Making AI Games? Learn These 34 Jargon Terms First
Nicolas Zullo made a free game development terminology cheat sheet, with 34 concepts plus illustrations and example prompts. Meta Alchemist suggests feeding it directly to an LLM and having it audit your game against the checklist.
Nicolas Zullo has released a free website gamedev-cheatsheet.com, which collects 34 development concepts that affect gameplay, each with an illustration and an example prompt. The site navigation is divided into 3D & rendering, Game feel, Systems, and Performance; you can already see terms like Raycasting, Normals, and Level of detail. At the top, you can leave your email to subscribe to new concepts. Nicolas says it is 100% free.
This solves a very specific problem. More and more people are making games with AI; models can write code and generate assets, but many people can't clearly articulate what they want. Nicolas says he has been hardcore prompting AI for three years, mainly in game development. His judgment is that the biggest lever is knowing the vocabulary, concepts, and fundamentals of game design. The difference in output can be huge. That judgment isn't mystical. If you say 'make the visuals look better,' the model can only guess; if you say 'check whether normal direction and lighting are consistent, and whether LOD transitions pop,' it has somewhere to start.
Meta Alchemist's tweet offers a more practical use. He puts the website link directly into the prompt and asks the LLM to comprehensively audit the game from the standpoint of these terms and systems, plan and execute everything that needs improvement, and then show before-and-afters on localhost and visually. The gist of the original wording is: 'can you audit everything in our game from the standpoint of these terms and systems comprehensively... plan and execute everything that needs improvement, and show me before-and-afters on everything in localhost and through visuals'. This is much more specific than asking the model to 'help me optimize the game'. Here the glossary works like a checklist: the model goes through it item by item, and the human accepts it item by item.
But don't mistake an LLM audit for a playtest. A commenter, Adam P, said in a reply that an LLM will happily tell you your coyote time is fine; playtesters won't. Coyote time is the grace period during which a character can still jump after running off a platform edge. Whether it feels right ultimately still has to be checked by a human with a controller. Terminology can help you describe the problem clearly, but it can't replace testing.
The examples on the site are also suitable to be rewritten directly into prompts. Raycasting is about casting an invisible ray from an origin along a direction to detect collisions or interaction points. Normals describe surface orientation and affect lighting and highlights. Level of detail is about using different polygon densities at different distances. There are many more under Game feel and Systems. Each concept has an illustration, making it friendly to people without a graphics background. For people with a development background, it's more of a reminder: don't focus only on code; the vocabulary of gameplay and rendering also affects model output.
In the replies, someone just wrote 'Z fighting' and paired it with a Dragon Ball Z running animation. Z fighting is the flickering caused by depth conflicts when two surfaces are too close together; it's not the same thing as Goku running, but it works well enough as a memory hook.
A commenter said they were vibecoding a game and thought this cheat sheet would be useful.
Meta Alchemist himself said he bookmarked it first, then actually put it in his browser, used it three times, and thought it would be helpful. Bookmarking doesn't equal using. The cheat sheet is free right now, and the cost of opening it is very low. Start by using it to audit the demo you're working on and see whether the model can point out three problems you hadn't noticed before.
发布时间: 2026-11-02 04:44