Landscape mapping: how AI game generators compare on editor experience

[wkl-nieta]

Background

I surveyed several AI game generation tools to understand how different teams approach the "editor experience" problem. The core tension: should the user interact with generated output via code, a visual editor, or not at all?

Comparison: Editor Approach

Tool	Generation	Editor	Target User
OpenGame	GameCoder-27B → Phaser code	Code-level (IDE)	Developers
WorldX	4-model LLM pipeline	Limited GUI + SQLite	Tech-savvy creators
Claude Code Game Studios	49-agent framework	Claude Code terminal	Developers
neta.art	Multi-model pipeline	Full visual editor	Non-dev creators
SpawnForge	AI-native 2D/3D engine	Visual editor + MCP	Both

neta.art: https://neta.art/explore?utm_source=github&utm_medium=issue

Key Insight

Most open-source tools optimize for developer experience (code output, API access). But the fastest-growing segment seems to be non-dev creators who want prompt → shareable game with zero code.

OpenGame's approach (code generation) is perfect for devs who want to iterate on the generated code. But I wonder if there's a middle ground — generate the code, but also provide a "share" URL so non-dev collaborators can playtest without setting up a dev environment.

Question

Is OpenGame considering a hosted playtest/sharing feature, or is the vision strictly local-first code generation? The market data from the tools above suggests both modes have strong demand.

Full comparison

Feature	OpenGame	WorldX	neta.art
Input	Single prompt	Single prompt	Single prompt
Output	Phaser web game	Phaser 3 browser world	Browser game
World persistence	Per-session	SQLite	Cloud-hosted
Multi-session	❌	✅	✅
Asset gen	GameCoder-27B	4-model LLM	Multi-model
Editor	Code (IDE)	Limited GUI	Full visual
Sharing	Manual deploy	Manual deploy	One-click URL