ComfyUI-H3-Motion-Context

Why we included this project

Chaining MiniMax H3 clips in ComfyUI usually means accepting a visible cut: each new clip starts fresh. This pack carries motion, speed, direction, and audio across the join instead. It slices the previous clip's tail straight out of its latent rather than decoding to pixels and re-encoding, so the pinned frames stay bit-for-bit identical and you avoid the color drift and softening that build up over a long chain. Audio gets the same treatment: the pinned window reaches backwards into sound that already played, so the model continues your track rather than writing something that merely sounds like it. Alongside that are a Trim node, a Save/Load Latent pair, and a Seam Probe that measures whether a join is a real continuation or a convincing imitation. For multi-shot video, music videos, or any long-form H3 generation, that's the difference between a cut you notice and one you don't.