Geplantes Veröffentlichungsdatum: Januar 2027

Infos zum Spiel

Draw a ladder. Get a noodle.

You draw what you need on a 32-pixel grid. A neural network inside the game looks at it and decides what you meant.

It is not very good at this.

That is the game

The game never checks what you drew — it checks what the thing can do. You need something to stand on: a bench works, a couch works, an anvil works, and so, it turns out, does the noodle.

So when your careful ladder arrives as a hot dog, the question stops being how do I draw a better ladder and becomes what can I do with a hot dog. The answer is usually: something.

Every stage has at least three ways through. It has to — the game cannot assume you got the object you asked for.

100 objects, all with opinions

Every object you use is something you drew a moment ago. It arrives with its own weight, its own physics and its own uses.

• Ladder — throw it and it locks where it lands. Climb it, or bridge a gap with it

• Anvil — heavy. Pins a pressure plate, and you can stand on it

• Watermelon — explodes on impact when thrown, or on a one-second fuse if you set it down

• Umbrella — a grappling hook. Swing from a ceiling

• Leaf — let go of it and it rises instead of falling

• Boomerang — flies straight, ignores gravity, and comes back

You will get some of these on purpose.

Ink

Drawing big costs ink, and the price climbs faster than the size does. The tank refills slowly. So: a bridge all the way across the gap, or one small platform to jump from? Worth thinking about before you spend half a tank on something that comes out as a pretzel.

What's in it

• 25 stages across five caves

• 100 drawable objects, each with its own behaviour and its own uses

• 22 achievements, including one for summoning every object in the drawing collection

• Two-player co-op — join with a room code and share one ink tank.

• A stage editor, and Steam Workshop to trade what you build

• English, 한국어, 日本語, 简体中文, Español, Português (Brasil), Русский

The model runs on your machine

Recognition ships inside the game. No account, no API key, no server, no waiting. The only place a drawing ever travels is your co-op partner's screen.

A small robot announces each drawing out loud as it appears. Sometimes it hedges. It should hedge more.

"Ah. Couch. Obviously." — on being shown a bench

Music by Kevin MacLeod , licensed under Creative Commons: By Attribution 4.0.
Doodle recognition trained on the Quick, Draw! Dataset by Google Creative Lab, licensed under CC BY 4.0.
Full license notices are in the game: Esc → Credits.

Offenlegung von KI-generierten Inhalten

Der Spieleentwickler beschreibt den Einsatz von KI-generierten Inhalten in diesem Spiel wie folgt:

Close Enough was built with generative AI. The code, the in-game text, the level layouts and all seven translations were written by an AI assistant (Claude) working to the developer's direction, and most of what you see on screen — the objects, the whole interface — is drawn by that code rather than loaded from image files.

The doodle recognition inside the game is not generative. It is a small image classifier bundled with the game, trained on Google's Quick, Draw! dataset of human-drawn doodles. It runs entirely on your own computer, sorts each drawing into one of 100 pre-defined objects and creates nothing. Your drawings are never uploaded, and the release build contains no code that could upload them.

The sprites, tiles, music, sound effects and font are licensed assets made by people, not by AI. Every one of them is credited inside the game under Esc → Credits.