Sign In

AnimeNAIV3StyleMIX(模拟NAIV3动漫风格)

293
1.8k
45
Updated: Dec 11, 2023
base modelanimemodelcute
Verified:
SafeTensor
Type
Checkpoint Merge
Stats
1,773
Reviews
Published
Dec 11, 2023
Base Model
SD 1.5
Hash
AutoV2
E93D60829D
default creator card background decoration
D0pa's Avatar
D0pa

recommended parameters:
Steps: 16, Sampler: DPM++ 2M Karras, CFG scale: 7,or Steps: 23, Sampler: Euler, CFG scale: 7.

Size: 512x792.

Hires upscaler: Latent (nearest-exact),Denoising strength: 0.6.

cuteVersion1

挑取了自己NAI3中手脚不崩坏的图,再分层融合了手脚相对正常的模型所产生的版本,尽可能保留了原有的上色风格。但由于要挑选正常的图片,导致这个版本的训练图片数目还不足,所以在某些场景下的泛化能力应该会很差。

Selected images with intact hands and feet from NAI V3, and then used Merging Block Weights to fuse versions generated by models with relatively normal limbs, attempting to preserve the original coloring style as much as possible. However, due to the need to select normal images, the number of training images for this version is insufficient, so the generalization ability in certain scenarios may be poor.