This glossary explains technical terms related to Artificial Intelligence used in Celestory and its documentation. Definitions are intentionally simplified to be accessible to everyone. --- ## A AI Agent (of the Graph) The name given to the Text tab chat once it becomes able to act. It is not a separate mode: as soon as the chosen model can use tools, every message triggers a "turn" during which the assistant adds blocks, wires them, writes dialogue and creates variables in the graph you have open. With a non-capable model, a banner warns you and the chat stays a plain conversation. AI Credit The unit measuring AI generation consumption in Celestory, written "©" in the interface. The monthly quota depends on the subscription: 1 credit (free trial), 5 (Personal account / Pro), 30 (Organization account / Business). Careful — a credit is not "one action": a Text tab chat message costs ¼ of a credit (4 messages = 1 credit), an Image/Video/Audio generation from the editor costs 1 credit, an in-game AI text request costs 0.01 credit and an in-game AI image request 0.05 credit. The © badge next to a model indicates its relative cost, but it is not what the monthly counter deducts. Failed generations are not charged. Requests triggered by players consume the creator's credits, never the player's. API (Application Programming Interface) Communication channel between two software programs. When Celestory sends your prompt to an AI model and receives a response, it uses an API. You don't need to understand how it works: Celestory handles it for you. Asset Graphic, sound, or text resource used in a project: character image, background, interface icon, sound effect… In AI, generating an asset means creating this resource automatically from a description. Audio Avatar Video automatically generated from a character (still portrait) whose lips and facial expressions move in sync with an audio file. Allows creating speaking characters without an actor or camera. --- ## C Competency Prompt (System Prompt / Role) Basic instructions given to the AI before the conversation, describing its role, constraints, and behavior rules. Example: « You are a sarcastic old wizard sage. You always answer in verse. You never give a direct answer. » The more precise the competency prompt, the more consistent the model will remain. Context → see Context Window Context Window The maximum amount of text an AI model can process in a single interaction, expressed in tokens. Think of it as the model's "working memory": beyond that limit, the model starts to "forget" the oldest exchanges in the conversation. Conversational Chat Textual discussion interface (like a messaging app) integrated into Celestory to interact with textual AI models. The history of exchanges is kept so the AI can keep track of the conversation. --- ## D Diffusion / Diffusion Model Type of AI architecture used to generate images and videos. The model starts with random noise (like snow on a TV screen) and progressively refines it until a coherent image corresponding to the prompt is obtained. Flux, LTX-2, Veo, Kling are diffusion models. --- ## F fal The platform through which Celestory routes **all** of its AI calls — text, image, video and audio, in the editor as well as in-game. For text, fal itself routes the request on to OpenRouter. Practical consequence: no provider API key lives in your project, and Celestory carries the billing. The "More info on this model" link next to the model picker opens the model's technical page at fal. FPS (Frames Per Second) Number of still images displayed per second in a video. 25 FPS = classic cinema style. 50 FPS = very fluid video. The higher the FPS, the smoother the video appears, but the larger the file size. Frame A still image that makes up a video. A 10-second video at 25 FPS is made up of 250 frames. In AI video generation, the "first frame" is the starting image and the "last frame" is the ending image. --- ## G Generation (in the Celestory context) The act of asking an AI model to create content (text, image, video, audio). Each generation consumes AI credits from the creator's monthly quota — see AI Credit for the breakdown. Image, Video and Audio generations are asynchronous: you retrieve them in the Generations tab, which does not refresh by itself. Graph (in Celestory) The main authoring space in Celestory, where you connect blocks to build your game or application logic without writing code. --- ## H HTML5 Web programming language (HyperText Markup Language, version 5). Used for creating interactive interfaces in browsers. In Celestory, the HTML5 block allows you to insert a mini-web page or animation directly into your game. With vibe coding, you can ask the AI to write this code for you in plain English. --- ## I Image Mask Black and white image overlaid on a photo or illustration to tell the AI which area to modify (black area = to regenerate, white area = to keep). Used for inpainting. Inpainting Technique consisting of regenerating only a specific area of an existing image, leaving the rest intact. Example: correcting the background of a portrait without modifying the character in the foreground. --- ## J JSON (JavaScript Object Notation) Text-based data structuring format widely used in computing. Looks like: {"key": "value", "number": 42}. In Celestory, the advanced mode of AI generators allows for entering parameters in JSON for full control over the request. --- ## L LLM (Large Language Model) Artificial Intelligence program trained on immense amounts of text to understand and generate human language. Examples: GPT-5, Claude, Mistral, Gemini, Llama, Grok. An LLM can write, summarize, translate, code, analyze, and answer questions. Loop Animation or sound designed to repeat indefinitely and smoothly, without an audible or visible break between the end and the beginning. A background music loop plays continuously throughout a game scene. --- ## M MCP (Model Context Protocol) A shared language that lets an online service (a database, an internal tool, a third-party service) declare "here are the actions you can ask me to perform", and lets an AI use them. In Celestory, MCP appears in two places: the **MCP block** in the graph, and the **Tools** tab of the AI Generation window, which equips the chat assistant. 🔺 Authentication there is manual (no authentication, API token or JSON headers): there is no one-click OAuth sign-in. Markdown Simple and lightweight text formatting format, readable both in plain text and graphically rendered. Examples: - # Title → Title - **bold** → bold - *italic* → italic - - item → • item - [link](url) → clickable link Widely used for documentation, notes, and AI chat interfaces. In Celestory, AI text responses are rendered in Markdown. Model / AI Model Artificial Intelligence software trained to perform a specific task (generating text, images, videos, audio). Different models have different strengths in terms of quality, speed, and cost. Motion Control AI video generation mode where a character's movements are dictated by a reference video. You provide a character image and a video of a real person moving, and the AI animates your character reproducing those movements. --- ## N Nocode Approach to developing applications without writing programming code, using visual interfaces (blocks, connectors, forms). Celestory is a nocode platform: you build your game by assembling blocks rather than coding. Negative Prompt Instruction given to an AI model to tell it what it must not include in the generation. Ex: « no text, no blur, no deformed hands ». More common in image and video generation than in text generation. NPC (Non-Player Character) In a game, any character controlled by the program rather than by the player: merchants, enemies, guides, witnesses… AI makes it possible for NPCs to hold dynamic dialogue that was not scripted in advance. --- ## O Open source Software whose source code is published freely and modifiable by everyone. In AI, an open-source model (like Meta's Llama or Black Forest Labs' Flux) can be downloaded, modified, and used without depending on a commercial service. OpenRouter API service that acts as a « router » to several different AI models (OpenAI, Mistral, Anthropic, Meta, Google…). Celestory reaches it **through fal** for all text generation — the editor chat as well as the in-game AI text request block — allowing access to many models through a single interface. --- ## P Prompt Instruction or question you send to an AI model to trigger a generation. The quality of the prompt is decisive for the quality of the result: a precise prompt (style, context, constraints) gives a better result than a vague prompt. --- ## Q Quota AI credit consumption limit defined by the creator's subscription plan. Shown at the top of every AI Generation tab as "Monthly usage — X/Y ©", with the reset date. The bar turns red when the limit is reached. The two small round buttons next to it (+ and ?) open documentation articles: they are not purchase buttons. To change plan, go to your account settings, Subscription tab. --- ## R Ratio / Aspect Ratio Proportions between the width and height of an image or video. - 16:9 → landscape format (television, computer) - 9:16 → portrait format (mobile, stories) - 1:1 → square format (social media) - 4:3 → classic landscape format (older screens) Reference Image Existing image that you provide to an AI model to guide its generation. The model draws inspiration from it to respect a style, composition, or character appearance. Role → see Competency Prompt --- ## S Seamless Quality of an animation or sound loop whose transition between the end and the beginning is imperceptible. A seamless loop can run indefinitely without the player hearing or seeing the break. Seed Starting random number that determines the result of an AI generation. By using the same seed with the same prompt, exactly the same result is obtained. Useful for reproducing a generation or generating slight variations of the same result. Sovereign Model AI model developed by a European company (e.g., Mistral, based in Paris) or in open source (e.g., Meta's Llama, Black Forest Labs' Flux), independent of the large American platforms like OpenAI, Google, or Anthropic. Outside Celestory, some of these models can be installed on your own servers. 🔺 Inside Celestory, however, there is no feature to download a model and run it locally: every call leaves from Celestory's servers, whatever model you pick. Speech Synthesis (Text-to-Speech / TTS) Technology that converts written text into spoken audio. In Celestory, the Audio tab's ElevenLabs TTS models generate realistic voices for characters, narrated from text you write. Sprite In game development, a graphic image representing a character, object, or interface element, usually with a transparent background so it can be overlaid on a scene. --- ## T Temperature Parameter that controls the level of creativity (or randomness) of a textual AI model. A low value (close to 0) gives more predictable and factual answers. A high value (close to 1) gives more creative and surprising answers. 🔺 In the **AI text request** block the default is **1** — the most creative and therefore the least reproducible: lower it for any factual use. 🔺 Never set it to exactly **0**: the server reads a 0 as "nothing was set" and silently replaces it with 0.5. Use 0.1 instead. Text Variable (in Celestory) Text data stored in your Celestory project, reachable and editable from the editor. Examples: the player's name, a character's rules, a style guide, a dialogue script. Variables can be attached to AI prompts in the Text tab to enrich them dynamically; each one shows its token weight. Token Basic unit used by AI models to process text. One token corresponds approximately to ¾ of a word in English. Examples: "hello" = 1 token, "antidisestablishmentarianism" = 3-4 tokens, a 10-word sentence ≈ 13 tokens. Models are billed and limited in the number of tokens processed. Tool / Tool call An action an AI model can ask to run instead of merely answering in text: adding a block, reading a table, calling a service. In Celestory tools come from two sources — the 29 built-in tools that manipulate the graph, and whatever your MCP servers expose. Only models flagged as tool-capable can use them; with the others, a banner warns you that the chat will stay a plain conversation. TTS (Text-to-Speech) → see Speech Synthesis --- ## V Vibe Coding Approach to creating code assisted by AI: instead of writing code manually, you describe in natural language (English) what you want to create, and the AI generates the corresponding code. In Celestory, vibe coding mainly applies to the HTML5 block: you describe an interface or animation, and the AI produces the ready-to-use HTML/CSS/JS code. --- ## W WebP Modern image format developed by Google, lighter than PNG and JPG while maintaining good quality. Supports transparency like PNG. Ideal for web images and game assets where performance is important. ---