This article is the current development log of PKC AI-ONE.
This article was analyzed using AI.
PKC AI-ONE — Key Feature Summary
Author: GPT
1. Overview
This document summarizes the core features of the PKC AI-ONE system running on an RTX 2060 Super (8GB). It explains the essential functions in a simple and easy-to-understand way, without complex technical terms.
2. Main Feature Summary
PKC AI-ONE is a fully local, integrated AI system that supports:
- Text interaction (LLM)
- Emotion analysis
- Image generation
- Vision-based image understanding
- TTS (Text-to-Speech)
- STT (Speech-to-Text)
✔ 1) Text Chat (LLM)
- Uses Llama-3.2-8B (GGUF model)
- Smooth real-time conversation via SSE streaming
- Combined pipeline of emotion analysis + language model
- Automatically adjusts response tone based on user emotion and writing style
✔ 2) Image Generation (Stable Diffusion)
- Based on Stable Diffusion 3.5 medium GGUF
- Generates 512×768 images
- Shows generation progress
- Korean prompts are automatically translated
- Cached prompts regenerate instantly
✔ 3) Vision AI (Image Understanding)
- Qwen2-VL model for image content analysis
- Model automatically loads when an image query is requested
✔ 4) File Upload → Analysis
- Automatically summarizes or analyzes image/text files
- Shows thumbnail previews
✔ 5) Emotion Analysis
- korean-emotion-kluebert-v2
- Detects emotions from user messages (e.g., joy, sadness, anger, neutral)
- Adjusts AI response tone accordingly
✔ 6) Session Management
- Saves conversation history
- Keeps separate logs per session
- Supports creating, deleting, renaming sessions
- Full JSON export/import supported
✔ 7) Browser UI Features
- STT (Speech-to-Text)
- TTS (Text-to-Speech)
- Image generation button
- Web search button
- Auto cleanup of old chat bubbles
- Fully mobile responsive
✔ 8) System Monitoring
- Real-time GPU / CPU / RAM usage display
- Shows model loading status
3. How the System Works (Simplified)
● 1) Loads only the required model
- Keeps the LLM active during text conversations
- Temporarily unloads the LLM during image generation to free VRAM
- Reloads it after work is completed
● 2) Image models load only when needed
- Prevents unnecessary VRAM usage
- Cache enables fast reuse after generation
● 3) Automatic conversation memory
- Stores user/AI conversation history in a local DB
- Helps maintain context across sessions
- AI remembers previous conversations stored in the DB
4. Conclusion
PKC AI-ONE provides the following features in a single system:
- Emotion analysis (korean-emotion-kluebert-v2)
- Text conversation (llama-3-Korean-Bllossom-8B-Q5_K_M.gguf)
- Image generation (sd3.5_medium-Q5_1.gguf)
- Image understanding (Qwen2-VL-2B-Instruct-Q4_K_M.gguf)
- File analysis (System)
- Session & log management (System)
- Web search (System)
- STT & TTS (Browser Feature)
In short, it is an all-in-one local AI tool running entirely on a personal PC.
'AI-ONE > English Translation' 카테고리의 다른 글
| Building a Local Multimodal LLM System with 8GB VRAM _01 (0) | 2026.01.04 |
|---|