antirez is writing a MiniMax-H3 video and audio inference engine in C and Metal for Apple Silicon, 2,363 stars in 14 days

GitHub·medium signalXBlueskyLinkedInCopy link

h3.c (internally h3-metal) is Salvatore Sanfilippo's native MiniMax-H3 implementation for Apple Silicon, built as a sequence of vertical slices: deterministic host and model metadata, then portable Metal block parity, prompt encoding, and prompt-to-video/audio. Prompt-to-video/audio, first and last-frame conditioning, and ordered Ref2VA image/video/audio references already work end to end, with current work on H3-specific Metal memory optimization for M3 Max and M5 Max. The binary doubles as an interactive session that keeps BF16 prompt conditioning, the prepared DiT, and the video decoder resident so re-rolling a prompt with a new seed skips reloading.