docs(10_Wiki): Topic_Business/General/Graphic/Programming을 Topics/ 하위로 이동
최상위 10_Wiki/Topic_*였던 4개 카테고리 폴더를 10_Wiki/Topics/Topic_* 로 재배치. 콘텐츠 변경 없음(순수 폴더 이동) — Topics/ 하위 나머지 폴더는 이미 지난 커밋에서 전부 정리된 상태(잔존 항목은 에이전트 운영 상태 및 사용자가 보존을 요청한 업데이트0615/무제 3.canvas 뿐).
This commit is contained in:
Antigravity Agent
@@ -0,0 +1,236 @@
|
||||
---
|
||||
id: wiki-2026-0508-generative-ai
|
||||
title: Generative AI
|
||||
category: 10_Wiki/Topics
|
||||
status: verified
|
||||
canonical_id: self
|
||||
aliases: [generative AI, gen-AI, generative model, LLM, image gen, video gen, audio gen, multimodal]
|
||||
duplicate_of: none
|
||||
source_trust_level: A
|
||||
confidence_score: 0.98
|
||||
verification_status: applied
|
||||
tags: [generative-ai, ai, llm, diffusion, multimodal, foundation-model]
|
||||
raw_sources: []
|
||||
last_reinforced: 2026-05-10
|
||||
github_commit: pending
|
||||
tech_stack:
|
||||
language: Python
|
||||
framework: Anthropic / OpenAI / Stability / HuggingFace
|
||||
---
|
||||
|
||||
# Generative AI
|
||||
|
||||
## 매 한 줄
|
||||
> **"매 새로운 content 의 의 의 의 model — 매 text, image, audio, video, code, 3D"**. 매 modern: Claude, GPT, Gemini, Llama (text), Midjourney/DALL-E/SD (image), Suno (audio), Sora/Veo (video). 매 transformer + diffusion 의 dominant.
|
||||
|
||||
## 매 핵심
|
||||
|
||||
### 매 modality
|
||||
- **Text**: GPT, Claude, Gemini, Llama.
|
||||
- **Image**: Stable Diffusion, Midjourney, DALL-E 3, FLUX, Imagen 3.
|
||||
- **Video**: Sora (OpenAI), Veo (Google), Runway, Pika.
|
||||
- **Audio / Music**: Suno, Udio, MusicLM.
|
||||
- **Speech**: ElevenLabs, OpenAI TTS, Whisper (STT).
|
||||
- **3D**: Meshy, Tripo, Luma Genie.
|
||||
- **Code**: Codex, CodeLlama, Claude.
|
||||
|
||||
### 매 architecture
|
||||
- **Transformer** (text, code).
|
||||
- **Diffusion** (image, video, audio).
|
||||
- **Latent diffusion** (SD).
|
||||
- **DiT** (Diffusion Transformer): SD3, Sora, FLUX.
|
||||
- **Mamba / SSM** (emerging).
|
||||
|
||||
### 매 modern (2025-2026)
|
||||
- **Frontier**: Claude Opus 4.7, GPT-5, Gemini 2 Ultra.
|
||||
- **Open**: Llama 3.x, Qwen 2.5, FLUX.
|
||||
- **Multimodal**: Sora, Veo 2, Genie 2.
|
||||
- **Reasoning**: o1, o3, R1.
|
||||
|
||||
### 매 응용
|
||||
1. **Productivity**: writing, coding.
|
||||
2. **Creative**: art, music, video.
|
||||
3. **Customer service**: chatbot.
|
||||
4. **Education**: tutor.
|
||||
5. **Marketing**: ad copy, image.
|
||||
6. **Research**: literature review.
|
||||
7. **Game**: NPC, content.
|
||||
|
||||
### 매 risk
|
||||
- **Hallucination**.
|
||||
- **Copyright** (training, output).
|
||||
- **Misinformation** (deepfake).
|
||||
- **Bias**.
|
||||
- **Energy use**.
|
||||
- **Job displacement**.
|
||||
|
||||
## 💻 패턴
|
||||
|
||||
### Text generation (Claude)
|
||||
```python
|
||||
from anthropic import Anthropic
|
||||
client = Anthropic()
|
||||
r = client.messages.create(model='claude-opus-4-7', max_tokens=1024,
|
||||
messages=[{'role': 'user', 'content': 'Write a haiku about AI'}])
|
||||
```
|
||||
|
||||
### Image (Stable Diffusion)
|
||||
```python
|
||||
from diffusers import StableDiffusionXLPipeline
|
||||
pipe = StableDiffusionXLPipeline.from_pretrained('stabilityai/sdxl-turbo', torch_dtype=torch.float16).to('cuda')
|
||||
img = pipe('a sunset over mountains', num_inference_steps=4).images[0]
|
||||
```
|
||||
|
||||
### FLUX (modern)
|
||||
```python
|
||||
from diffusers import FluxPipeline
|
||||
pipe = FluxPipeline.from_pretrained('black-forest-labs/FLUX.1-schnell', torch_dtype=torch.bfloat16).to('cuda')
|
||||
img = pipe('photorealistic forest', num_inference_steps=4).images[0]
|
||||
```
|
||||
|
||||
### Video (Sora-like)
|
||||
```python
|
||||
# 매 OpenAI Sora API (when available)
|
||||
client.videos.generate(model='sora-1', prompt='a cat playing piano', duration_s=10)
|
||||
```
|
||||
|
||||
### Audio (Suno-like)
|
||||
```python
|
||||
# 매 commercial APIs
|
||||
suno_client.generate_song(prompt='upbeat synth-pop', duration_s=180)
|
||||
```
|
||||
|
||||
### TTS (ElevenLabs)
|
||||
```python
|
||||
import elevenlabs
|
||||
audio = elevenlabs.generate(text='Hello world', voice='Adam', model='eleven_multilingual_v2')
|
||||
```
|
||||
|
||||
### 3D (Tripo / Meshy)
|
||||
```python
|
||||
# 매 image → 3D model
|
||||
mesh = tripo_client.image_to_mesh('input.png')
|
||||
mesh.save('output.glb')
|
||||
```
|
||||
|
||||
### Multimodal (Claude vision)
|
||||
```python
|
||||
client.messages.create(model='claude-opus-4-7', max_tokens=1024,
|
||||
messages=[{'role': 'user', 'content': [
|
||||
{'type': 'image', 'source': {'type': 'base64', 'media_type': 'image/jpeg', 'data': img_b64}},
|
||||
{'type': 'text', 'text': 'Describe this image'},
|
||||
]}])
|
||||
```
|
||||
|
||||
### Agent (multi-step)
|
||||
```python
|
||||
def agent_loop(goal, tools, max_steps=10):
|
||||
history = [{'role': 'user', 'content': goal}]
|
||||
for _ in range(max_steps):
|
||||
r = client.messages.create(model='claude-opus-4-7', tools=tools, messages=history)
|
||||
if r.stop_reason == 'end_turn': return r
|
||||
# 매 execute tool, append result
|
||||
```
|
||||
|
||||
### Watermark (C2PA)
|
||||
```python
|
||||
from c2pa import Signer
|
||||
Signer(cert).sign('output.png', claims={
|
||||
'generator': 'AI', 'model': 'flux-1-schnell',
|
||||
})
|
||||
```
|
||||
|
||||
### Prompt engineering
|
||||
```python
|
||||
def well_formed_prompt(task, context, examples=[], format='json'):
|
||||
return f"""## Context
|
||||
{context}
|
||||
|
||||
## Examples
|
||||
{format_examples(examples)}
|
||||
|
||||
## Task
|
||||
{task}
|
||||
|
||||
## Output format
|
||||
{format}"""
|
||||
```
|
||||
|
||||
### RAG-augmented gen
|
||||
```python
|
||||
def rag_generate(question, retriever, llm):
|
||||
docs = retriever.retrieve(question, k=5)
|
||||
context = '\n'.join(d.text for d in docs)
|
||||
return llm.generate(f"Context:\n{context}\n\nQuestion: {question}\nAnswer with citations:")
|
||||
```
|
||||
|
||||
### Fine-tune (LoRA)
|
||||
```python
|
||||
from peft import LoraConfig, get_peft_model
|
||||
config = LoraConfig(r=16, lora_alpha=32, target_modules=['q_proj', 'v_proj'])
|
||||
model = get_peft_model(base_model, config)
|
||||
# 매 train on task data
|
||||
```
|
||||
|
||||
### Generation cost monitoring
|
||||
```python
|
||||
def cost_track(usage):
|
||||
pricing = {'claude-opus-4-7': {'in': 15/1e6, 'out': 75/1e6}}
|
||||
cost = usage.input_tokens * pricing[model]['in'] + usage.output_tokens * pricing[model]['out']
|
||||
return cost
|
||||
```
|
||||
|
||||
### Eval (LLM-as-judge)
|
||||
```python
|
||||
def llm_judge(output, criteria):
|
||||
prompt = f'Rate {criteria}. Response: {output}. Output JSON: score 0-10.'
|
||||
return json.loads(judge.generate(prompt))['score']
|
||||
```
|
||||
|
||||
### Brand safety
|
||||
```python
|
||||
def brand_safe(output):
|
||||
return classify_toxicity(output) < 0.05 and not has_competitor(output) and has_brand_voice(output)
|
||||
```
|
||||
|
||||
## 매 결정 기준
|
||||
| 상황 | Tool |
|
||||
|---|---|
|
||||
| Best text quality | Claude Opus 4.7 / GPT-5 |
|
||||
| Cost-aware text | Claude Sonnet / GPT-4o-mini |
|
||||
| Best image | FLUX / Midjourney v7 |
|
||||
| Fast image | SDXL Turbo / FLUX schnell |
|
||||
| Video | Sora / Veo 2 |
|
||||
| Audio | Suno / ElevenLabs |
|
||||
| Local | Llama 3.x + SDXL local |
|
||||
| Code | Claude / Codex |
|
||||
|
||||
**기본값**: 매 frontier API + 매 RAG + 매 prompt eng + 매 LLM-judge eval + 매 brand safety + 매 cost track.
|
||||
|
||||
## 🔗 Graph
|
||||
- 부모: [[AI]] · [[Foundation-Models]]
|
||||
- 변형: [[Transformer_Architecture_and_LLM_Foundations|LLM]] · [[Diffusion-Models]] · [[Multimodal-LLM]]
|
||||
- 응용: [[Generative-Adversarial-Networks]] · [[Stable-Diffusion]]
|
||||
- Adjacent: [[RAG]] · [[Fine-tuning]] · [[Prompt_Engineering|Prompt-Engineering]] · [[Ethics & AI]]
|
||||
|
||||
## 🤖 LLM 활용
|
||||
**언제**: 매 모든 productivity, creative, customer-facing.
|
||||
**언제 X**: 매 deterministic compute. 매 IP-strict (with care).
|
||||
|
||||
## ❌ 안티패턴
|
||||
- **Hallucination 의 ship**: 매 verify.
|
||||
- **No watermark**: 매 misinformation.
|
||||
- **No copyright check**: 매 legal risk.
|
||||
- **Single model lock-in**: 매 API down → outage.
|
||||
- **No cost monitoring**: 매 bill shock.
|
||||
|
||||
## 🧪 검증 / 중복
|
||||
- Verified (Anthropic, OpenAI, Stability, Google docs).
|
||||
- 신뢰도 A.
|
||||
|
||||
## 🕓 Changelog
|
||||
| 날짜 | 변경 |
|
||||
|---|---|
|
||||
| 2026-04-20 | Auto |
|
||||
| 2026-05-08 | Phase 1 |
|
||||
| 2026-05-10 | Manual cleanup — modalities + 매 text / image / video / audio / 3D / agent code |
|
||||
Reference in New Issue
Block a user