비교 대상 : 베타( https://github.com/aria1th/stable-diffusion-webui/commit/632b9a556e35a471b1479e4d1f0b1d39e5854de6 ) 버전과 d4790fa버전
사용 하이퍼네트워크 - Momoko https://arca.live/b/hypernetworks/61478319
대상 프롬프트
hyper detailed, ((realistic)) ,artistic illustration, ((bishoujyo)), white skin, golden hair, wearing crown, princess wearing white dress, princess is sitting in throne, beautiful cute face, ((Rembrandt lighting)) from stained glass and inside cathedral, sunny outside, [japanese loli : photo, ((loli)), (intricate_detail) , finely detailed, only 1girl,summer, sweating, first-rate female body, sweating, best quality, solo focus, and close up for face and upper body shot, single scene: 0.25] drawn by tamano kedama and asanagi and momoko
Negative prompt: lowres,bad anatomy,bad hands,text,error,missing fingers,extra digit,fewer digits,cropped,worst quality,low quality,normal quality,jpeg artifacts,signature,watermark,username,blurry,missing arms,long neck,Humpbacked,malformed limbs,poorly drawn,poorly drawn hands,mutilated,missing legs,{unclear eyes},{more than 2 thighs},deformed,extra legs,mutated hands and fingers,three legs,bad face,error,extra limbs,worstquality,low quality,{bad anatomy}, {{malformed}}, giant, ((multiple views))
Steps: 40, Sampler: Euler a, CFG scale: 2.5, Seed: 3568245996, Size: 512x512, Model hash: 925997e9, Hypernet: momoko-1221-style-32000, ENSD: 31337
베타 버전
HypernetworkModule(
(linear): Sequential(
(0): Linear(in_features=640, out_features=1280, bias=True)
(1): Hardswish()
(2): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)
(3): Dropout(p=0.3, inplace=False)
(4): Linear(in_features=1280, out_features=1280, bias=True)
(5): Hardswish()
(6): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)
(7): Linear(in_features=1280, out_features=640, bias=True)
(8): Hardswish()
(9): LayerNorm((640,), eps=1e-05, elementwise_affine=True)
이전 버전
HypernetworkModule(
(linear): Sequential(
(0): Linear(in_features=640, out_features=1280, bias=True)
(1): Hardswish()
(2): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)
(3): Dropout(p=0.3, inplace=False)
(4): Linear(in_features=1280, out_features=1280, bias=True)
(5): Hardswish()
(6): LayerNorm((1280,), eps=1e-05, elementwise_affine=True)
(7): Linear(in_features=1280, out_features=640, bias=True)
(8): Hardswish()
(9): LayerNorm((640,), eps=1e-05, elementwise_affine=True)
)
완전히 동일한 것을 확인합니다
