沐阳 H3 · 动作音频自动分段（内部）

Every time the Director slices its action reference video into per-segment clips, it also has to slice the soundtrack - and the two cuts have to land at the same moment in time, or the motion transfer inherits a permanent offset between picture and sound. 沐阳 H3 · 动作音频自动分段（内部） (H3ReferenceAudioClip) is that audio twin of 沐阳 H3 · 动作视频自动分段: given the same start_frame and frame_count you'd give the video slicer, it cuts the audio to match.

The mechanism is simple and correct: it converts the frame range into samples using the frame rate - start_frame * sample_rate / fps gives the sample offset, frame_count * sample_rate / fps gives the length - then slices the waveform. fps defaults to 24, matching the pack's video timeline. Like its video counterpart, it pads if the requested chunk runs past the end of the audio, so a tail segment doesn't come back shorter than expected.

Why does it exist as a separate node rather than riding along inside the video slicer? Because in this pack audio and video flow through different sockets and different nodes - the reference video is an IMAGE batch, the soundtrack is an AUDIO dict with waveform and sample rate - and the Director's graph handles them as parallel lanes that only get recombined at the conditioning step. Keeping the slicing explicit in each lane makes the sync contract visible: same start frame, same count, same fps, and the two come out aligned.

Inputs

• audio - the soundtrack as an AUDIO dict. It requires a real waveform and sample rate, and errors clearly if you hand it something else.
• start_frame / frame_count - the same frame range you gave the video slicer; frame_count defaults to 125.
• fps - 24.0 default; set it to match the video's actual rate.

Output: 分段动作音频 (AUDIO), the synchronized slice.

Install and context

Standard pack install: ComfyUI Manager search "ComfyUI-MiniMaxH3-Myang", or git clone https://github.com/civilcoco/ComfyUI-MiniMaxH3-Myang into custom_nodes, restart. No extra Python deps.

This is an internal Director node, so you'll meet it inside expanded motion-transfer graphs rather than place it yourself. If a transferred segment's audio drifts from its picture, check that the audio slicer and video slicer are using the same start_frame, frame_count and fps - that's the whole sync contract. And remember H3's local weights carry a territory-restricted community license (US/EU/UK/South Korea excluded).