Kijai

/

SkyReels-V1-Hunyuan_comfy

like 184

If that's with the native workflow, ComfyUI itself needs to be the latest nightly version, if it's with the wrapper then same for that.

I have the updated version, but still get the error

I have the updated version, but still get the error

Only way to get that error is if the nodes are not up to date, do a re-install to make sure.

The issue lies in your workflow, see where you are passing the HunyuanVideo Encode output. They should point to img_cond_latents. You can try using the workflow provided with ckpts. https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/blob/main/skyreels_hunyuan_I2V_native_example_01.json

RuntimeError: Error(s) in loading state_dict for HunyuanVideo:
size mismatch for img_in.proj.weight: copying a param with shape torch.Size([3072, 32, 1, 2, 2]) from checkpoint, the shape in current model is torch.Size([3072, 16, 1, 2, 2]).
My GPU is 2080Ti
Please help me

Make sure you update comfyui, torch & cuda

Run the update_comfyui.bat from the update folder

Go back to your python_embeded folder

Click on the file directory bar at the top, type cmd then hit enter

In cmd type "python.exe -m pip install --upgrade torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu126"

Also note that if you use ComfyUI manager to upgrade, it's probably set to update ComfyUI to release versions not nightly so you need to hit the change version button (below the update buttons)

Same error here: size mismatch for img_in.proj.weight: copying a param with shape torch.Size([3072, 32, 1, 2, 2]) from checkpoint, the shape in current model is torch.Size([3072, 16, 1, 2, 2]).

I have windows (Nvidia) beta from https://www.comfy.org/download

• Python Version 3.12.8 (main, Jan 14 2025, 22:49:36) [MSC v.1942 64 bit (AMD64)]
• Embedded Python false
• Pytorch Version 2.6.0+cu126
• ComfUI v0.3.14.
• Comfyui frontend 1.9.19.
• Comfyui_desktop v.0.4.20.
• Comfyui_Manager 3.17.7.
• "you need to hit the change version button (below the update buttons)" No such button in C Manager or settings
• "Run the update_comfyui.bat from the update folder" no such folder in Comfyui
• "You can try using the workflow provided with ckpts. https://huggingface.co/Kijai/SkyReels-V1-Hunyuan_comfy/blob/main/skyreels_hunyuan_I2V_native_example_01.json" not working
• "Go back to your python_embeded folder" no such folder in Comfyui

Problem still persists: Given groups=1, weight of size [3072, 32, 1, 2, 2], expected input[1, 16, 13, 68, 68] to have 32 channels, but got 16 channels instead

Uninstalled Comfy, reinstalled but not the desktop version.

• Python Version 3.12.8 (tags/v3.12.8:2dc476b, Dec 3 2024, 19:30:04) [MSC v.1942 64 bit (AMD64)]
• Embedded Python true
• Pytorch Version 2.6.0+cu126
• ComfUI v0.3.15.
• Comfyui frontend 1.9.18.
• Comfyui_Manager 3.25.1.

@Genzo9319 does your cuda version 12.6 to match your pytorch cu126, nvidia GPU

Given groups=1, weight of size [3072, 32, 1, 2, 2], expected input[1, 16, 13, 68, 68] to have 32 channels, but got 16 channels instead

This error simply means either that your ComfyUI does not have the required code changes to load the model. This was added to ComfyUI main repository 4 days ago in this commit:

https://github.com/comfyanonymous/ComfyUI/commit/acc152b674fd1c983acc6efd8aedbeb380660c0c

It's possible that whichever way you're using to install ComfyUI is not using the latest code, as some releases update only periodically and not instantly.

Or if you are using the wrapper, the nodes not being up to date, or wrong workflow where the image is not passed into the image_cond_latents.

Sorry actually upon re-reading the errors, if it's:

expected input[1, 16, 13, 68, 68] to have 32 channels

this should be workflow issue, the model is loaded so things should be up to date, but the image is not being passed correctly.

copying a param with shape torch.Size([3072, 32, 1, 2, 2]) from checkpoint, the shape in current model is torch.Size([3072, 16, 1, 2, 2]).

This one means ComfyUI or the wrapper nodes if you're using them are not up to date.

Problem solved.
The Image Noise Augmentation node was not up to date. Perhaps I had it bypassed in an earlier edited worklow (my bad).
But I still think the Desktop Comfy should not be used for this, not up to date enough.
Thank you tsolful and Kijai (Keep up with the good work Champ).

The latest Comfyui version has been updated to solve all issues