You don't need a local image-to-video pipeline to test two stills

I have lost more evenings to “just testing a pair of frames” than I have to actually publishing a clip. The usual path looks serious: clone a ComfyUI graph, pin a model, fight a CUDA wheel, wait for a