FULL STORY

Phosphene Enables Local MiniMax Video Models on Mac

Phosphene rolled out continuous updates to run MiniMax H3 video generation locally on Apple Silicon Macs via the MLX framework, significantly lowering hardware barriers.

2026-08-04 ~ 2026-08-08 · 2 episodes · 7 posts

Episode 1 · Phosphene 3.4.0 Brings MiniMax H3 Video Generation to Mac Locally (2026-08-04, 5 posts)

Phosphene 3.4.0 has been released, with developer @cocktailpeanut enabling local MiniMax H3 video generation on Apple Silicon Macs via the MLX framework, eliminating the need for cloud or API keys. This marks a rapid community effort to localize MiniMax H3 after its open-source release, offering a new offline option for high-end video generation.

Confirmed

  • Version and model: Phosphene 3.4.0 adds MiniMax H3 as a second local video generation engine, alongside the existing LTX-2.3.
  • Hardware requirement: Running MiniMax H3 locally requires a Mac with 64GB RAM.
  • Features: Supports joint dialogue and sound generation in both text and image modes.
  • Generation speed: According to the workbench log, generating a multimodal AI character video takes approximately 36 minutes.

Why it matters

This update breaks the absolute dependence on cloud computing for high-end video generation. Developer @cocktailpeanut, using the MLX framework, has brought the complete multimodal generation workflow, including sound and visuals, to the local environment, providing a viable solution for creators who prioritize privacy and offline workflows.

Episode 2 · Phosphene Update Enables Local Hailuo H3 Video Generation on Mac (2026-08-08, 2 posts)

Apple Silicon-native AI tool Phosphene updated to v3.6.0, enabling local Hailuo H3 video generation using Turbo mode. Tests show an M4 Max chip can generate a 5-second video with synchronized audio in just 8 minutes.