This is a motion transfer workflow in ComfyUI using DiffusionGemma PromptBuilder nodes + LTX 2.3 to transplant movements from a source video onto a still image character. It allows movement while maintaining the appearance with concise instructions. DiffusionGemma NVFP4 automatically provides precise prompting at the node level, which allows you to achieve the desired result effortlessly. Essentially, you can just use a video and an image as a prompt. Optionally, you can add additional context - but it isn't necessary.While the image I used in the example was busier/noisier than it should be for this type of workflow, DiffusionGemma was still able to guide the generation nicely. You'll need the nodes for this workflow; you can get them here: <a target="_blank" rel="ugc" href="https://github.com/exportAnything/ComfyUI-DiffusionGemmaPromptBuilder">https://github.com/exportAnything/ComfyUI-DiffusionGemmaPromptBuilder</a>The nodes are still a WIP. I'll upload more workflows for different use-cases here in the coming days.

Files with SHA256: e801b9a4f519cac07168b9b3182710352bd086115ea24d84fb125cf7818d727c