目前加上此lora后会对语义理解尤其角度有明显的勿扰,

#12
by 52Hz-p - opened

不知道是我参数问题还是什么问题,我测试了很多参数,很多模型,加与不加Hyper的图在语义理解上有很明显的错误尤其在有视角等带有方向性的token时候,仅此而已,其他都很完美,包括速度质量,目前加速模型里可以说最好的了

ByteDance org
edited Apr 26

Hi, @52Hz-p
Thanks for you attention! We will try our best to improve it.
You could also share your experience on the way of tuning parameters with others here! 👏

ByteDance org
edited Apr 30

Hi, @52Hz-p
We have uploaded the CFG-preserved hyper-SD15 LoRA and hyper-SDXL LoRA just now, higher cfg and negative prompts may be helpful in your usecase, looking forward to your use and feedback!

image.png

You can test resize it into rank1

Sign up or log in to comment