meeting-103 v2.1 clean R7+R8 🏆 BREAKTHROUGH: dpmpp_2m karras = pastel 82.43%

R7 (3/5 success, R6-2 sweet spot 확인):
- 7-1: PuLID 0.85 (default ref) → 64.45% (less than R6-2)
- 7-2: PuLID 0.6 (default ref) → 65.93%
- 7-3: seed 999777 → CUDA error
- 7-4: Stage 3 Refiner d=0.20 → CUDA error
- 7-5: Pony V6 60갑자 calligraphy → 18.09% (검정 회피 OK)

R8 (5/5 success, 🏆 거대한 발견):
- 8-1: seed 200000 → 65.73% (variance -3.93pp)
- 8-2: seed 555555 → 64.30%
- 8-3: guidance 2.5 → 62.71% (낮은 guidance = 파스텔 감소)
- 8-4: guidance 5.0 → 69.99% (R6-2와 동일)
- 8-5: dpmpp_2m karras → 🏆 82.43%  전체 best (+12.77pp)

 핵심 발견 2: 샘플러 변경의 dramatic effect
   euler/simple (R6-2 69.66%) → dpmpp_2m/karras (R8-5 82.43%) = +12.77pp

누적 진행 (R9 legacy → R8):
44.5% → 48.5% → 49.99% → 50.47% → 56.41% → 69.66% → 82.43% = +37.93pp

자산 라이브러리:
- workflows: winner_auto 11 + winner_auto_high 4 + winner_auto_breakthrough 1
- 8-5 워크플로 = 타로/12지신/별이 시리즈 재활용 표준 베이스
This commit is contained in:
choijaewook
2026-05-20 00:15:56 +09:00
parent e1cb7eb21b
commit f34be5f58f
29 changed files with 2134 additions and 24 deletions

View File

@@ -29,7 +29,7 @@
],
"models": {
"pony-v6": {
"rounds_used": 8,
"rounds_used": 9,
"dimensions": {
"byeolyi": [],
"hanja": [],
@@ -48,7 +48,8 @@
30.8,
14.1,
29.9,
77.3
77.3,
18.09
],
"auto_black": [
0.61,
@@ -62,7 +63,8 @@
0.41,
0.17,
0.17,
0.03
0.03,
0.02
]
},
"avg_scores": {},
@@ -78,7 +80,7 @@
"verdict": "파스텔 평균 30.4% (12회) / 검정 평균 0.23%"
},
"flux-dev": {
"rounds_used": 37,
"rounds_used": 44,
"dimensions": {
"byeolyi": [],
"hanja": [],
@@ -117,7 +119,14 @@
48.65,
52.99,
69.66,
15.09
15.09,
64.45,
65.93,
65.73,
64.3,
62.71,
69.99,
82.43
],
"auto_black": [
3.11,
@@ -151,7 +160,14 @@
3.7,
0.5,
0.05,
12.46
12.46,
0.01,
0.19,
0.03,
0.05,
0.0,
0.47,
0.0
]
},
"avg_scores": {},
@@ -218,12 +234,20 @@
"variant": "PuLID 0.7 + Redux 1.0 (ref byeolyi-default.png)",
"pastel_pct": 69.66,
"black_pct": 0.05
},
{
"round": "R8-clean",
"id": "8-5",
"variant": "R6-2 baseline + dpmpp_2m karras",
"pastel_pct": 82.43,
"black_pct": 0.0,
"_BREAKTHROUGH": true
}
],
"verdict": "파스텔 평균 32.1% (27회) / 검정 평균 3.65%"
},
"pulid-flux": {
"rounds_used": 31,
"rounds_used": 38,
"dimensions": {
"byeolyi": [],
"hanja": [],
@@ -255,7 +279,14 @@
50.36,
49.85,
52.99,
69.66
69.66,
64.45,
65.93,
65.73,
64.3,
62.71,
69.99,
82.43
],
"auto_black": [
0.73,
@@ -282,7 +313,14 @@
0.86,
0.08,
0.5,
0.05
0.05,
0.01,
0.19,
0.03,
0.05,
0.0,
0.47,
0.0
]
},
"avg_scores": {},
@@ -358,6 +396,14 @@
"variant": "PuLID 0.7 + Redux 1.0 (ref byeolyi-default.png)",
"pastel_pct": 69.66,
"black_pct": 0.05
},
{
"round": "R8-clean",
"id": "8-5",
"variant": "R6-2 baseline + dpmpp_2m karras",
"pastel_pct": 82.43,
"black_pct": 0.0,
"_BREAKTHROUGH": true
}
],
"verdict": "파스텔 평균 36.9% (22회) / 검정 평균 2.56%"
@@ -399,7 +445,7 @@
"verdict": "Stage 2 인페인트 6차 실패 — Qwen 계열 GGUF Q4 환경 호환성 결함 확정"
},
"flux-redux": {
"rounds_used": 14,
"rounds_used": 21,
"dimensions": {
"byeolyi": [],
"hanja": [],
@@ -420,7 +466,14 @@
50.36,
49.85,
52.99,
69.66
69.66,
64.45,
65.93,
65.73,
64.3,
62.71,
69.99,
82.43
],
"auto_black": [
0.27,
@@ -436,7 +489,14 @@
0.86,
0.08,
0.5,
0.05
0.05,
0.01,
0.19,
0.03,
0.05,
0.0,
0.47,
0.0
]
},
"avg_scores": {},
@@ -503,6 +563,14 @@
"variant": "PuLID 0.7 + Redux 1.0 (ref byeolyi-default.png)",
"pastel_pct": 69.66,
"black_pct": 0.05
},
{
"round": "R8-clean",
"id": "8-5",
"variant": "R6-2 baseline + dpmpp_2m karras",
"pastel_pct": 82.43,
"black_pct": 0.0,
"_BREAKTHROUGH": true
}
],
"verdict": "파스텔 평균 49.0% (11회) / 검정 평균 0.34%"
@@ -589,6 +657,8 @@
"R3-clean",
"R4-clean",
"R5-clean",
"R6-clean"
"R6-clean",
"R7-clean",
"R8-clean"
]
}