Last Frame Extender

A workflow that takes the last frame of a generated clip and uses it to seed the next one, chaining generations so a few seconds of video can be extended into a longer continuous shot.

The project called for walkthroughs of different locations and with the 15 second limitation of all video models at the time, I needed a way to generate multiple clips that could be stitched together seamlessly to produce 2-3 min videos. When people were in the image, they were mostly background so the workflow handled them on its own. The LLM reads the frame and writes the motion from it. For shots I needed to actually direct, I added an input that joins my own direction to the generated motion prompt.