Load Pose Guider

The Load Pose Guider node is a one-input loader that does exactly one job: load the pose_guider.pth file so the rest of the AnimateAnyone pipeline can see what a pose even looks like. If you're following the example workflow, it's one of the first nodes in the graph and one of the easiest to get right - and to get wrong, because it depends on a model file you have to fetch by hand.

What the pose guider actually is

In the AnimateAnyone architecture, the pose guider is the lightweight conditioning network - roughly the ControlNet analog in this pipeline. It's a small convolutional encoder that takes your OpenPose skeleton frames and projects them into feature space the denoising UNet can be guided by. In this implementation it's a PoseGuider(320, block_out_channels=(16, 32, 96, 256)) network, which is small beans compared to the UNets it feeds. That's deliberate: the whole point of AnimateAnyone's design was that pose conditioning shouldn't cost you a full ControlNet's worth of weights and VRAM.

The file it loads comes from the Moore-AnimateAnyone pretrained weights, hosted on Hugging Face as patrolli/AnimateAnyone. You download pose_guider.pth from there and drop it into the pack's pretrained_weights/ folder.

Inputs and output

There's a single input, pose_guider_model_path, a text field that defaults to ./pretrained_weights/pose_guider.pth relative to the pack's install folder. Relative to what, exactly, trips people up: that default only works if the file sits inside ComfyUI/custom_nodes/ComfyUI-AnimateAnyone-Evolved/pretrained_weights/. If you've put the model somewhere else, set the full path here.

The single output is pose_guider (type POSE_GUIDER), which you wire straight into the Pose Guider Encode node - not into the sampler. Load Pose Guider → Pose Guider Encode → Sampler is the chain.

Installing it

The node is part of the ComfyUI-AnimateAnyone-Evolved pack, so install the pack and the model together:

cd ComfyUI/custom_nodes
git clone https://github.com/MrForExample/ComfyUI-AnimateAnyone-Evolved
pip install -r requirements.txt

Then grab pose_guider.pth from patrolli/AnimateAnyone on Hugging Face and place it at ComfyUI-AnimateAnyone-Evolved/pretrained_weights/pose_guider.pth. ComfyUI Manager can install the pack itself (search "AnimateAnyone Evolved"), but it won't download this model for you - none of the pack's four .pth files are auto-downloaded, and forgetting this one is the most common reason the Pose Guider Encode node fails with a missing-file error.

Gotchas

Watch the console on first load. If you see an import failure on the [AnimateAnyone] nodes, the pack's dependencies aren't in your environment - run the pip install -r requirements.txt step, and if diffusers is the culprit, pip install --force-reinstall diffusers>=0.26.1 per the README. And remember this is a 2024-era pack: it works fine, but it expects the old file layout, so if your ComfyUI version moved model folders around, set the absolute path in this node and you're set.