned physics" says the model picked up the rule, and that is testable. The test has never been whether the outputs look right on the footage it was trained on.

Bingyi Kang and colleagues ran one in 2024, in a paper called How Far is Video Generation from World Model: A Physical Law Perspective. They trained video models on simple simulated scenes of balls moving and colliding, then asked for speeds the footage never showed. Figure 8.7 runs that test twice: once on speeds the footage covered, and once on speeds past them.

the speeds the footage covered

the bench

One model and a bench of scenes. Score it.

scenes on the bench

8

passed

0 of 8

the model

unchanged

Fig. 8.7Here is that test as a scorecard. Each tile is one scene, ordered by speed, and the shaded ones are the speeds the footage covered. Press Score it, then switch the bench to scenes from past where the footage ends and score it again. Same model, same procedure, and the only thing that changed is where the scenes came from. The speeds and the errors are illustrative.

Everything inside the band fits with having learned physics. It fits just as well with having learned the answers to the questions it was asked, and Kang's group concluded their models were doing the second: matching each new scene to the training scenes it most resembled and copying what happened there. Nobody tests outside the band, because that is where the footage runs out. I call that the band problem.

So the defensible claim is narrower. These systems copy a great many physical regularities across the situations they were trained on. Whether that is the same as having the rule, no demo you have seen can settle, because every one of them was inside the band.

What they are for

None of that argues against using them. Answering "it has not learned physics" with "so it is a toy" is the mirror-image error, and I see it nearly as often.

For training other systems a generated world is useful. It avoids contact with the real environment and costs GPU time instead. It resets at once, and it can produce a thousand versions of a scene that would take a week to stage. Press the buttons in the figure below a few times and you'll see what I mean.

a real stage

ready

versions so far

0

a generated world

ready

versions so far

0

Same scene on both sides. Press Another.

stage · versions

0

stage · time spent

0 days

stage · real contacts

0

stage · status

ready

generated · versions

0

generated · time spent

0 frames of GPU time

generated · real contacts

0

generated · status

ready

Fig. 8.8One scene, two ways of getting more of it. On the left is a real stage: press Another and a day goes by and one new version arrives. On the right is a generated world: press Another and a version appears in the time it takes to draw a frame, with the lighting, the clutter and the camera moved. Press Reset on both and watch which one is ready. The counts and timings are illustrative.

NVIDIA, whose chips train most of these models, built Cosmos for this. Cosmos is a set of video generators. The 2025 paper is called Cosmos World Foundation Model Platform for Physical AI. It makes physically-aware video as training data for robots and vehicles, where real data is dearest.

NVIDIA sells Cosmos beside its explicit, hand-built simulators as one platform, and has not declared either the replacement for the other. I'd take that as a hint about where things stand.

For being a world you can move around in, they are already good. No other approach lets you walk into the picture. I think that is worth more than the sceptics admit and less than the launches imply.

For being the model an agent plans inside, raw video generators are an awkward shape. A planner may need to score a hundred action sequences, and rendering each one is expensive. Systems can instead distil a learned latent, a short internal code that stands in for the frames, and plan in that.

one action sequence, eight stepsplan in pixels×103,200 unitsplan in a latent×10400 unitsthe chosen onedrawn once, after choosing×1

10 sequences and the pixel bill is most of the chart. The latent planner paid a fraction and drew one.

Sequences

10

Frames rendered, pixels

10 × 8 = 80

Codes scored, latent

10 × 8 = 80

Frames drawn, latent

8

Cost, pixels

3,200 units

Cost, latent

400 units

Fig. 8.9A planner has to score many action sequences before it picks one. Drag the number of sequences up and watch the two bills. Planning in pixels renders every frame of every sequence, and its bar runs off the chart. Planning in a latent, a short code standing in for each frame, scores the same sequences for a fraction of that, and only the chosen one is ever drawn. The costs are illustrative.

Where the seams are

There are three seams where a demo can mislead you.

Consistency is learned, not enforced. Nothing stops the room from changing, and usually it does not. But usually is a statistical property of a trained system, and a guarantee is a property of an engine.

no runs yetone mark per run
what holds the door in place

Nothing here is holding the door in place. Press and see how often it comes back anyway.

runs

0

held

0 of 0

guarantee

no, this is a rate

Fig. 8.10One room, one door, and a walk away and back every time you press. Press Again a few times and the door comes back where it was, which is what usually looks like. Keep pressing and one run comes back with the door somewhere else. Then flip the switch to an engine holding the scene and there is nothing left to fail. The rate is illustrative.

The interesting moments are the vague ones. A model trained on pixels is pushed towards the average of the possible futures wherever the future is open. We saw in chapter 7 what that average looks like: a picture of something that cannot happen.

The future is most open when something is about to happen. So the steering-wheel test is hardest to pass exactly where passing it matters. Step along the clip in the figure below and you'll see why.

123456moment 1givenasked for leftasked for righthow far apart they should have come outhow far apart they did

every moment, both bars

It obeys here. Both commands end up in the same corridor, so obeying cost it nothing.

moment

1 of 6

at stake here

almost nothing

obeyed

yes, and it did not matter

Fig. 8.11One clip with a junction in it, and the steering-wheel test run at every moment along the way. Step through the moments and read the two bars: how far apart the two commands should have come out, and how far apart they did. In the corridor it obeys, and nothing was at stake. At the junction the two commands come out on top of each other. The moments and the numbers are illustrative.

The scores do not compose. Benchmarks measure visual quality, geometry, action fidelity and persistence one at a time. There is no accepted rule for combining them into one ranking.

So I will not tell you whether Genie 3 beats GameNGen. A claim about improvement belongs to one particular test. Click through the measures below and watch the winner change.

System A0.91
System B0.62
System C0.74

Frame quality, which is what demos are made of and what a per-pixel score rewards.

Best by this measure

System A

Best overall

There is no agreed way to combine these, so the question does not currently have an answer.

Fig. 8.12The scores are invented and the systems are not real, but the disagreement is. Click through the four measures and watch the winner change. There is no accepted way to combine them into one ordering, so "better" only ever means better at one of these.

So, are video models world simulators? My answer is that they are somewhere you can be, and that is new and real. Whether one obeys you, and whether it holds the rules rather than the footage, are two separate measurements, and the demo you saw did not make either of them. Ask for the steering-wheel test, and ask what was outside the band.

Hopefully this chapter helped. There is a short quiz below if you want to check the ideas stuck. If I have got something wrong, or you know a better source than the ones I used, the About page has a corrections section and I would be glad to hear from you.

Try it

1 / 8Score 0

What change makes a video model capable of being steered, without proving that it will obey?

Pick one

Fig. 8.13Eight questions on the move and the claim. If a demo has just impressed you, go straight to the middle three.

Sources

Fig. 8.14I've ordered these for someone building on this rather than for historical completeness. Kang and the OpenAI report sit together as claim and measurement.

This chapter is free and open source. A star helps the next person find it.

Star on GitHub22
Star on GitHub22