How to train LoRA with input conditional image?

[Adenialzz]

Hi, thanks for open-sourcing. I wonder how can I train IC-LoRA with input conditional image?

[flankechen]

same question, and maybe the training is cond image->vae half, gt image->vae half?

[bozagina]

same question, and maybe the training is cond image->vae half, gt image->vae half?

Hi, does the training with input conditional image use aitoolkit?

[Orannue]

@Adenialzz @flankechen @bozagina Hello, I’m facing the same problem. Have you found a solution?

[bozagina]

@Adenialzz @flankechen @bozagina Hello, I’m facing the same problem. Have you found a solution?

I have no idea; it's still a mystery to me.

[flankechen]

@Adenialzz @flankechen @bozagina Hello, I’m facing the same problem. Have you found a solution?

seems no current implementation(AI toolkit/kohyass sdscripts) with cond image.
but check this out, it propose that without iclora, their prompting and attention enhance works with cond image input.
https://diptychprompting.github.io/

[JoshonSmith]

@Adenialzz @flankechen @bozagina Hello, I’m facing the same problem. Have you found a solution?

seems no current implementation(AI toolkit/kohyass sdscripts) with cond image. but check this out, it propose that without iclora, their prompting and attention enhance works with cond image input. https://diptychprompting.github.io/

In fact, I can not generate the result in the flow figure of this paper

[flankechen]

@Adenialzz @flankechen @bozagina Hello, I’m facing the same problem. Have you found a solution?

seems no current implementation(AI toolkit/kohyass sdscripts) with cond image. but check this out, it propose that without iclora, their prompting and attention enhance works with cond image input. https://diptychprompting.github.io/

In fact, I can not generate the result in the flow figure of this paper

yeap, this paper show really cherry-pick result. no official code released.
but in my iclora test and my implementation of dypitch attn, it helps.

[flankechen]

@Adenialzz @flankechen @bozagina Hello, I’m facing the same problem. Have you found a solution?

seems no current implementation(AI toolkit/kohyass sdscripts) with cond image. but check this out, it propose that without iclora, their prompting and attention enhance works with cond image input. https://diptychprompting.github.io/

In fact, I can not generate the result in the flow figure of this paper

and https://github.com/ali-vilab/ACE_plus this new paper, that release flux fill based lora, shows really good result.