AIBRIEF
The Redis creator's open source native port of the MiniMax H3 video model runs short clips locally on M3 Max and M5 Max, with a persistent interactive session for iterating on shots.
antirez (Salvatore Sanfilippo, creator of Redis) has released h3-metal, a native Metal port of the open-source MiniMax-H3 video model that runs short clips locally on Apple Silicon, without a cloud round trip. The port is H3-specific Metal work, built as a sequence of working vertical slices, not a fork of the model and not a general-purpose local video-AI runtime.
Current focus is performance and memory optimization on M3 Max and M5 Max. Portable Metal block parity, prompt encoding, prompt-to-video and audio, first/last-frame conditioning, and ordered Ref2VA image, video, and audio references all work end to end. The balanced preset generates 22 frames at 24 fps, about 0.92 seconds of video.
The load-bearing piece is the interactive session. Running ./h3 -d ./MiniMax-H3 --width 512 --height 512 --steps 6 keeps BF16 prompt conditioning, the prepared DiT, and the video decoder resident across prompts, so a developer iterating on a shot never re-loads or re-encodes. Session commands include !status, !seed, !seconds, !show, !save, !cache, plus !first/!last anchors and !ref-image (Ref2VA) with positional <Picture 1>, <Picture 2> naming.
An early Hacker News user on M5 Pro 64GB reports the same Q5_K_M GGUF path through ComfyUI takes over an hour for a nine-second 480x864 clip at 20 steps. On the M5 Max 128GB, that user recorded an image+audio render at 74.58 seconds. The next domino is not a longer preset. It is a sparse attention mode, hinted at in a MiniMax Reddit AMA, that antirez is now testing as an optional --sparse-attention flag.