The large model can run on consumer GPUs and I'd argue is more relevant to the hobbyist use cases because I'm not working out the efficiency of running thousand

The large model can run on consumer GPUs and I'd argue is more relevant to the hobbyist use cases because I'm not working out the efficiency of running thousands of users - just my own videos.