Prompt Search


프롬트 학습관련 질문 2

June 27, 2023, 11:42 p.m.

favorite

31


parameters Best Quality, masterpiece, high raw DPI, intricate details, (American 14yo loli girl from California high_end middle school in style by Murata Renji:1.2), (character design by Murata Renji:1.3), fashion forward, ((Edgar Degas)), teen fashion magazine, forever 21, dynamic sassy poses, teen vogue, slender healthy fit, short brown hair, (girl drawn in Impressionist drawing technique), (((angry kuso_gaki))), (((open smirking mouth))), (((disgusted, raged, disappointing, in_pain))), (marked by intricate layers that achieved depth and luminosity with contrast from use of shadow, brushwork ranged from smooth glazing strokes to impasto for textural richness, ((skillful use of chiaroscuro)), the interplay of light and shadow, brought volume and drama to his subjects, radiant quality in skin tones and fabrics, perfect anatomical precision with dramatic dynamic action poses, extensive palette, combined with an understanding of color theory, enabled vibrant harmonies and contrasts with symphony of layers, shadow, and deep colors that underpinned the emotional and grand narratives), ((grabbing small teddybear doll))
negative_prompt (worst quality, low quality:1.3), 3D face, cropped, lowres, text, jpeg artifacts, signature, watermark, username, blurry, artist name, trademark, title, oriental, asian, multiple view, Reference sheet, bad anatomy, red nose, ((adult, grown_up)), kemonomimi, kumamim
steps 20 content_copy
sampler Euler a content_copy
cfg_scale 8 content_copy
seed 2163261948 content_copy
size 1024x1216 content_copy
model_hash ac3de0bb2a content_copy PROMPT SEARCH: PTSearch XL 3D v15 open_in_new
Copy All

여러가지 프롬트를 조합해 새로운 캐릭터를 만드는 경우 만들어지는 이미지 중 원하지 않는 결과물을 지우고 원하는 결과물만을 남겨 놓아 가는 방식으로 학습을 시키면 이러한 누적된 이미지들이 학습을 좀더빨리 그리고 효과적으로 시키는데 도움이 될까요?

favorite 31

Comments

캐릭터를 만드는걸 말씀하시는건가요?
아니면 새로운 모델 학습하는걸 말씀하시는건가요?

어느경우든 간에 일반적으로 데이터를 정제하는게 어느정도 도움은 될지 모르겠지만,
모델의 출력을 다시 학습 입력으로 넣는다는 흐름상으로 볼때,
과적합이 발생해서 문제가 발생할 가능성이 높아보입니다.

June 29, 2023, 12:56 p.m.

학습인하여 목표치를 향해 가는데 샛길로 가다가 시간이 더 소요 하거나 잘못된 길로 빠질 수 있다는 거군요. 답변 감사합니다.

June 29, 2023, 1:35 p.m.