Editor AI

AI-powered level generation for Geometry Dash. Describe your level in plain text and watch AI build it in the editor.

Consider supporting the project by starring it on GitHub

Quick Start

1. Download the latest release
2. Open the game and navigate to the Geode page
3. Press the manual download button and select the file
4. Open the editor and press the AI button
5. Click the gear icon to configure your AI provider

AI Setup

All settings are configured in-game via the gear icon in the AI popup. No need to dig through Geode mod settings.

EditorAI Platinum (Free — recommended)

EditorAI Platinum is a community proxy that lets anyone run the official EditorAI Ollama models over the internet for free.

1. Open settings → set Provider to ollama
2. Switch to the Provider tab and enable Platinum
3. Available models are auto-detected from the Platinum network

Want to donate your computer to the network? Follow the instructions in the Platinum repository.

Ollama (Free — recommended for 6+ GB VRAM)

Uses ~5.2 GB storage. Fully local and private — no internet required after setup.

1. Download and install Ollama
2. Pull the official model: ollama pull entity12208/editorai:qwen
   • deepseek — more creative  |  qwen — more powerful  (each ~5.2 GB)
3. Open settings → set Provider to ollama
4. Your installed models are auto-detected on the Provider tab

Claude (Paid)

1. Get an API key at console.anthropic.com
2. Open settings → set Provider to claude → switch to Provider tab
3. Select a model and paste your API key

Mistral AI (Paid)

1. Get an API key at console.mistral.ai
2. Open settings → set Provider to ministral → switch to Provider tab
3. Select a model and paste your API key

HuggingFace (Free tier available)

1. Get a token at huggingface.co/settings/tokens
2. Open settings → set Provider to huggingface → switch to Provider tab
3. Enter a model ID and paste your token

ChatGPT (Paid)

1. Get an API key at platform.openai.com/api-keys
2. Open settings → set Provider to openai → switch to Provider tab
3. Select a model and paste your API key

Gemini (Not recommended — tight rate limits)

1. Get an API key at aistudio.google.com/api-keys
2. Open settings → set Provider to gemini → switch to Provider tab
3. Select a model and paste your API key

Usage

1. Open the editor of any level
2. Press the AI button in the editor toolbar
3. Type a description of what you want (e.g. "medium difficulty platforming level")
4. Click Generate and wait — a timer shows elapsed time
5. AI-generated objects appear as blue ghost previews
6. Choose Accept, Edit, or Deny:
   • Accept — makes objects permanent
   • Edit — makes objects solid so you can move/delete them, then press Done
   • Deny — removes all generated objects
7. Optionally rate the generation 1-10 to help the AI learn your style

Use the up/down arrows next to the text input to recall previous prompts. Press Cancel during generation to abort.

Settings

Click the gear icon in the top-left of the AI popup to access all settings, organized into three tabs:

Tab	Settings
General	Provider, difficulty, style, length, max objects, spawn speed
Provider	Model selection, API keys, Ollama/Platinum config, timeout
Advanced	Triggers & colors, rate limiting, feedback & learning

Advanced Features

Enable Triggers/Colors in the Advanced settings tab to unlock:

• Color Triggers — smooth color transitions for background, ground, and objects
• Move Triggers — moving platforms and obstacles with easing curves
• Rotate Triggers — spinning decorations and obstacles around center points
• Alpha Triggers — fade objects in and out
• Toggle Triggers — show/hide groups of objects
• Pulse Triggers — flash colors on groups for visual emphasis
• Spawn & Stop Triggers — chain trigger sequences and cancel animations
• Speed Portals — control game speed (0.5x to 4x)
• Player Visibility — show/hide the player icon and trail
• Color Channels — assign objects to color channels for coordinated color changes
• Group IDs — assign objects to groups for trigger targeting
• Multi-Activate — make orbs, pads, and triggers fire on every touch

These features work best with smarter models (Claude, Gemini, Ollama qwen, or Platinum).

Blueprint Preview

Generated objects appear as blue-tinted ghost previews (40% opacity, non-interactable) before you commit them. This lets you see what the AI made before it affects your level. The Edit button makes objects solid so you can rearrange or delete individual pieces before pressing Done.

Feedback & Learning

When enabled (on by default), a rating popup appears after you accept/edit/deny a generation. Your ratings and optional text feedback are stored locally and injected into future AI prompts as examples — the AI learns what you like and avoids what you don't. Feedback examples are matched by similarity to your current request (difficulty, style, length).

Error Codes

If something goes wrong, the error message includes a code like [EAI-G2001]. Share this code when reporting issues — it encodes the provider, error category, and specific problem, making diagnosis much faster.

Fine-Tune Your Own Model

The training/ folder contains scripts to fine-tune a Qwen Instruct model on GD level data using QLoRA, then export it to Ollama. See training/README.md for details.

Support

Join the Editor AI Discord server for support and ideas.