ComfyUI Node
MiniMax H3 Audio Reroll (Experimental) / 音频重生成（实验性）
Experimental: condition a fresh H3 sampling run on an already-generated video latent, so a new soundtrack can be produced without re-authoring the picture. The video is strongly preserved but resampled, not frozen -- use video_passthrough downstream for a bit-identical mux. / 以已生成的视频潜空间作为参考重新采样音频；画面会被强力保留但仍会重采样。如需逐位一致的画面，请在下游使用 video_passthrough 混流。
MiniMax H3 Audio Reroll (Experimental) / 音频重生成（实验性）
- conditioning
- av_latent
- positive
- latent
- video_passthrough
- info
◄video_cond_strength1.000►
◄reset_audio_latenttrue►
Categorymodel/conditioning/minimax
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| conditioning | CONDITIONING | Conditioning from an H3 conditioning node, with the soundscape/music text changed. / 来自 H3 条件节点的条件，其中环境音或音乐文字已经修改。 | |
| av_latent | LATENT | The AV latent from the run whose picture you want to keep. / 需要保留画面的原始运行所产生的音视频潜空间。 | |
| video_cond_strength | FLOAT | 1.0000.5–1 | Noise augmentation on the video reference. 1.0 feeds it perfectly clean for maximum preservation; lower values let the picture drift. The DiT keyframe default is 0.999. / 视频参考的噪声增强值。1.0 表示完全干净并最大程度保留画面；较低值会允许画面变化。DiT 的关键帧默认值为 0.999。 |
| reset_audio_latent | BOOLEAN | true | Start the audio stream from zeros so the sampler writes a genuinely new track. Off reuses the old audio latent and biases the result toward the original. / 从零开始音频流，让采样器生成全新音轨；关闭时复用旧音频潜空间，结果会偏向原音轨。 |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| positive | CONDITIONING | — |
| latent | LATENT | — |
| video_passthrough | LATENT | — |
| info | STRING | — |