Made with the zeroscope model at 576x320, rendered 2-3 second clips and spliced them together with some audio. The timing is a bit rough, I threw it together the other night after generating a bunch of footage that came out pretty creepy and I'm still getting my editing chops together. Prompts like "horror movie footage", "nightmarish creature running towards viewer" produce some great results.

The model is here: https://huggingface.co/cerspense/zeroscope_v2_576w

I tried to upscale it with the XL model, but I keep getting out-of-memory errors and haven't found a working solution yet.