Allow Wan21_HuMo extra_conds to pass concat_latent_image

[drozbay]

This is the same code change as this older PR. I am resubmitting after rebasing the code changes to a more recent version of ComfyUI.

To re-summarize the changes: This small change simply allows the concat_latent_image to pass through the extra_conds in the HuMo model code so that it can be used for I2V. This can be accomplished in base ComfyUI by chaining a WanImageToVideo node with a WanHuMoImageToVideo node.

Even though the HuMo model was not designed for direct I2V, it still works passably and there are many circumstances in which it is useful. Because the WanHuMoImageToVideo node doesn't have an input for a start image, only people who know what they are doing may stumble upon this capability. I have already developed several useful workflows that takes advantage of this technique.

Attached is a very simple HuMo I2V workflow that works only when this change is in place. I tested this WF on a fresh ComfyUI installation with no custom nodes and only this code change in place. It performs a simple I2V HuMo generation with a start image, and then performs one continuation with a second sampler using the last frame as the start image for the second generation. There is definitely a contrast artifact on the second generation and this is a known limitation of HuMo I2V.

ComfyUI_00023_.mp4

Workflow: droz_HuMoImageToVideo_I2VBaseComfyUI_v1.json

Start Image:

Audio source: https://commons.wikimedia.org/wiki/File:Found-von_Goethe.ogg

System Info
OS
win32
Python Version
3.13.9 (tags/v3.13.9:8183fa5, Oct 14 2025, 14:09:13) [MSC v.1944 64 bit (AMD64)]
Embedded Python
true
Pytorch Version
2.9.1+cu130
Arguments
ComfyUI\main.py --windows-standalone-build
RAM Total
61.66 GB
RAM Free
49.55 GB
Devices
Name
cuda:0 NVIDIA GeForce RTX 5090 : cudaMallocAsync
Type
cuda
VRAM Total
31.84 GB
VRAM Free
30.12 GB
Torch VRAM Total
0 B
Torch VRAM Free
0 B