# 👑 MORGANA SANDBOX ENGINE v4.0 [The Monolithic Offline AI Playground]
The era of predefined, rigid setups is officially over. Version 4.0 transforms the framework into a fully dynamic, image-agnostic AI Sandbox core. Named in honor of the dark, gothic elegance of Morgana, this engine breathes life into ANY companion entirely by extracting prompt tokens, workflows, and character data directly from your active root PNG file!
> "I am absolutely shocked by what the AI can achieve here — the way the engine dynamically adapts and generates everything offline is just incredibly cool!"
---
### 💻 Technical Architecture & Release Concept
During a deep architectural refactor, the entire codebase has been completely overhauled. The engine has been stripped of all internet-dependent libraries, network clutter, and file duplication, unifying the complete backend application layer into a single, high-performance offline monolithic core dima_v2.py).
The system is fully autonomous and operates 100% locally. A smart auto-date-modified parser automatically scans the root folder upon initialization, extracting the character's Identity and ComfyUI node network on the fly. This gives the user complete freedom to morph the sandbox world simply by swapping a single PNG asset in the root directory, completely eliminating manual backend file modification.
In live auto-chat mode, a non-destructive 512x512 latent shield is enforced. It dynamically intercepts empty latent node calls and optimizes them, preventing melted human anatomy, body duplication, and distorted faces while leaving native step counts, CFG scales, and custom planners of advanced architectures (Flux / Create) fully intact.
Description
👑 VERSION 3.5 [THE UNIVERSAL AGNOSTIC UPDATE]
The era of heavy, restrictive 13 GB monolith archives is officially over. Version 3.5 introduces a complete core architecture refactoring, turning Seraphina into a lightweight, fully adaptable sandbox shell that integrates seamlessly with your personal ComfyUI setups and handles unconstrained thematic generations.
What's New in v3.5:
🚀 80% Less Disk Space: The core distribution size has been reduced from 13 GB down to just ~2.5 GB! We completely wiped duplicate diffusion transformers, clips, and vae tracks from the repository.
🔞 Built-In Uncensored Core Matrix: Integrated the custom local GGUF-layer Gemma-3 Heretic running via Ollama. Complete narrative freedom, zero structural filtering, and native support for complex adult (NSFW) story-driven dialogue mechanics.
🎨 Choose Your Own Models: No more hardcoded generation scripts. Run ANY model setup you want (Flux, SDXL, Pony, SD 1.5) inside your personal ComfyUI, generate a test layout frame, and copy that PNG file into the game root folder. The engine will instantly hijack its API nodes and stream live gameplay text directly into your custom workspace environment.
💬 Streamlined Conversational Flow: Seraphina now generates perfectly balanced, concise responses consisting strictly of 2-4 sentences to heavily boost text-reading narrative dynamics.
🌐 Bulletproof Language Lock: Enhanced system context interceptors to ensure flawless native multilingual switching (RU, EN, ZH, JA, ES) matching the player's incoming token sequence without prompt pollution.
🛠️ Two-Stage Smart Batches: Split runtime scripts into '1_Setup_Studio.bat' for clean one-time local brain compilation and '2_Run_Studio.bat' for instant production server deployment with integrated hardware status checks.
How to Launch v3.5:
1. Open ComfyUI, generate a test image using your absolute favorite custom model pipeline, and copy that PNG file into the game root directory.
2. Run '1_Setup_Studio.bat' to parse your workflow and compile the uncensored local brain inside Ollama.
3. Run '2_Run_Studio.bat' to start the live game stream, open the web viewport, and unleash your creative freedom!