ControlNet and Kohya HRFix interop support

Is there an existing issue for this?

• I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

我经常使用高分辨率的图生图，会使用CN的lineArt和Tile两个模型的样子。而且我发现，小分辨率的时候（1000px），SD生成的图，光影和结构更加正确、优秀，超过2000px的时候就会变得很平、很容易出错的样子，无论是结构上，还是光影。

然后我使用了内置的Kohya HRFix，它的工作机制就是先缩放分辨率迭代出正确的结构，然后再提升会原本的分辨率，而且可以控制参与的步骤的样子。在文生图（1500x1500，甚至2000x2000以上），能得到结构正确且非常优秀的画面。
我就想着试着搭配CN模型一起，进一步控制画面的样子，然后就出现了报错的样子，在Kohya HRFix起作用的阶段中，CN就会出现
warning control could not be applied torch.Size([2, 1280, 11, 8]) torch.Size([2, 1280, 32, 22])
warning control could not be applied torch.Size([2, 1280, 11, 8]) torch.Size([2, 1280, 32, 22])
warning control could not be applied torch.Size([2, 1280, 11, 8]) torch.Size([2, 1280, 32, 22])
warning control could not be applied torch.Size([2, 1280, 22, 15]) torch.Size([2, 1280, 64, 43])
warning control could not be applied torch.Size([2, 1280, 22, 15]) torch.Size([2, 1280, 64, 43])
warning control could not be applied torch.Size([2, 640, 22, 15]) torch.Size([2, 640, 64, 43])
warning control could not be applied torch.Size([2, 640, 43, 29]) torch.Size([2, 640, 128, 86])
warning control could not be applied torch.Size([2, 640, 43, 29]) torch.Size([2, 640, 128, 86])
然后并不起作用的样子。因为这个分辨率和CN设定的原本的分辨率不匹配的样子。

所以我特别希望CN也能自动匹配和Kohya HRFix的流程。

Proposed workflow

1. 到图生图/文生图，写好参数和设置
2. 启用Kohya HRFix Integrated，设置好参数
3. 启用CN，设置好参数和模型
4. CN能和HRFix一起正常运转

Additional information

No response

Reproduce the documented text-to-image or image-to-image workflow with Integrated Kohya HRFix and ControlNet, then inspect the ControlNet integration at the HRFix stage where the reported tensor-shape warnings occur. Done means ControlNet applies during HRFix without those warnings and preserves the configured conditioning.