Google’s GameNGen demonstrated a playable neural simulation of classic Doom: a diffusion model generated each new frame from recent frames and the player’s actions, rather than a conventional engine rendering polygons and executing the original game code. In short human-tested clips, evaluators were only slightly better than chance at telling GameNGen footage from original Doom. That is impressive evidence of visual similarity—not proof that an entire game, its rules, or its executable was perfectly recreated.
The project was publicly reported in August 2024 and published as an ICLR 2025 paper, so it is research rather than a new commercial release in 2026. See the GameNGen project page and paper.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
DOOM Eternal: Standard Edition - PlayStation 4 | $27.49 | Buy on Amazon |
| 2 |
|
DOOM: The Dark Ages – Xbox Series X | $39.95 | Buy on Amazon |
| 3 |
|
DOOM: The Dark Ages – PlayStation 5 | Buy on Amazon | |
| 4 |
|
Doom - Xbox One | $27.99 | Buy on Amazon |
| 5 |
|
DOOM + DOOM II (Limited Run Games #144) - for Playstation 5 | $44.62 | Buy on Amazon |
What GameNGen actually built
GameNGen is best understood as a learned visual simulator. It predicts the next game frame from a short history of previous frames plus the player’s input, then feeds that prediction back into the next step. Repeating the process creates an interactive sequence instead of a passive video.
The system does not reproduce Doom’s source code, renderer, physics engine, or complete symbolic game state. It learns the visual consequences of actions from gameplay data. “Without a game engine” therefore means that conventional rendering and simulation were replaced by a neural model; software infrastructure and specialized hardware were still required.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute#1 Best Overall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
How training worked
- An reinforcement-learning agent learned to play Doom.
- The researchers recorded gameplay trajectories, including images and actions.
- A diffusion model learned to predict subsequent frames from that information.
- Training and conditioning methods were used to make long, autoregressive rollouts more stable.
This is not a general AI being prompted to invent Doom from scratch. The model was trained on an existing game and interaction data from an agent playing it.
Is it really playable?
Yes. The project shows people controlling Doom simulated by GameNGen, rather than merely watching a pre-rendered movie. Google reports more than 20 frames per second on a single TPU. A contemporary Heise report described approximately 50 frames per second at reduced visual quality; that figure is secondary reporting, not the project’s principal published operating point.
“Playable” establishes responsive interaction and visual continuity in the demonstration. It does not establish compatibility with the original executable, mods, network play, save files, deterministic replays, or every feature of a commercial Doom port.
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.
What the “indistinguishable” claim means
The human evaluation tested short clips of original Doom and GameNGen output. Raters performed only slightly better than random chance when asked which was which, according to the ICLR 2025 paper. That supports a careful statement: GameNGen clips were difficult to distinguish from original footage under that specific test.
It does not show that unlimited play is indistinguishable, that the underlying rules are identical, or that the model is ready to replace production engines. A brief clip can look convincing while concealing state errors, accumulated drift, or unusual-input failures.
GameNGen versus a conventional game engine
| Capability | Conventional engine | GameNGen-style neural simulation |
|---|---|---|
| World state | Stores entities, positions, health, inventory, and rules explicitly. | Predicts visual outcomes from recent images and actions; complete state is not inherently exposed. |
| Rendering | Calculates geometry, textures, lighting, particles, and animation. | Generates the next image directly. |
| Physics and collisions | Designed systems provide explicit collision and physics behavior. | No inherent guarantee of exact collision or physics consistency. |
| Repeatability | Can support deterministic runs, saves, and replays when designed to do so. | Exact repeatability is not guaranteed by frame prediction alone. |
| Content pipeline | Supports scripting, level tools, debugging, modding, and certification workflows. | Those capabilities are not supplied by the simulator itself. |
| Generalization | New levels and mechanics can be authored within the engine. | New content or mechanics may require additional training or adaptation. |
“AI-generated game” can refer to AI-made assets inside an ordinary engine, AI-written code for an ordinary game, or a neural model that directly generates frames. GameNGen is primarily the third category.
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.
Why the result matters technically
Real-time neural rendering
More than 20 frames per second on one TPU shows that a diffusion-based environment model can operate interactively rather than only offline. The hardware context matters: one TPU is not the same as running locally on an ordinary gaming PC.
Longer rollouts
The researchers report stable sessions lasting multiple minutes. Maintaining coherent visual behavior over that duration is harder than predicting one isolated frame, because every generated frame becomes input to the next prediction.
Free tools Windows power users keep installed
One-click scans. No signup required.
Pixel similarity, with an important caveat
The project reports a next-frame PSNR of 29.4, described by the authors as comparable to lossy JPEG compression. PSNR measures pixel-level similarity; it does not measure input latency, game rules, fun, accessibility, determinism, or whether an enemy, door, or item remains logically consistent.
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
Where a neural simulator can fail
- Limited visual memory: A short temporal context is not the same as a complete map and state database.
- State drift: Small prediction errors can accumulate during autoregressive play.
- Inconsistent revisits: A room, switch, weapon, or enemy may not remain exactly as it was after the player leaves and returns.
- Unusual inputs: Rapid turning, uncommon routes, or situations underrepresented in training can expose weaknesses.
- Latency and quality trade-offs: Higher frame rates may require lower visual quality, and responsiveness depends on the deployment setup.
- Narrow benchmark: Classic Doom is comparatively constrained; success there does not demonstrate readiness for modern open worlds, multiplayer games, or unfamiliar mechanics.
- Reproducibility and rights: A commercial system would need predictable behavior, suitable hardware economics, and answers about training-data and output licensing for a copyrighted game.
Useful tests go beyond asking whether a clip looks real: revisit the same room, check whether doors and ammunition retain state, repeat an identical input sequence, play for several minutes, try unexpected interactions, and measure input-to-frame latency.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What this approach could enable
Neural environments could be useful for rapidly prototyping interactive scenes, generating varied worlds, training AI agents, or creating personalized game-like experiences. They may also offer a way to simulate environments when building and rendering a fully coded world is expensive.
Those are implications, not products delivered by GameNGen. A conventional engine remains the safer choice for competitive multiplayer, deterministic esports, precise physics, modding, persistent campaigns, save/load systems, accessibility tooling, debugging, console certification, and predictable performance.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
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
Availability in 2026
Nothing in the GameNGen project materials establishes a downloadable consumer version of the Doom simulator or a general-purpose neural game engine. Readers should treat it as a research demonstration.
Google’s later projects are related in theme but are not GameNGen releases. Project Genie is described as an early prototype for creating and exploring generated worlds, with access indicated for Google AI Ultra subscribers in the United States who are at least 18. It is not a GameNGen download or a faithful Doom clone. Genie 3 is presented as a separate world-model research direction.
Google Antigravity is an agentic coding environment, not a neural-frame simulator. Google’s post gives an example estimate of $916.92 in API usage for a multi-agent FreeDoom operating-system experiment and describes a $200-per-month Google AI Ultra plan for a preview feature at the time of that post; prices and availability should be checked on official pages before relying on them.
The bottom line on “AI-generated Doom”
GameNGen is a meaningful demonstration that a neural model can generate an interactive, visually convincing simulation of Doom in real time. The evidence is strongest for short-clip visual plausibility and demonstrated interactivity. It is not evidence that Google recreated the original executable, guaranteed the same game state, or made conventional engines obsolete.
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.
