Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Yes: a platformer can offer one million distinct levels without storing one million maps. The game stores a generator, physics rules and a numeric seed; the generator builds a level when needed, while a reachability system rejects layouts the player cannot complete. Cloudberry Kingdom demonstrates the physics-aware search approach, and Spelunky demonstrates a more compact template-and-solution-path approach. In both cases, the difficult part is not producing enough combinations. It is making every combination readable, beatable and appropriately difficult.
Why is one million levels mainly a quality problem?
A million is a small target for a combinatorial generator. Jordan Fisher, Cloudberry Kingdom’s lead programmer, wrote that “The level tree is massive. Each node has millions of children” (Game Developer, 2012). If each decision can choose among block types, widths, moving phases, ceilings, hazards and decoys, the number of complete paths through that tree quickly exceeds one million.
As an Amazon Associate I earn from qualifying purchases.
Free tools Windows power users keep installed
One-click scans. No signup required.
The hard constraint is quality. A random platformer can be unique and still be impossible, tedious or unfair. Fisher reduced the design requirement to three tests: “There are three things a good procedural algorithm needs to nail:” feasibility (can you beat it?), interesting design (do you want to beat it?) and appropriate skill level (is it a good challenge?). Those tests interact: adding a fireball can make a jump impossible, while removing a safe landing can turn a visually busy section into an easy one if the player can simply run underneath it.
Sampling random maps until one happens to work is inefficient because the set of provably playable maps is much smaller than the set of all maps. A useful generator constructs a candidate around movement constraints, proves or strongly constrains a route, then adds variation that cannot invalidate it.
Popular on Amazon
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
What makes platformer generation harder than maze generation?
A maze can usually guarantee connectivity by checking whether two cells are linked. A platformer must guarantee a sequence of physical actions. Gravity, horizontal momentum, friction, jump height and variable jump timing determine whether the character can reach a platform, not merely whether a platform exists.
- A gap that is reachable from a standing jump may be unreachable after a forced landing or a fireball hit.
- A moving platform can occupy the correct destination at the wrong time, intersecting the jump arc.
- A ceiling can make a previously legal full jump require a shorter jump.
- An obstacle near the end can invalidate an earlier route if it leaves no safe state for the next section.
That dependency is why feasibility must be checked during construction rather than after a complete map is assembled. A generator needs a model of the player’s state, including position, velocity and jump phase, and it must update reachable states as geometry changes.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
How does Cloudberry Kingdom’s AI generate beatable levels?
Cloudberry Kingdom uses a design AI to propose geometry and a player AI to prove that the proposed geometry can be traversed. The method searches a huge space instead of drawing a finished map and hoping it works.
1. Treat the level as a search tree
The empty level is the root. Each child adds one block or structural element. A depth-first walk chooses a promising child, extends it and backtracks when the branch cannot remain feasible. The child choices can include block type, width, bouncy or moving behavior, orbit phase, ceiling placement and decoy platforms.
2. Use a heuristic to avoid blind branching
Because each node has millions of possible children, the search orders candidates with heuristics. The aim is not to enumerate every map; it is to find a legal, interesting branch quickly, then continue refining it. Backtracking can remove several recent decisions when a later constraint exposes an earlier mistake.
3. Ask a player model whether the next block is reachable
The player AI precomputes destinations from a standing character, then tracks different movement states: standstill versus full-speed running, and full versus half jumps. Those states produce different maximum gaps. When a new block is proposed, the AI checks whether the player can reach it and whether the new geometry destroys the already-valid prefix.
This is a physics-aware proof, not a topological path check. Altering gravity, jump height or friction changes the reachable-state model, so the same construction system can generate a different family of legal levels.
4. Add danger after the route exists
Once a feasible skeleton is found, the generator can add hazards that do not intersect the player AI’s successful path. Fisher also restricted the AI’s behavior to shape the game’s feel: never running forward below half speed encourages flow, while limiting the movement range creates more convoluted routes.
Also popular on Amazon
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
5. Tune for a body, not an abstract avatar
Cloudberry Kingdom’s physics and hero factory lets the generator rebuild levels for different characters. A high-jump, low-friction character and a jetpack character with a low maximum fall require different geometry. The game can therefore combine a character’s movement model with a fresh level rather than treating the avatar as a cosmetic swap.
What did the development effort reveal?
Fisher said the AI work began two years before graphics and gameplay were added (GamingLives, 2013). Four or five months taught it to make beatable levels; the remaining work taught it to make levels with the intended difficulty. He later said the project had “hundreds of parameters” for interestingness and “thousands of parameters” for difficulty (Game Developer, 2012).
Those parameters include gap distance, required take-off position, full or half jump, fireball count, speed and direction, emitter synchronization, safe spots and whether continuous running is required. A genetic algorithm was tried, then abandoned because an opaque optimizer could damage the delicate feasibility constraints. Three years and thousands of man-hours of beta testing eventually mapped that coupled parameter space onto a continuous difficulty rating, from easy to humanly impossible (Game Developer, 2012).
Pwnee Studios identified the remaining difficulty problem precisely: obstacle count is not a reliable score. “A level can have a million obstacles, but if none of them can even touch you, that makes it easy,” the studio told videogametourism.at (2014). Death-rate feedback could adjust difficulty, but changing it during a level weakened the player’s sense of accomplishment, so the practical design kept a level’s challenge stable.
How are levels delivered to the player?
The game can generate and validate the next level in the background while the player is still on the current one (GamingLives, 2013). The same player AI can demonstrate a solution route. Fisher reported seeing the AI fail only twice in four years of development, while stressing that the game’s potential catalog has no published exact integer count. His memorable description was “Mario On Speed.”
Story mode uses selected or baked procedural levels. Arcade Escalation provides a Tetris-like, effectively unbounded difficulty ramp; Time Crisis adds a clock; and Free Play exposes controls for length, clouds, falling blocks, elevators, swinging maces, fireballs and difficulty. These modes are possible because the generator exposes parameters instead of a fixed collection of authored maps.
How does Spelunky generate levels without searching raw geometry?
Spelunky takes a different route: it guarantees a coarse route through a grid, fills rooms from authored templates, then lets procedural traps, monsters and items create local surprises. The original game’s campaign contains 16 generated levels, and each level uses a 4×4 room grid.
1. Establish the room graph
The entrance is on the top row and the exit on the bottom. Room types encode how the solution can continue: a type-0 room is off the path; type 1 connects left to right; type 2 connects left, right and bottom (and can connect upward when the room above is also type 2); type 3 connects left, right and top. This creates a navigable route before individual traps are considered.
Also popular on Amazon
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
2. Select and vary a room template
Each room is 10 tiles wide by 8 tiles high: 80 tiles represented as an 80-character string. Darius Kazemi described about 50 basic layouts for each tileset (2009). The room’s static structure is approximately 80% hand-authored and 20% randomly generated, with special characters acting as hooks for variation.
3. Run separate obstacle and monster passes
Spikes, enemies and items are placed after the route and terrain exist. Local geometric predicates can ask whether a brick is beneath a ledge, whether a space is large enough for a creature or whether a trap would block the guaranteed route. Off-path rooms can be intentionally isolated; reaching them may require bombs, which turns optional exploration into a resource decision.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC 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 & 11Kazemi’s source-code reading gives the Giant Spider rule as a 1-in-40 chance under qualifying bricks, followed by a lockout that prevents additional spiders in that level (2009). A later chapter of Procedural Content Generation in Games summarizes the chance as 20% and says it becomes zero after the first spawn. Those figures conflict; Kazemi’s original code reading is the closer account of the implementation, while the textbook number should be treated as a later summary rather than a correction.
Why do a small number of rooms feel larger?
Players learn the recurring room shapes but cannot predict the trap and creature combinations. Derek Yu described the goal as giving each component a simple, readable “personality” and allowing those pieces to combine into complexity (Game Developer, 2017). A frog, piranha or monkey is understandable alone; the same creature crossing a randomly assembled floor plan can create a new timing problem. Mike Cook called Spelunky “the most beautiful and perfect application of procedural generation” because it balances a creator’s design rules with a digital apprentice (Baker, 2016).
Spelunky 2 extends the idea with liquids and particles, including lava and water, and a second layer reached through a door, increasing the explorable space without abandoning the route-and-template foundation (PCMag, 2020).
How do the two architectures compare?
| Architecture | Solvability guarantee | Authored content | Typical feel | Difficulty control |
|---|---|---|---|---|
| Cloudberry Kingdom | Player-AI reachability proof during a physics-aware search | Raw geometry searched and assembled from many parameters | Precision arcade flow, described by Fisher as “Mario On Speed” | Thousands of tuned variables collapsed to a continuous difficulty rating |
| Spelunky | Topological solution path through a 4×4 room grid | About 50 room layouts per tileset, roughly 80% authored and 20% varied | Exploratory roguelike platforming with emergent trap combinations | Area, enemy and trap rules rather than one universal slider |
| Infinite Mario Bros. section assembly | Template connectivity, but less comprehensive physical proof | Five section templates: straight, hill, tubes, jump and cannons | Fast, recognizable Mario-like sections with limited border emergence | Light random variation and section choices |
The choice depends on the game’s promise. Use Cloudberry’s approach when exact jump timing and a wide difficulty continuum are central. Use Spelunky’s approach when recognizable authored spaces, optional routes and interaction between simple entities matter more than proving every possible movement state.
Recommended Free Tools
How can you build a platformer with one million levels?
A practical implementation can combine both families: guarantee a route with chunks or templates, then use a physics simulator and critics to reject unsafe variations.
- Define the movement contract. Record gravity, acceleration, maximum speed, friction, jump impulse, variable-jump behavior, collision box and abilities such as wall jumps or jetpacks. A generator cannot reason about a jump it cannot quantify.
- Represent player states. Track at least position, horizontal velocity, vertical velocity, grounded status and jump phase. Precompute reachable landing intervals for common states such as standing, full-speed running, full jump and half jump.
- Create a guaranteed macro-route. Choose either a room graph like Spelunky’s, authored chunks anchored at potential player positions, or a sequence of action rhythms. Every route segment should expose an entry state and an exit state.
- Generate geometry around those states. Place platforms, ceilings and moving elements only where the reachable intervals permit them. For a moving platform, include its phase and speed in the state check rather than testing its position at one instant.
- Run a player simulation after each extension. If no valid state reaches the new segment, backtrack and choose another child. Do not wait until the whole level is complete; late rejection wastes the search effort and can hide the decision that caused the failure.
- Add hazards with local safety predicates. Test fireball trajectories, enemy envelopes, trap activation and safe landing space against the proven route. Keep hazards that create choices or timing challenges, and reject those that seal every route.
- Separate variation from structure. Keep hand-authored chunks responsible for recognizable composition, while randomizing decoration, enemy timing, item placement and alternate branches. This preserves design intent without making every run predictable.
- Score the result on several axes. Measure route length, required jumps, recovery opportunities, timing windows, hazard density, optional rewards and the number of valid player states. A single obstacle count cannot represent difficulty.
- Freeze the seed and version the rules. Once a level is published, store its seed and the generator version. Rebuilding with the same rules should reproduce the same map, while a new generator version should be treated as a different catalog.
What is a seed, and why does it make a million levels practical?
A seed is a number used to initialize the generator’s pseudorandom sequence. Given the same seed, algorithm and content data, the game can rebuild the same level on demand. The saved object is therefore a compact recipe rather than a complete tile map.
Also popular on Amazon
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Widely Compatible: Transform any USB-C port into a USB-A port and USB-A port into a USB-C port, and connect up a wide range of USB-C(including MacBook, iPad Pro/Air/Mini, iPhone models with USB C ports, Samsung Galaxy series, and Google Pixel series. ) and USB A devices(including external hard drives, phones, mice, printers, and more)
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
This is why players can share seeds in games such as Minecraft, and why a small download can produce thousands or millions of arrangements (PCMag, 2020; GameHaunt, 2025). A seed does not guarantee quality by itself: the generator still needs constraints and validation. It only makes storage, sharing and replay practical.
No Man’s Sky illustrates the scale of seed-based catalogs, although it is not a platformer: a 64-bit seed permits 18,446,744,073,709,551,616 possible worlds, and visiting each for one second would take 585 billion years (GameHaunt, 2025). Chris Baker reported that a team of 13 built that universe (2016). A million platformer levels is a much smaller use of the same compression principle.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →How do you tune difficulty without making levels unfair?
Difficulty is a high-dimensional relationship between the player, the movement model and the level’s timing. A level with many hazards can be harmless if they never intersect the route; a nearly empty level can be brutal if one jump demands a precise edge take-off.
Use measurable demands, not decoration counts
- Required horizontal distance and the width of the landing window.
- Whether the player must begin from rest, full speed or a particular jump phase.
- How many consecutive actions are needed without a safe reset.
- Hazard speed, direction, synchronization and overlap with the player’s arc.
- Availability of recovery platforms, alternate routes and readable warnings.
Map many controls to one player-facing rating
Cloudberry Kingdom’s team spent three years and thousands of man-hours turning coupled variables into a continuous rating from easy to humanly impossible (Fisher, 2012). The lesson is architectural: expose one difficulty slider to players, but keep the underlying generator free to adjust many correlated parameters. A slider should change demands such as timing windows and safe-space frequency, not simply multiply the number of enemies.
Keep the rating stable during a stage
Adaptive systems can use death rate to select a later level, but changing a level’s challenge while the player is inside it can make success feel arbitrary. A better contract is to choose the difficulty before generation, validate the resulting map, and preserve that contract until the stage ends.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why do random platformer levels feel unfair, and how do you fix that?
Unfairness usually comes from a mismatch between what the generator proves and what the player actually needs. A route may exist in a geometric model but depend on an untelegraphed moving-platform phase, a collision-box edge case or a jump state the game never teaches.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches- Model the shipped physics. Use the same gravity, friction, collision shapes and variable-jump rules in validation and gameplay.
- Validate timing, not only destinations. Moving hazards and platforms need phase-aware simulation.
- Reserve reaction time. Avoid placing a hazard immediately after a forced landing unless the player can see it before committing.
- Provide recovery states. A missed optional challenge should not automatically erase the only route forward.
- Teach each rule before combining it. Introduce a creature, trap or movement demand in isolation, then allow procedural overlap after the player has learned its behavior.
- Replay failed seeds. Deterministic reconstruction lets designers inspect the exact sequence that caused a bad outcome instead of chasing a one-time random event.
What other procedural methods are useful?
Rhythm groups
Rhythm-group systems first generate a sequence of player actions, convert those actions into geometry, stitch groups together and score the result with critics. Designer controls can include path shape, rhythm types, geometry frequency, coins, jumps per group and beat pattern. This is useful when the intended experience is a sequence of movement phrases rather than a collection of rooms.
Occupancy-regulated extension
Mawhorter and Mateas’s occupancy-regulated extension grows a level from human-authored chunks anchored at potential player positions. The prototype library contained 42 chunks, with the largest measuring 10×10 tiles (2010). It prioritizes variety; playability is not guaranteed without an additional validation pass.
Also popular on Amazon
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Noise, grammars and cellular methods
Perlin noise, Simplex noise, Wave Function Collapse, L-systems, cellular automata, binary-space partitioning and grammars are broad procedural tools. For a 2D platformer, they are most useful for decoration, biome shape or constrained chunk choice. They do not replace a movement-aware solvability check.
What does “one million levels” really mean?
It means the generator can produce at least that many distinct outputs under its chosen seed range and rules; it does not mean a developer authored one million complete files. Spelunky’s roughly 50 room layouts per tileset can produce a large number of campaigns because room order, orientation, trap placement, enemies and items multiply the combinations. Cloudberry Kingdom searches a much larger raw geometry space, but no published source gives an exact integer count of unique Cloudberry levels. Fisher said he had not done the calculation and described the potential catalog as larger than the number of particles in the known universe (GamingLives, 2013).
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Cloudberry Kingdom launched on PlayStation 3, Xbox 360, Wii U and Windows in 2013. Its Steam version was delisted in May 2020; RedoApps released versions for PlayStation 4 and PlayStation 5 on 12 December 2025 and Nintendo Switch on 2 February 2026 (Wikipedia, 2026). Those dates describe availability, not the size of the procedural catalog.
What should a developer verify before shipping?
- Can the validator reproduce the exact physics and collision behavior used by the game?
- Does every generated route include a recoverable entry and exit state?
- Are moving hazards checked across their full cycle rather than one sampled frame?
- Can the same seed be rebuilt after saving, sharing or restarting?
- Are generator and content versions recorded so old seeds remain interpretable?
- Does the difficulty score account for timing, movement state and safe space?
- Are authored templates carrying composition while procedural systems supply variation?
- Can a player understand why a failure happened, even when the layout is new?
The central design principle is simple: use procedural generation to multiply authored rules, not to remove design. A guaranteed route supplies feasibility, readable templates supply identity, and physics-aware validation keeps the million-level promise from becoming a million ways to fail.
Frequently Asked Questions
Can two players play the same procedural level together?
Yes, if both systems use the same seed, generator version, physics settings and content set. Networked games also need synchronized random-number usage; an extra random call on one machine can otherwise produce divergent layouts.
Do procedural levels need to be generated online?
No. A deterministic generator can build them locally from a seed, which supports offline play and compact save files. Online services are useful for sharing seeds, leaderboards or moderation, but they are not required to create the map.
Can hand-authored platformer levels still use procedural generation?
Yes. A common hybrid keeps the critical route and visual grammar authored, then randomizes safe decorations, enemy timing, rewards and optional branches. This retains a designed rhythm while preventing exact memorization.
What happens when a generator finds no level at the requested difficulty?
It can relax secondary constraints, choose another macro-route, widen the allowed movement states or fall back to a validated template. The important rule is to change the search space deliberately rather than silently shipping an unverified map.
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.
