Player not loading? Watch on YouTube
This tutorial explains a ComfyUI workflow that animates a character image using motion from a reference video. It uses LTX 2.3 with IC LoRA Union Control, which the speaker describes as supporting Canny, depth and pose control maps. Setup includes downloading the LoRA into ComfyUI's LoRA folder and installing the workflow's custom node packs through ComfyUI Manager.
The speaker first adjusts the character image to match the reference video's opening pose using an image-editing model. A node extracts the first frame, and Open Pose creates its control map. The speaker reports better results with this preparation, though the workflow also has a switch to disable it.
The initial example uses LoRA strength 0.8, an output length of 81 frames, 720 × 1280 dimensions and 24 fps. Depth Anything V2 supplies the video control map. A separate guide strength controls how strongly that map influences generation, and the workflow matches the reference video's orientation.
The examples also show limitations. Some movements differ from the reference; a second character has hand artifacts and a changed face. The speaker lowers LoRA strength to 0.5, reruns the image edit and switches to DW pose preprocessing, then judges the result better. Pose transfer remains imperfect, and the speaker cautions that image pose editing can fail in some cases.
