Closed Creative-comfyUI closed 2 months ago
you are probably using the wrong image encoder
Here is what I did
I use IPadapter Model Loader and Clip Vision with Ipadapter Stye & Composition SDXL, and it tried all the IPadapter model and all the Clip Vision model. With some of them I get the error I mentioned. For the Clip Vision, it seems to always happen for some of them, no matter what I did. It is not always the same error. (I follow your example)
I thought that using a model that is SD with a composition SDXL was the problem. It is not the case as it seems, I explain in the general discussion, I explain that we can mix SD and SDXL without problems, but not always. Look at what I wrote, as it follows that discussion in a certain sense
https://github.com/cubiq/ComfyUI_IPAdapter_plus/discussions/510
This is the image for iploader bug (the 79k works perfectly that is why I presume the bug came for the IpLoader)
This is the image for clipVision bug
Thanks
base sdxl model requires vit-g encoder, sdxl plus the vit-h
I have this error. What it is the problem ? Thanks The checkpoint model is an SDXL Latent size 1024 and Ipadapter model is idapadpt_sdxl.safetensors
Error occurred when executing IPAdapterStyleComposition: Error(s) in loading state_dict for ImageProjModel: size mismatch for proj.weight: copying a param with shape torch.Size([8192, 1280]) from checkpoint, the shape in current model is torch.Size([8192, 1024]).