Damian

Real2Playable is an open-source (MIT License) AI-assisted 2D Game Asset Compiler designed for indie game developers, solo creators, and game-jam participants. It compiles everyday photographs of real-world objects (such as desk toys, potted plants, mugs, or tools) into standardized, transparent, measured, and animation-ready 2D game asset packages that can be dropped directly into any 2D game engine (Phaser 3, Godot 4, Unity, etc.). Unlike conventional AI image generators or chat interfaces, Real2Playable functions as a technical asset compiler and verification pipeline: - Web Studio (React + TypeScript): A desktop-first workbench for image uploads, interactive object masking, origin/pivot placement, procedural motion presets configuration, and alpha-channel quality inspection. - Orchestration Backend (FastAPI + Python): Bridges multi-stage Meta models via the official Responses API: - SAM 3.1 (Segment Anything Model): Handles high-precision foreground object segmentation from raw photos and background re-masking on generated artwork. - Muse Image 1.0: Stylizes the segmented cutout into cohesive game art styles (Storybook Toy, Retro 8-bit Pixel Art, or 3D Cartoon Clay). - Muse Spark 1.3: Produces descriptive, schema-compliant semantic metadata (asset names, asset types, and descriptive tags). - Deterministic Compiler: Normalizes graphics to an RGBA 512×512 canvas, calculates pixel-accurate bounding boxes, generates procedural single-sprite transform motions (idle_bob, interact_pulse, impact_flash), bakes an optional 4-frame sprite atlas, generates SHA-256 checksum manifests, and packages everything into an open .asset.zip. - Independent Consumer Playground (Phaser 3): An entirely decoupled reference game app that imports exported ZIP packages purely in the client browser, validates them across 4 integrity checkpoints, and renders animated sprites with zero server API calls. Problems Solved: 1. The "Raw AI Image ≠ Game Asset" Gap 2. Missing Origin / Pivot Coordinates & Uncalibrated Geometry 3. Generative Multi-Frame Hallucinations vs. Predictable Motion 4. File Fragmentation and Proprietary Engine Lock-In 5. Multi-Stage Pipeline Fragility & Wasted Inference Costs 6. The "Black-Box Demo" Problem (Lack of Portability Proof)