# 🎨 Seraphina Artist Engine v3.5 (Z-Image Turbo Edition)
100% FREE & FULLY OFFLINE INTERACTIVE MULTILANGUAGE CHAT + PHOTO-RENDER
No internet connection required! Perfect for country houses, summer camps, or places without network access. Powered by lightweight Z-Image Turbo (Wan 2.1) by 6tZ.
---
### 🌍 MULTILANGUAGE CORE (5 Languages / 5 国语言)
The UI, chat logic, character name, and hidden prompt-generation states automatically adapt to 5 languages: EN, ZH, JA, ES, RU.
* [EN] Seraphina is a vivid 21-year-old amateur artist girl who snaps casual sketches on her phone and talks to you like a real friend. Completely optimized for 8 GB VRAM.
* [ZH] 采用轻量化架构。玩家将与21岁的傲娇少女画家莎拉菲娜一起互动，她会用手机随手记录美丽的乡村风景。针对 8 GB VRAM 进行了极限优化。
* [JA] 21歳の少し生意気なアマチュア画家の女の子「セラフィナ」と会話しながら風景を生成。8 GB VRAM 向けに完全最適化され、遅延ゼロを実現。
* [ES] Seraphina es una chica de 21 años, artista aficionada, que toma bocetos del campo con su teléfono y habla contigo como una amiga real.
* [RU] Серафима — живая 21-летняя художница-любительница, которая делает дачные наброски на телефон и общается с вами как реальный сверстник. Ноль фризов.
---
### 🔥 8 GB VRAM OPTIMIZATION (Zero Lag):
* Smart Memory Bridge: The moment you click "Render", Ollama instantly flushes out of VRAM, giving 100% of GPU power to ComfyUI.
* Monolithic CLIP: Ultra-lightweight CLIP GGUF Q3 (1.8 GB) stays cached in VRAM, eliminating Python's garbage collection lags.
* Auto-Reset: Session history clears out cleanly on every launch, protecting KSampler from old context overflow.
* Clean Scenery: Fixed KSampler nodes. No photography slang, ISO stats, or weird artifacts in the picture.
---
### 🤝 ACKNOWLEDGMENTS & CREDITS / БЛАГОДАРНОСТИ
* Special Thanks to 6tZ: Huge credit to 6tZ for developing the incredible, lightweight, and high-performance Z-Image Turbo checkpoint line. Thank you for pushing the boundaries of local AI rendering and making this release possible!
* ComfyUI & Ollama Communities: Thanks to all open-source developers providing the GGUF ecosystem and fast pipeline nodes that made this fully offline dacha experience possible.
Description
### 🪐 VERSION 3.5: Standalone Studio Edition
This archive contains the complete, fully responsive standalone engine optimized for stable offline rendering, image analysis, and immediate character compliance.
---
### 🌐 STEP 1: Download Original Checkpoint by 6tZ
Download the preferred format of the original **Z-Image Turbo** model directly from the author's page:
👉 **[Download original Checkpoint on CivitAI](https://civitai.com)** (Model Author: *6tZ**)*
* **Preset A — int8 version (5.74 GB) [RECOMMENDED]**: For **RTX 2000, 3000**, and entry-level **RTX 4050 / 5050** cards. True 8-bit quality without VRAM pressure.
* **Preset B — nvfp4 version (3.24 GB)**: For **RTX 4000 and RTX 5000** series only. Blazing fast 4-bit render leveraging native tensor cores.
⚠️ **CRITICAL RULE / ВАЖНОЕ ПРАВИЛО:**
You **MUST rename the downloaded file to zImageTurboNSFW_110.safetensors** and place it in the ComfyUI/models/checkpoints/ directory.
---
### 📦 STEP 2: Quick Start & Installation
1. Extract the contents of your downloaded studio ZIP archive.
2. **CRITICAL FOR OLLAMA:** Before launching anything, look at your Windows system tray (near the clock). Right-click the Ollama icon and select **"Quit Ollama"**. The installer must run with Ollama completely closed!
3. Put the renamed checkpoint zImageTurboNSFW_110.safetensors into your checkpoints folder.
4. Put the text encoder file Q3_K_M.gguf into ComfyUI/models/clip/.
5. **CRITICAL (Do BEFORE leaving the internet area):** Run *1_Setup_Studio.bat** at least once with the internet connected to pull the brain core (baytout3/qwen3.5-uncensored:9b) and lock all 33 character layers directly into the GPU.
6. Unplug your network cable, launch *2_Run_Studio.bat** and enjoy the 100% offline vision experience anywhere (Dacha, Camp, Offline areas)!
---
💡 Our gratitude goes to *6tZ** for the amazing base checkpoint that powers this standalone installation tool.*