MiniMax H3 Audio Lock (best-effort, unverified):
MiniMaxH3AudioLock places a reference sound into the audio portion of a MiniMax H3 audio-video latent, which is an intermediate representation used before the final render. It first encodes the reference sound, then mixes that encoded sound with the audio already present. You can use it to influence the entire clip with a repeating reference or to influence only the beginning. The video portion is passed through unchanged. This is a best-effort implementation whose behavior has not been verified against an original public version of the node, so listen to the rendered result before relying on it.
MiniMax H3 Audio Lock (best-effort, unverified) Input Parameters:
av_latent
A required LATENT containing the MiniMax H3 video and audio streams you want to process. The node changes its audio stream while retaining its video stream. It requires a MiniMax H3 audio-video latent in the expected format; an ordinary latent or one with incompatible dimensions produces an error. No default value is defined.
audio_vae
A required VAE used to encode the reference audio into the form needed for mixing with the latent audio stream. If the reference clip’s sample rate differs from the VAE’s audio sample rate, the node resamples the clip before encoding it. Use the audio VAE intended for your MiniMax H3 workflow. No default value is defined.
mode
A required choice with the options lock and blend; the default is lock. In lock mode, the encoded reference repeats as needed to cover the full target audio duration. In blend mode, it affects only the beginning, up to the shorter of the reference and target durations; later audio remains unchanged. In either mode, strength determines how much of the fitted reference replaces the existing audio.
strength
A required FLOAT controlling the mix between the existing latent audio and the fitted reference. The minimum is 0.0, the maximum is 1.0, the default is 0.35, and the adjustment step is 0.01. At 0.0, the audio values remain unchanged; at 1.0, the affected portion uses the encoded reference instead of the existing audio. Values between those limits combine the two. In blend mode, audio after the reference-covered opening remains unchanged regardless of strength.
audio_1
An optional AUDIO reference clip; no default value is defined. When connected, it takes priority over audio supplied by references. The node uses the first waveform in the supplied audio batch, resampling it if needed before encoding. Although this socket is optional, you must provide reference audio through either audio_1 or a non-empty references audio list.
references
An optional H3_REFS bundle; no default value is defined. If audio_1 is not connected, the node uses the first item in the bundle’s audios list as its reference clip. This lets you use a compatible references bundle without a separate reference-splitting step. Other entries in the bundle are not used by this node.
MiniMax H3 Audio Lock (best-effort, unverified) Output Parameters:
av_latent
A LATENT containing the original video stream paired with the processed audio stream. Connect this to the next compatible step in your MiniMax H3 workflow. The returned latent contains the samples data; the source code does not carry through any other fields that may have been present in the input latent.
audio
An AUDIO output containing the reference clip selected from audio_1 or, when that is absent, from references. This is the selected input clip, not a decoded preview of the processed latent audio. Use it when another node also needs that reference.
report
A STRING summary stating the selected mode, strength, encoded reference length, and target audio length in latent time units. It also warns that the original node behavior is unverified. The reported lengths help you check whether a short reference will repeat in lock mode or affect only the opening in blend mode; they are not audio durations in seconds.
MiniMax H3 Audio Lock (best-effort, unverified) Usage Tips:
- Start with the default
strengthof0.35, render a short test, and adjust by listening. Increase it for more reference influence or reduce it to retain more of the existing audio. - Choose
lockwhen you want the reference to influence the whole clip. A shorter reference repeats to fill the target duration, so listen for audible repetition in the render. - Choose
blendwhen you want to influence the opening while leaving the remainder of the latent audio as it was. A short reference affects only a short opening. - Connect
audio_1when you need to choose a specific clip directly. If you usereferencesinstead, check that its firstaudiosentry is the clip you intend. - Keep the original
av_latentavailable while comparing renders: the node changes its audio stream, and its best-effort behavior should be judged from the final sound rather than the report alone.
MiniMax H3 Audio Lock (best-effort, unverified) Common Errors and Solutions:
MiniMax H3 Audio Lock (best-effort, unverified): no reference audio was supplied -- wire a clip into audio_1, or connect a references bundle whose audios[] is non-empty.
- Explanation: Neither
audio_1nor the first entry ofreferences.audiosprovided a clip. - Solution: Connect an
AUDIOclip toaudio_1, or connect anH3_REFSbundle with at least one audio item. If using a reference splitter, connect itsaudio_1output rather than an emptyvideo_audio_1output.
MiniMax H3 Audio Lock (best-effort, unverified) expects a MiniMax H3 AV latent (video+audio NestedTensor)
- Explanation: The supplied
av_latentdoes not contain the two-part video-and-audio structure this node expects. - Solution: Connect a MiniMax H3 audio-video latent from the appropriate workflow stage, rather than a video-only or ordinary latent.
MiniMax H3 Audio Lock (best-effort, unverified): av_latent doesn't look like a MiniMax H3 AV latent
- Explanation: The latent has video or audio dimensions that do not match the format checked by this node.
- Solution: Check that
av_latentcomes from a compatible MiniMax H3 workflow and has not been replaced or reshaped by an incompatible node.
Missing required input: audio_vae or av_latent
- Explanation: A required connection is absent, so the node cannot encode the reference or process the target latent. The exact interface message may vary by ComfyUI build.
- Solution: Connect both a compatible audio VAE and a MiniMax H3 audio-video latent, then run the workflow again.
An audio encoding or tensor shape error appears during execution
- Explanation: The supplied audio, VAE, or latent may be incompatible even if the sockets connect. The source code does not define one specific error message for these underlying failures.
- Solution: Verify that the reference is a valid
AUDIOclip, the VAE is the one intended for this workflow, and the target is a compatible MiniMax H3 audio-video latent. Test those connections with a known-working clip and latent to isolate the mismatch.
