wikeeyang

/

Flux2-Klein-9B-True-V2

like 131

you image example is useless , sorry
4 step original
25 your model

the model can add Turbo or Distill LoRA then use less steps.

i see, so your model is a base_model ... nevertheless the comparison is not really comparable

i see, so your model is a base_model ... nevertheless the comparison is not really comparable

Yes, my model is a base model. I think it only makes sense to compare if it can produce higher-quality images. Thanks.

还是存色差 LCS 好像没作用，有案例？

还是存色差 LCS 好像没作用，有案例？

可以参考我的工作流，颜色和锐度参数，需要尝试和调整，另外，目前没有办法完全消除色差，只是通过各种办法使色差减少到尽可能小的范围。

感觉在一致性的保持上没有V1好

感觉在一致性的保持上没有V1好

也有网友这样反馈，主要原因是 V2 优化了构图和美学，微调幅度大了一些，对图像编辑与重绘的一致性有一些影响。可以尝试添加一致性 LoRA 来保持更好的重绘一致性。

感觉在一致性的保持上没有V1好

我也遇到了，感觉编辑模型还是要首先保证一致性

感觉在一致性的保持上没有V1好

也有网友这样反馈，主要原因是 V2 优化了构图和美学，微调幅度大了一些，对图像编辑与重绘的一致性有一些影响。可以尝试添加一致性 LoRA 来保持更好的重绘一致性。

brother are you in touch with lrzjason or not? he worked on consistency loras as well as him "https://huggingface.co/dx8152".

感觉在一致性的保持上没有V1好

也有网友这样反馈，主要原因是 V2 优化了构图和美学，微调幅度大了一些，对图像编辑与重绘的一致性有一些影响。可以尝试添加一致性 LoRA 来保持更好的重绘一致性。

brother are you in touch with lrzjason or not? he worked on consistency loras as well as him "https://huggingface.co/dx8152".

take a look as below, compare from lrzjason:

you image example is useless , sorry
4 step original
25 your model

Could you resolve your own problem and compare the 25-step 9b-True to the 25-step 9b-Base? We'd like to see complex scenes not portraits.

Thank you.

there is also always the question how much a lora or finetune was focused on young ladies compared to just ... everything, which is obviously a lot more effort. .
We all know that is a always main focus of things, but it makes a lot of those kind of pointless for general consistency use.

And i obviously agree, comparing a distilled model/model+lora to a full 25 step base run isn't really fair.