Generate text, images and content with AI in Celestory.
Describes the AI Generation window (Cmd+Shift+G), its 6 tabs (Text, Video, Image, Audio, Generations, Tools), the Monthly usage gauge and the monthly AI credits per plan: Free trial, Personal (Pro), Organization (Business).
Explains how AI reaches players through the AI text request and AI image request blocks, which call Celestory servers (no API key in the project), bill the creator's credits and are restricted to the Organization (Business) plan.
Explains the Text tab chat, which can modify your project: with a tool-capable model, the assistant adds and wires blocks, writes dialogue and creates variables through 29 graph tools, in at most 8 rounds per answer.
Explains the Tools tab, which connects MCP servers the assistant can use during a Text tab conversation. Add a connector with a title and server address, allow each tool; authentication is manual (none, API token or JSON headers).
Glossary of AI terms used in Celestory, in simple words: AI Agent of the Graph, AI Credit (© badge, quota per plan, cost per action), and other terms, listed alphabetically for beginners.
Explains the Image tab: Text to image and Image to image modes, the model picker, the form built from the chosen model, and asynchronous generation whose results are retrieved in the Generations tab, each one spending credits.
Explains the Video tab and its 6 generation modes, such as Image to video, Text to video and Start and end frame, with the model deciding the fields. Only one generation can run at a time in this tab.
Explains the Audio tab, mostly ElevenLabs models routed through fal: Text to Speech, Music, Sound Effects and dialogue. There is no mode dropdown: the chosen model determines the fields shown.
Explains the Previous Generations tab, where all Image, Video and Audio results arrive: status pills, the Update, Add to files, Copy and Show details buttons, the Editor and In-Game filters, and Load More for older entries.
Explains how to constrain the AI with competency prompts (style guides, methodologies attached as text variables), with the vibe coding use case for the HTML5 block, and gives advice on monitoring AI use.
This article will be available soon.
Describes how the AI text request block behaves: inputs prompt, system prompt, model (ministral-3b or gemini-2.5-flash), temperature (1 by default) and result, and the traps, notably no On Error output and no answer length setting.
Shows how to plug an MCP connector into the AI assistant so it gains extra tools (reading Notion pages, querying a business service, triggering an action). It is done in the Tools tab of the AI Generation window (Cmd+Shift+G) with Add a connector.
Shows how the AI assistant, in the Text tab of the AI Generation window (Cmd+Shift+G), builds your graph while it answers: blocks appear and wire themselves. The chosen model must be able to use tools, otherwise a banner warns you.
soon
