Give the video model a character, a scene and a sound to work from.
MiniMax H3 Reference-to-Video + Singularity are now on NanoGPT: guide clips with image/video/audio references, or animate a starting image.
3-15 seconds, with native stereo audio.
Reference-to-Video:
https://nano-gpt.com/media?mode=video&model=minimax-h3%2Freference-to-video
Singularity:
https://nano-gpt.com/media?mode=video&model=minimax-h3-singularity%2Fimage-to-video