Recently, I’ve been frequently coming across videos online that show something interesting: transforming people in photos into stylized avatars and automatically generating actions such as waving or dancing for the avatar. It’s genuinely fascinating. However, these videos are mostly created using closed-source commercial apps that are either limited or not free. To achieve full creative control over video generation, I built a comprehensive workflow in ComfyUI using entirely open-source models. All you need to do is input a photo of a person, choose your desired style, and specify an action. From there, you can locally generate unlimited videos of the stylized avatar performing the specified actions in sync with the original character—offering more flexibility and creative potential than those apps!

It’s the difference between giving someone fish and teaching them how to fish. In this video, I’ll first explain the thought process behind building the workflow, the implementation of each module, and any alternative approaches you might consider, then walk you through each step to achieve the final effect. Hope this helps you build your own workflows and improve together. Next, I’ll demonstrate how to use the complete workflow, ensuring you can get started right away—ready to use straight out of the box.

Workflow：https://drive.google.com/file/d/1ls-0LorP0W5hNmBwqVoofDu2c2OaPjYj/view?usp=sharing

CustomNodes：
ComfyUI_LayerStyle: https://github.com/chflame163/ComfyUI_LayerStyle
ComfyUI-RMBG: https://github.com/1038lab/ComfyUI-RMBG

Models：
Kontext：https://huggingface.co/Comfy-Org/flux1-kontext-dev_ComfyUI/tree/main/split_files/diffusion_models、
Wan2.1: https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/diffusion_models
FusionX LoRA：https://huggingface.co/vrgamedevgirl84/Wan14BT2VFusioniX/tree/main/FusionX_LoRa
FusionX VACE：https://huggingface.co/QuantStack/Wan2.1_T2V_14B_FusionX_VACE-GGUF/tree/main