Open cqflgl opened 5 months ago
对于Turbo和LCM模型需要将Style Strength Ratio调低一些以保证更好的identity. For Turbo and LCM models, the Style Strength Ratio needs to be lowered to ensure better identity.
Please refer to this URL: https://zenn.dev/neoai/articles/cedda3d17f0a70, for using LCM-LoRA with our PhotoMaker. This blog also provides many analyses. It is interesting.
总体效果还是令人惊艳的,感谢开源!生成时如果能参考原图的脸型就更好了。
原图是长脸,基底模型是SG161222/RealVisXL_V3.0_Turbo, prompt = "closeup portrait, girl img, from front, short black wavy hair, pale complexion, white lace, sky-blue dress, puff sleeves, earings, necklace, simple pure color background", negative_prompt = "nsfw, lowres, bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, blurry, username, text", 可以看出,生成结果的脸型和原图还是有明显区别的。
原图:![原图](https://github.com/TencentARC/PhotoMaker/assets/62335372/33a65698-846e-47d1-b8a3-aef891170c5e)
生成图:
![RealVisXL-2](https://github.com/TencentARC/PhotoMaker/assets/62335372/765b162f-38ff-48d3-8799-052b86fe2ff8)