Player not loading? Watch on YouTube
This tutorial uses the Advanced Live Portrait wrapper for the LivePortrait model in ComfyUI to adjust facial expressions, head direction and gaze after generating an image. The speaker uses these controls to make characters appear to interact in a scene. The creator runs ComfyUI with Stable Diffusion on an RTX 5070 Ti. The expression controls belong to this ComfyUI wrapper.
The setup starts in ComfyUI Manager: search for expression, install Advanced Live Portrait, then restart ComfyUI. The demonstrated workflow builds on an earlier image-to-image tutorial and uses Efficiency Nodes. An existing image can also go straight from a Load Image node into the expression editor.
The wiring correction is useful: the image to edit belongs in src image. The sample image input supplies a reference expression; connecting only that input leaves the editor without a source image and produces the error shown in the tutorial. Image sharpening is optional.
The speaker demonstrates combining controls, but warns that high values can distort some source images. For output that loses face quality, he adds ReActor's face restoration node. He describes its installation as complex and restoration as optional. Available choices shown include CodeFormer, GFPGAN 1.3 and 1.4, and GPEN BFR. He prefers the first two in his examples because GPEN changes the face too much. These are observations from his workflow, rather than a general benchmark.
