MiniMax H3 workflow with StoryBoard and Official/Third-party IR Refiner — v1.0 — v1.2(nodepack 1.3.0+)

Fix AdaLN pre-bake and LoRA compatibility, including precision differences that caused pre-bake cache misses and slower sampling.

Speed up selected paths: positive/negative text encoding now shares streamed groups, disk reads are concurrent, and quantized encoder weights are prefetched.

Rework the ring-buffer PIN logic: keep the contiguous pinned home pool, remove RAM-to-RAM staging copies, and fall back cleanly to pageable transfers when cudaHostRegister fails.

Improve the Storyboard frontend UI.

Add a negative prompt input to Simple Prompt and shared text encoding for positive/negative conditioning.