Free tools Windows power users keep installed
One-click scans. No signup required.
Google’s GameNGen simulates classic DOOM without a conventional game engine by using a diffusion model to generate each next frame from recent frames and player actions. The ICLR 2025 paper reports 20 frames per second on a single TPU and multi-minute stability, but the result is a DOOM-specific research prototype, not a universal game-engine replacement.
That distinction matters. GameNGen is not merely an AI agent playing the original executable while a normal renderer handles the visuals. The neural model produces the imagery and learns the visual consequences of movement, aiming, shooting, and navigation from recorded gameplay.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
DOOM Eternal: Standard Edition - PlayStation 4 | $27.49 | Buy on Amazon |
| 2 |
|
DOOM: The Dark Ages – Xbox Series X | $39.99 | Buy on Amazon |
| 3 |
|
DOOM: The Dark Ages – PlayStation 5 | Buy on Amazon | |
| 4 |
|
Doom - Xbox One | $26.99 | Buy on Amazon |
| 5 |
|
DOOM + DOOM II (Limited Run Games #144) - for Playstation 5 | $44.48 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
Key takeaways
- GameNGen is a Google research prototype that generates DOOM’s next visual frame with a neural model instead of running a conventional game loop.
- The model uses recent frames and player actions as context, learning from gameplay collected by a reinforcement-learning agent.
- According to the ICLR 2025 paper, GameNGen runs at 20 frames per second on a single TPU and remains stable during extended, multi-minute sessions.
- The paper reports a next-frame prediction PSNR of 29.4 and human judgments that were only slightly better than random chance when distinguishing short real and simulated clips.
- GameNGen is a DOOM-specific proof of concept, not evidence that commercial game engines have already been replaced generally.
What is Google’s GameNGen?
Google’s GameNGen is a research system that simulates the classic first-person shooter DOOM by using a neural model as the engine for the demonstrated environment. The system does not simply play the original game while recording its output; the model generates the game imagery itself in response to player input.
The project describes a future in which games could be represented substantially by learned model weights rather than manually authored game code. That is an ambitious research direction, while the published evidence remains limited to a specific DOOM demonstration.
#1 Best Overall
- Gain access to the latest demon-killing Tech with the DOOM Slayer's advanced praetor suit, including a shoulder-mounted flamethrower and the retractable wrist-mounted DOOM Blade
- Upgraded guns and mods, such as the Super shotgun's new distance-closing meat hook attachment, and abilities like the double Dash make you faster, stronger, and more versatile than ever
- You can't Kill demons when you're Dead, and you can't stay alive without resources. These tools are the key to your survival and becoming the ultimate demon-slayer
- A new class of (destructible) demon
- Battle mode is the new 2 versus 1 multiplayer experience built from the ground up at id software
The authors characterize GameNGen as “the first game engine powered entirely by a neural model.” That wording is the authors’ description of their research system, not proof that every kind of game engine can now be replaced by a neural model.
Did Google recreate DOOM with AI?
GameNGen recreates the visual and interactive consequences of DOOM gameplay in a learned simulation, but the published work should not be described as a commercial remake or a general-purpose replacement for DOOM.
A conventional DOOM executable contains explicit rules for movement, weapons, enemies, collisions, level state, and rendering. GameNGen instead learns from recorded gameplay how screens tend to change after actions such as moving, aiming, shooting, and navigating. The result is best understood as a generated, playable DOOM-like interaction driven by a neural model.
The official GameNGen project page presents the demonstration and its headline performance, while the ICLR 2025 conference paper provides the technical method and evaluation details.
How does GameNGen generate DOOM in real time?
GameNGen uses a two-stage pipeline: first, a reinforcement-learning agent plays DOOM and supplies gameplay data; second, a diffusion model learns to predict the next frame from recent frames and the player’s actions.
Rank #2
- Developed by id Software, DOOM: The Dark Ages is the prequel to the critically acclaimed DOOM (2016) and DOOM Eternal that tells the epic cinematic origin story of the DOOM Slayer’s rage.
- In this third installment of the modern DOOM series, players will step into the blood-stained boots of the DOOM Slayer, in this never-before-seen dark and sinister medieval war against Hell.
- A dark fantasy/sci-fi single-player experience that delivers the searing combat and over-the-top visuals of the incomparable DOOM franchise, powered by the latest idTech engine. With a customizable difficulty system, it’s the perfect entry point whether you’re new to the franchise or a long time fan.
- As the super weapon of gods and kings, shred enemies with devastating favorites like the Super Shotgun while also wielding a variety of new bone-chewing weapons, including the versatile Shield Saw.
- Experience the origin story of the DOOM Slayer’s rage in this epic, cinematic, and action-packed story.
- Gameplay collection: An RL agent learns to play the target environment, and its sessions are recorded.
- Action-conditioned prediction: A diffusion model receives a sequence of previous visual frames together with new player actions.
- Autoregressive generation: The model generates one frame, then uses that generated frame and later input to produce the next frame.
- Stability improvements: Conditioning augmentations reduce error accumulation during repeated generation, while decoder fine-tuning improves visual details and text.
In practical terms, the model repeatedly draws the next screen image rather than asking a traditional renderer to calculate the scene from explicit game state. That design creates the central challenge: every generated frame must preserve enough information about location, objects, enemies, and prior actions for the interaction to remain coherent.
How fast and stable is GameNGen?
According to the ICLR 2025 paper, GameNGen runs at 20 frames per second on a single TPU and remains stable over extended, multi-minute play sessions. The project page summarizes the result as running at more than 20 frames per second on one TPU; the conference-paper wording is the more precise figure for technical reporting.
| Measure | Published result | What it means |
|---|---|---|
| Inference speed | 20 frames per second on one TPU | Interactive visual generation was demonstrated on accelerator hardware. |
| Project-page headline | More than 20 frames per second on one TPU | A rounded project summary; not a universal consumer-device benchmark. |
| Next-frame prediction | PSNR of 29.4 | An image-quality measurement for predicted frames, not proof of pixel-perfect identity. |
| Long-session behavior | Stable over extended, multi-minute play sessions | The generated interaction retained enough continuity for the reported demonstration. |
The 20-frames-per-second result applies to the reported single-TPU setup. It should not be presented as a guaranteed frame rate on a gaming PC, phone, console, or ordinary cloud instance.
How convincing are the generated visuals?
The paper reports a next-frame prediction PSNR of 29.4, but PSNR alone does not establish that the simulated images are identical to frames from the original game. PSNR measures a particular form of image similarity and does not capture every aspect of interactivity, timing, or gameplay correctness.
In another evaluation, human raters were only slightly better than random chance at distinguishing short clips from the real game and simulated clips, even after five minutes of autoregressive generation under the reported experimental conditions. The result suggests that the system can maintain convincing visual continuity for a substantial demonstration, not that every generated frame is indistinguishable or that the model has reproduced the original game’s complete internal state.
Rank #3
- Developed by id Software, DOOM: The Dark Ages is the prequel to the critically acclaimed DOOM (2016) and DOOM Eternal that tells the epic cinematic origin story of the DOOM Slayer’s rage.
- In this third installment of the modern DOOM series, players will step into the blood-stained boots of the DOOM Slayer, in this never-before-seen dark and sinister medieval war against Hell.
- A dark fantasy/sci-fi single-player experience that delivers the searing combat and over-the-top visuals of the incomparable DOOM franchise, powered by the latest idTech engine. With a customizable difficulty system, it’s the perfect entry point whether you’re new to the franchise or a long time fan.
- As the super weapon of gods and kings, shred enemies with devastating favorites like the Super Shotgun while also wielding a variety of new bone-chewing weapons, including the versatile Shield Saw.
- Experience the origin story of the DOOM Slayer’s rage in this epic, cinematic, and action-packed story.
What is the difference between GameNGen and a conventional game engine?
GameNGen replaces much of the demonstrated environment’s conventional game-loop implementation with a learned mapping from visual context and actions to the next rendered frame. A traditional engine explicitly manages state and applies authored rules before rendering the result.
| Dimension | GameNGen | Conventional engine |
|---|---|---|
| Representation | Learned model weights plus training data | Explicit game code, rules, and assets |
| Rendering | Neural next-frame generation | Deterministic or rule-driven rendering pipeline |
| State and consistency | Maintained through learned visual and action context | Maintained through explicit game-state management |
| Compute profile | Neural inference on accelerator hardware, demonstrated on one TPU | Engine execution and rendering on the target platform’s CPU/GPU stack |
| Generalization | Demonstrated for the trained DOOM environment | Designed to support many projects and content configurations |
| Debuggability and control | Model behavior can be difficult to inspect or modify directly | Developers can edit authored rules and inspect explicit state |
This comparison explains both the appeal and the limitation of the approach. Learned weights could encode visual behavior without a manually authored rendering loop, but developers may lose the direct control, inspectability, and portability that established engines provide.
Does GameNGen mean game engines are going away?
No. GameNGen does not show that conventional game engines are about to disappear. The research demonstrates that a neural model can generate an interactive visual simulation for a particular environment, but it does not establish broad support for arbitrary commercial games, production workflows, development costs, or platform deployment.
Traditional engines remain valuable when developers need predictable physics, exact rules, reproducible behavior, efficient asset reuse, tooling, debugging, multiplayer synchronization, accessibility systems, or deployment across diverse hardware. A neural simulator may eventually offer different advantages, but the dossier provides no independently sourced statistic showing development-cost savings or a market-wide effect on the gaming industry.
What are GameNGen’s main limitations?
GameNGen’s main limitation is scope: the reported system was trained from gameplay generated by an RL agent and demonstrated on a specific existing game. The research does not show that a text prompt can create a complete game, that arbitrary games can be simulated, or that the model understands every underlying game rule.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #4
- A Relentless Campaign: There is no taking cover or stopping to regenerate health as you beat back Hell's raging demon hordes
- Return of id Multiplayer: Dominate your opponents in DOOM's signature, fast-paced arena-style combat
- Near-Limitless Gameplay: Doom SnapMap – A Powerful, but Easy-to-Use Game and Level Editor That Allows for Limitless Gameplay Experiences on Every Platform
- Entertainment Software Rating Board (ESRB) Content Description: Blood and gore, intense violence, strong language
- Training dependence: The model’s behavior depends on the gameplay distribution and visual patterns represented in its training data.
- Learned memory: Long interactions require the model to preserve spatial and causal continuity through visual context rather than explicit authoritative state.
- Hardware qualification: The published frame-rate result specifies a single TPU and should not be generalized to consumer hardware.
- Metric qualification: PSNR 29.4 is an image-quality metric, not a claim of complete visual or gameplay identity.
- Production uncertainty: The sources do not provide a commercial-engine benchmark, operating-cost comparison, or readiness study.
Why is GameNGen important despite those limits?
GameNGen matters because it moves generative modeling beyond isolated images and short videos into a setting where every output must respond to user actions and remain coherent over time. A player can immediately expose failures that are easy to hide in a single image: a changed map, a forgotten enemy, an impossible weapon result, or a camera that loses its position.
The reported multi-minute stability indicates that the model preserved sufficient visual and action continuity for the published playable demonstration. That is a meaningful proof of concept for neural simulation, even though it is not a general replacement for authored game logic.
What should readers call the result?
The most accurate description is that GameNGen is a research prototype that generates an interactive simulation of classic DOOM using a neural next-frame model. “AI simulates DOOM without a conventional game engine” is fair when the phrase is qualified by the specific demonstration and hardware conditions.
Descriptions such as “Google released a full AI remake of DOOM,” “AI can now create any game from text,” or “game engines are obsolete” go beyond the evidence supplied by the project and paper.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Frequently Asked Questions
What is Google’s GameNGen?
GameNGen is a Google research prototype that simulates classic DOOM by generating each next visual frame from recent frames and player actions. It does not run the original game in the background, and it is not presented as a commercial remake or general-purpose engine.
Best Value
- DOOM + DOOM II on a region-free physical disc.
- Includes: DOOM, DOOM II, TNT: Evilution, The Plutonia Experiment, Master Levels for DOOM II, No Rest for the Living, Sigil & Sigil II, Legacy of Rust (a new episode created in collaboration by id Software, Nightdive Studios and MachineGames).
- A new Deathmatch map pack featuring 25 maps
- Total of 187 mission maps and 43 deathmatch maps in DOOM + DOOM II
- # of Players: Single System 1-4, Local wireless 1-8, Online 1-16
Is GameNGen actually playable?
Yes, GameNGen is interactively controllable in the reported demonstration. The published paper reports 20 frames per second on a single TPU and stability during extended, multi-minute play sessions.
How does GameNGen generate DOOM in real time?
GameNGen uses an RL agent’s recorded DOOM gameplay to train a diffusion model that predicts the next frame conditioned on visual history and player actions. The model then generates frames autoregressively as new input arrives.
Does GameNGen mean traditional game engines are going away?
No. GameNGen demonstrates a neural simulation for a specific DOOM environment, but the research does not prove that arbitrary commercial games can be recreated this way or that conventional engines are generally obsolete.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesThe Bottom Line
GameNGen shows that a neural model can generate a controllable DOOM simulation at about 20 frames per second on a single TPU and remain stable for multi-minute sessions. It is an important research proof of concept, but the evidence does not establish a general-purpose, commercial replacement for conventional game engines.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
