blur/canny/depth controlnets for sd3.5_medium support

[yixingjia]

I only have a legacy GPU with 24 GB vram, When I try to run sd3.5_large with controlenet blur it reports RuntimeError: CUDA error: CUBLAS_STATUS_ALLOC_FAILED when calling cublasCreate(handle) Seems lack of vram.

Then I try to run the sd3.5_medium model with the same controlenet

it report Loading SD3 model sd3.5_medium.safetensors...
Skipping key 'context_embedder.bias' in safetensors file as 'context_embedder' does not exist in python model
Skipping key 'context_embedder.weight' in safetensors file as 'context_embedder' does not exist in python model
W: shape mismatch for key controlnet_blocks.0.bias, torch.Size([1536]) != torch.Size([2432])
W: shape mismatch for key controlnet_blocks.0.weight, torch.Size([1536, 1536]) != torch.Size([2432, 2432])
W: shape mismatch for key controlnet_blocks.1.bias, torch.Size([1536]) != torch.Size([2432])
W: shape mismatch for key controlnet_blocks.1.weight, torch.Size([1536, 1536]) != torch.Size([2432, 2432])
W: shape mismatch for key controlnet_blocks.10.bias, torch.Size([1536]) != torch.Size([2432])
W: shape mismatch for key controlnet_blocks.10.weight, torch.Size([1536, 1536]) != torch.Size([2432, 2432])
W: shape mismatch for key controlnet_blocks.11.bias, torch.Size([1536]) != torch.Size([2432])

Any suggestions? Do we have a medium version controlnets for sd3.5_medium?

[yixingjia]

I find the follow:
All currently released ControlNets are compatible only with Stable Diffusion 3.5 Large (8b).
source: https://huggingface.co/stabilityai/stable-diffusion-3.5-controlnets
So this is expected....