74a0415e8c
R3 (4/5 success): - 3-1: PuLID 0.7 + Redux 0.85 → 50.47% (R2 best 49.99% 갱신) - 3-2: PuLID 0.6 + Redux 0.7 → 49.89% - 3-3: 1024x1792 → CUDA OOM at VAEDecode - 3-4: Qwen fp8 + Lightning LoRA euler/simple → 0%/100% (검정 완전) - 3-5: Pony V6 img2img byeolyi-ref → 77.3% (img2img 0.65 denoise, 원본 색 보존) R4 (3/5 success, 별이 새 best): - 4-1: PuLID 0.7 + Redux 1.0 (max) → 56.41% ★ 전체 best - 4-2: PuLID 0.85 + Redux 0.7 → 50.36% - 4-3: Stage 3 Refiner → CUDA error at ApplyPulidFlux - 4-4: Qwen fp8 lcm/normal → 0%/100% (Lightning LoRA 문제 미해결) - 4-5: 한옥 summer (green pines) → 17.7%/3.97% 학습: - Redux strength 0.5 → 0.7 → 0.85 → 1.0 단조 증가 (56.41% sweet spot) - PuLID 0.7 fixed, Redux 1.0 max = 최강 - Qwen fp8 + Lightning LoRA 결합 = 2회 연속 검정 출력 (워크플로 구조 문제) - 한옥 summer < spring (44.45%) — 색감 자체가 차분 진행: R9 legacy 44.5% → R1 48.5% → R2 49.99% → R3 50.47% → R4 56.41% (+11.91pp) 자산 라이브러리: - workflows: winner_auto 7 + partial 7 - prompts/failed: 9 패턴 (Qwen 계열 2건 추가)
118 lines
4.1 KiB
JSON
118 lines
4.1 KiB
JSON
{
|
||
"qwen-image-2512-GGUF-Q4-SamplerCustomAdvanced-v4": {
|
||
"attempts": 4,
|
||
"patterns": [
|
||
"EmptyLatentImage",
|
||
"EmptyQwenImageLayeredLatentImage",
|
||
"ModelSamplingAuraFlow+EmptySD3LatentImage",
|
||
"SamplerCustomAdvanced+ModelSamplingAuraFlow"
|
||
],
|
||
"all_results": "5KB 검정 latent",
|
||
"hypothesis": "Qwen-Image GGUF Q4_K_S + CUDA 13.0 + PyTorch 2.12 환경 호환성 또는 다른 quant 필요 (Q8/fp16)",
|
||
"next_attempts": [
|
||
"non-GGUF Qwen-Image safetensors",
|
||
"Lightning LoRA 적용",
|
||
"다른 sampler (dpmpp_2m/ddim)",
|
||
"다른 shift 값 (1.73/5.0)"
|
||
]
|
||
},
|
||
"pulid-flux-fusion-concat-max-R2": {
|
||
"fusion_modes_tested": [
|
||
"concat",
|
||
"max"
|
||
],
|
||
"weight": 0.85,
|
||
"result": "워크플로우 에러로 실행 안 됨 (history 누락)",
|
||
"next_attempts": [
|
||
"fusion=mean weight 변동 (R1 mean 0.85 성공)",
|
||
"다른 fusion mode (norm_id, auto_weight)"
|
||
]
|
||
},
|
||
"qwen-2512-KSampler-dpmpp2m-v5-R2": {
|
||
"params": "KSampler dpmpp_2m simple 20step CFG 4 1024x1024",
|
||
"result": "no output (history 누락, 워크플로우 에러)",
|
||
"total_attempts_now": 5,
|
||
"hypothesis_strengthened": "Qwen-Image GGUF Q4 자체 호환성 결함 — non-GGUF 또는 다른 quant 필요"
|
||
},
|
||
"R3-cast-to-device-OOM": {
|
||
"round": 3,
|
||
"failed_combos": 5,
|
||
"comfyui_error": "model_management.cast_to_device — dtype/VRAM 부담",
|
||
"vram_free_at_error": "4.6GB / 12GB total",
|
||
"all_workflows_affected": [
|
||
"PuLID mean w0.5",
|
||
"PuLID mean w1.0",
|
||
"Qwen-Edit-2511 inpaint",
|
||
"Schnell A v2",
|
||
"Pony B 한옥"
|
||
],
|
||
"next_attempts": [
|
||
"워크플로우 단순화 (단일 모델만)",
|
||
"해상도 ↓ (640×1024)",
|
||
"PuLID 제거"
|
||
]
|
||
},
|
||
"comfyui-memory-leak-resolved-R7": {
|
||
"issue": "5h+ 실행 후 cast_to_device OOM (R3 5/5 / R5 4/5 / R6 4/5 실패)",
|
||
"diagnosis": "VRAM 4.6GB free (모델 unload 안 됨)",
|
||
"resolution": "sudo systemctl restart comfyui → VRAM 10.98GB free → R7 5/5 100% 성공",
|
||
"operational_protocol": "긴 R&D 세션 시 5라운드마다 ComfyUI 재시작 권장"
|
||
},
|
||
"qwen-family-gguf-Q4-environment-incompatibility-CONFIRMED-6attempts": {
|
||
"models_tested": [
|
||
"qwen-image-2512-Q4",
|
||
"qwen-image-edit-2511-Q4"
|
||
],
|
||
"samplers_tested": [
|
||
"KSampler euler simple",
|
||
"SamplerCustomAdvanced",
|
||
"KSampler dpmpp_2m simple",
|
||
"KSampler euler simple Edit"
|
||
],
|
||
"patterns_tested": [
|
||
"EmptyLatent",
|
||
"EmptyQwenLayered",
|
||
"ModelSamplingAuraFlow+SD3",
|
||
"SamplerCustomAdvanced+ModelSampling",
|
||
"1024x1024 KSampler dpmpp_2m",
|
||
"Qwen-Edit 인페인트 denoise 0.7",
|
||
"Qwen-Edit denoise 0.85"
|
||
],
|
||
"verdict": "GGUF Q4 + CUDA 13.0 + PyTorch 2.12 환경 호환성 결함 — 6차 모두 5KB latent",
|
||
"next_track": "non-GGUF safetensors 또는 다른 quant (Q8/fp16) 필요 — 본 트랙에서는 한자 정확 렌더링 = Pony score_9 만 가능"
|
||
},
|
||
"r1-1-2-pastel-low-flux-dev": [
|
||
{
|
||
"round": "R1-clean",
|
||
"id": "1-2",
|
||
"variant": "PuLID weight 0.85 mean delayed_start 0.05 (concat-replacement)",
|
||
"pastel_pct": 9.99,
|
||
"black_pct": 13.98,
|
||
"note": "auto-classified pastel-low (delayed start_at + concat replacement)"
|
||
}
|
||
],
|
||
"flux-dev-pulid-redux-1024x1792-VAEDecode-OOM-R3": [
|
||
{
|
||
"round": "R3-clean",
|
||
"id": "3-3",
|
||
"variant": "PuLID 0.7 + Redux 0.7 @ 1024x1792",
|
||
"error": "CUDA invalid argument at VAEDecode (memory pressure with 768->1024 upscale)"
|
||
}
|
||
],
|
||
"qwen-fp8-lightning-lora-scheduler-simple-cfg1-black-output-R3": [
|
||
{
|
||
"round": "R3-clean",
|
||
"id": "3-4",
|
||
"variant": "Qwen fp8 + Lightning LoRA 4-step euler/simple cfg=1.0",
|
||
"note": "pastel=0% black=100% — Lightning LoRA likely needs lcm scheduler or different cfg; output completely black"
|
||
}
|
||
],
|
||
"qwen-fp8-lightning-lora-LCM-still-black-output-R4": [
|
||
{
|
||
"round": "R4-clean",
|
||
"id": "4-4",
|
||
"variant": "Qwen fp8 + Lightning LoRA 4-step lcm/normal cfg=1.0",
|
||
"note": "pastel=0% black=100% — lcm scheduler 도 해결 X, Lightning LoRA strength_clip=1.0 부적절 의심"
|
||
}
|
||
]
|
||
} |