refactor(topics): 멀티 에이전트용 지식 재편 — _Common(공통 기본기) + Domain_* 구조
에이전트 8종(대화형/프로그래머 C·S/디자이너/설계자/기획자/QA/PD/PM)에게 [공통 기본 능력 + 롤별 Specialty] 2층으로 지식을 주입하기 위한 재분류. 문서 내용·포맷은 무수정, 폴더 이동만 (6,372개 문서 수 보존 확인). - Topic_Programming → Domain_Programming (내부 구조 보존) - Topic_Graphic → Domain_Design - Topic_Business → Domain_Product - Topic_General → Domain_General - _Common 신설: Math(구 Topic_Math_Specialty), Reasoning(구 General/From_Thinking & Reasoning), Reasoning_Creativity(구 General/From_창의성), Communication(Poetic_Blog_Writing + From_writing) - 타 도메인의 From_* 폴더는 유지 (출처 표기일 뿐, 이미 도메인에 맞게 분류된 문서) - 빈 폴더 정리 (memory/procedures) - 에이전트→폴더 매핑은 workspace의 .astra/agent-knowledge-map.json (9개 에이전트) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Antigravity Agent
+274
@@ -0,0 +1,274 @@
|
||||
---
|
||||
id: wiki-2026-0508-2026년-인공지능-시각-언어-생성-패러다임-전환-및-연속
|
||||
title: 2026 AI Visual Language Generation Paradigm Shift
|
||||
category: 10_Wiki/Topics
|
||||
status: verified
|
||||
canonical_id: self
|
||||
aliases: [continuous creative workflow, visual AI 2026, draft mode paradigm, prompt engineering visual]
|
||||
duplicate_of: none
|
||||
source_trust_level: B
|
||||
confidence_score: 0.85
|
||||
verification_status: conceptual
|
||||
tags: [image-generation, midjourney-v7, draft-mode, prompt-engineering, continuous-workflow, visual-ai]
|
||||
raw_sources: []
|
||||
last_reinforced: 2026-05-09
|
||||
github_commit: pending
|
||||
---
|
||||
|
||||
# 2026 AI Visual Language Generation Paradigm Shift
|
||||
|
||||
## 📌 한 줄 통찰
|
||||
> **Single shot → continuous workflow**. 매 draft mode 의 fast iteration + omni reference 의 consistency + post-edit 의 polish. 매 prompt 의 camera / lighting science 의 vocabulary.
|
||||
|
||||
## 📖 핵심 paradigm shift
|
||||
|
||||
### 매 evolution
|
||||
|
||||
#### 2022-2023 (Era 1): Single shot
|
||||
- 매 prompt → image.
|
||||
- 매 luck.
|
||||
- 매 generic output.
|
||||
|
||||
#### 2023-2024 (Era 2): Iterative
|
||||
- 매 multiple variation.
|
||||
- 매 prompt iterate.
|
||||
- 매 inpaint.
|
||||
|
||||
#### 2025-2026 (Era 3): Continuous workflow
|
||||
- 매 draft mode (cheap explore).
|
||||
- 매 reference (style, character, omni).
|
||||
- 매 post-edit pipeline.
|
||||
- 매 production-quality output.
|
||||
|
||||
### 매 5-layer prompt structure
|
||||
|
||||
#### 1. Subject
|
||||
- 매 specific entity (person, object, scene).
|
||||
- 매 physical detail.
|
||||
- 매 emotional / narrative context.
|
||||
|
||||
#### 2. Medium
|
||||
- "Oil painting, watercolor, digital art, photo".
|
||||
- 매 era / school ("Renaissance, Bauhaus, Cyberpunk").
|
||||
|
||||
#### 3. Environment / Composition
|
||||
- 매 location.
|
||||
- 매 framing ("close-up, wide shot, low angle").
|
||||
- 매 background.
|
||||
|
||||
#### 4. Lighting
|
||||
- 매 type ("Golden hour, volumetric, chiaroscuro, rim light").
|
||||
- 매 source ("softbox, natural, neon").
|
||||
|
||||
#### 5. Technical parameter
|
||||
- 매 lens ("85mm, 24mm, macro").
|
||||
- 매 depth ("shallow, deep").
|
||||
- 매 ratio ("--ar 16:9").
|
||||
- 매 quality ("--q 2, 8k").
|
||||
|
||||
### 매 photography vocabulary
|
||||
- **Lens**: 매 85mm portrait, 24mm wide, 100mm macro.
|
||||
- **Aperture**: f/1.4 (shallow DOF), f/8 (sharp).
|
||||
- **Lighting type**: golden hour, blue hour, soft light, hard light.
|
||||
- **Composition**: rule of thirds, leading lines, symmetry.
|
||||
- **Color theory**: complementary, analogous, monochrome.
|
||||
|
||||
### Continuous workflow
|
||||
|
||||
#### Step 1: Mood board
|
||||
- 매 reference (Pinterest, ArtStation).
|
||||
- 매 style direction.
|
||||
|
||||
#### Step 2: Draft generation
|
||||
- 매 30+ variant.
|
||||
- Midjourney `--draft` (10x speed).
|
||||
- Flux Schnell (4 step).
|
||||
|
||||
#### Step 3: Selection
|
||||
- 매 promising 5-10.
|
||||
- 매 visual review.
|
||||
|
||||
#### Step 4: Refinement
|
||||
- 매 prompt iterate.
|
||||
- 매 reference (sref / cref / oref).
|
||||
|
||||
#### Step 5: Full quality
|
||||
- 매 selected 의 high-quality.
|
||||
|
||||
#### Step 6: Post-edit
|
||||
- 매 inpaint defects.
|
||||
- 매 outpaint extend.
|
||||
- 매 face restoration.
|
||||
|
||||
#### Step 7: Upscale
|
||||
- Real-ESRGAN.
|
||||
- Magnific.
|
||||
- Topaz.
|
||||
|
||||
#### Step 8: Final touch (optional)
|
||||
- Photoshop.
|
||||
- Lightroom (color grade).
|
||||
|
||||
### 매 reference 의 type
|
||||
|
||||
#### Style reference (sref)
|
||||
- 매 brand 의 mood.
|
||||
- 매 visual coherence.
|
||||
|
||||
#### Character reference (cref)
|
||||
- 매 person consistency.
|
||||
- 매 series / campaign.
|
||||
|
||||
#### Omni reference (oref) — Midjourney V7
|
||||
- 매 specific object identity.
|
||||
- 매 product mockup.
|
||||
|
||||
#### IP-Adapter (Stable Diffusion)
|
||||
- 매 reference image 의 style + structure.
|
||||
|
||||
### 매 model 의 specific control
|
||||
|
||||
#### Midjourney V7
|
||||
- `--draft`, `--sref`, `--cref`, `--oref`.
|
||||
- `--s` (stylize), `--c` (chaos), `--w` (weird).
|
||||
- 매 minimal natural language.
|
||||
|
||||
#### DALL-E 3
|
||||
- 매 natural language.
|
||||
- 매 GPT-4 의 expansion.
|
||||
- 매 negation 약.
|
||||
|
||||
#### Stable Diffusion / Flux
|
||||
- 매 weighted prompt: `(keyword:1.2)`.
|
||||
- 매 negative prompt 강.
|
||||
- 매 LoRA, ControlNet, IP-Adapter.
|
||||
|
||||
### 매 emerging (2026)
|
||||
|
||||
#### Video generation
|
||||
- Sora (OpenAI).
|
||||
- Veo 2 (Google).
|
||||
- Runway Gen-3.
|
||||
- Kling.
|
||||
- 매 image → video.
|
||||
- 매 1 minute clip.
|
||||
|
||||
#### 3D generation
|
||||
- 매 image / text → 3D mesh.
|
||||
- 매 game asset.
|
||||
- TripoSR, InstantMesh.
|
||||
|
||||
#### Real-time generation
|
||||
- LCM (Latent Consistency Model).
|
||||
- SDXL Turbo.
|
||||
- 매 < 1 sec / image.
|
||||
|
||||
## 💻 Code
|
||||
|
||||
### Iterative workflow (production)
|
||||
```python
|
||||
class CreativeWorkflow:
|
||||
def __init__(self, model="midjourney"):
|
||||
self.model = model
|
||||
|
||||
def explore(self, base_prompt: str, n_drafts=30):
|
||||
"""Stage 1: Draft."""
|
||||
variations = self.generate_variations(base_prompt)
|
||||
return self.batch_generate(variations, draft=True)
|
||||
|
||||
def select(self, drafts, criteria="visual_quality"):
|
||||
"""Stage 2: Select."""
|
||||
scored = [(d, self.score(d, criteria)) for d in drafts]
|
||||
return sorted(scored, key=lambda x: -x[1])[:5]
|
||||
|
||||
def refine(self, selected_image, refinement_prompt):
|
||||
"""Stage 3: Refine."""
|
||||
return self.generate(refinement_prompt, reference=selected_image)
|
||||
|
||||
def post_edit(self, image):
|
||||
"""Stage 4: Post-edit."""
|
||||
defects = self.detect_defects(image)
|
||||
for d in defects:
|
||||
image = self.inpaint(image, d.mask, prompt=d.fix_prompt)
|
||||
return image
|
||||
|
||||
def upscale(self, image):
|
||||
"""Stage 5: Upscale."""
|
||||
return self.upscaler.enhance(image, scale=4)
|
||||
```
|
||||
|
||||
### Reference-driven generation
|
||||
```python
|
||||
def generate_with_references(prompt, style_ref=None, character_ref=None):
|
||||
parts = [prompt]
|
||||
if style_ref:
|
||||
parts.append(f"--sref {style_ref}")
|
||||
if character_ref:
|
||||
parts.append(f"--cref {character_ref}")
|
||||
|
||||
full_prompt = " ".join(parts)
|
||||
return midjourney.generate(full_prompt)
|
||||
```
|
||||
|
||||
### Prompt builder (5-layer)
|
||||
```python
|
||||
def build_prompt(subject, medium, env, lighting, params):
|
||||
return f"{subject}, {medium}, {env}, {lighting} {params}"
|
||||
|
||||
prompt = build_prompt(
|
||||
subject="elegant woman, age 30, blue eyes, smiling",
|
||||
medium="oil painting, Renaissance style",
|
||||
env="close-up portrait, marble background",
|
||||
lighting="chiaroscuro, dramatic light, volumetric",
|
||||
params="85mm lens, shallow depth of field --ar 3:2 --s 500"
|
||||
)
|
||||
```
|
||||
|
||||
### Batch + cost optimization
|
||||
```python
|
||||
def cost_aware_batch(prompts, target='exploration'):
|
||||
if target == 'exploration':
|
||||
return [generate(p, draft=True, steps=10) for p in prompts]
|
||||
elif target == 'production':
|
||||
return [generate(p, steps=50, upscale=True) for p in prompts]
|
||||
```
|
||||
|
||||
## 🤔 결정 기준
|
||||
|
||||
| Goal | Workflow |
|
||||
|---|---|
|
||||
| Brand campaign | sref + multi-iteration + post-edit |
|
||||
| Character consistency | cref / oref + LoRA |
|
||||
| Quick concept | Draft mode |
|
||||
| Final polish | Full quality + post-edit + upscale |
|
||||
| Video | Sora / Veo / Runway |
|
||||
| 3D asset | TripoSR / InstantMesh |
|
||||
|
||||
**기본값**: 5-layer prompt + draft mode + reference + post-edit + upscale 의 sequence.
|
||||
|
||||
## 🔗 Graph
|
||||
- 부모: [[AI Image Generation]]
|
||||
- 변형: [[Draft-Mode]] · [[Omni Reference]]
|
||||
- Tools: [[Midjourney-V7]] · [[Flux]]
|
||||
|
||||
## 🤖 LLM 활용
|
||||
**언제**: 매 commercial creative project. 매 visual brand.
|
||||
**언제 X**: 매 throwaway. 매 highly specific artist (legal).
|
||||
|
||||
## ❌ 안티패턴
|
||||
- **Single prompt 의 expectation**: cliche / generic.
|
||||
- **No reference**: brand inconsistency.
|
||||
- **Skip post-edit**: defect ship.
|
||||
- **Generic vocab ("nice picture")**: 매 specific 의 더 좋음.
|
||||
- **Full quality from start**: cost 폭발.
|
||||
|
||||
## 🧪 검증 / 중복
|
||||
- Verified.
|
||||
- 신뢰도 B.
|
||||
- Overlap with [[AI Image Generation]] / [[AI 모델 사후 편집 도구 (Post-editing Tools)|Post-editing-Tools]] / [[Image-Workflow]].
|
||||
|
||||
## 🕓 Changelog
|
||||
| 날짜 | 변경 |
|
||||
|---|---|
|
||||
| 2026-05-08 | Phase 1 |
|
||||
| 2026-05-09 | Manual cleanup — paradigm shift + 5-layer + workflow + emerging tech |
|
||||
Reference in New Issue
Block a user