When an existing LTXV latent already contains video and audio, connect it to video_latent on LTXVConcatAVLatent and connect the replacement to audio_latent. The node’s documented behavior is to retain the input’s video stream and replace its audio stream. Confirm the inputs’ types and inspect the resulting latent and noise_mask; an AV output label alone does not prove which video content was retained.
Set up the replacement
- Identify the existing joint AV latent and the new audio latent before wiring anything. Use the node’s actual type labels, not color alone, to tell latent inputs apart. Record a short identifier for each input so you can compare them after the operation.
- Add
LTXVConcatAVLatent. Connect the existing AV latent tovideo_latent, not toaudio_latent. Connect the replacement audio latent toaudio_latent. The input name matters: the documented special case is triggered by a video input that already contains both streams. - Check that the new audio can be fitted to the video latent. The node adjusts audio length by trimming longer audio or padding shorter audio with zeros. It reports an error if the latents cannot be fitted, including when they differ across more than one dimension or have incompatible batch or channel dimensions. Do not infer compatibility from matching filenames or an apparent duration alone.
- Route the
latentoutput to the intended downstream AV workflow. Record whether each input has anoise_mask, then inspect the output mask as part of acceptance. A combined AV latent is the documented output format; the mask may also be combined or supplied with all-ones values where an input mask is absent.
Use a before-and-after stream worksheet
The table is a hypothetical example of an author-recommended comparison, not a node execution or guaranteed output. The input descriptions and observations are illustrative. Record your own identifiers and mask observations from the workflow you are reviewing.
| Check | Hypothetical before | Hypothetical after | Decision |
|---|---|---|---|
| Video identity | Existing AV input: scene-A | Output video compared with scene-A | Pass only if the intended video stream is retained |
| Audio identity | Existing AV input: audio-old; replacement: voice-new | Output audio compared with voice-new | Pass only if the replacement is the intended stream |
| Output type | Input is a joint AV latent | Output type label is LATENT; stream content is checked separately | A generic type label does not establish stream identity |
| Noise mask | Existing input has a mask; replacement mask status is recorded | Output mask presence and observed content are recorded | Note differences; do not assume the mask is unchanged |
| Fit result | Audio latent is shorter than the video latent | Workflow completes; padded area is checked for the intended generation behavior | Treat length handling as documented behavior, not proof of acceptable sound |
For a practical identity check, compare the known input and output stream descriptions or inspect them with the tools available in your ComfyUI workflow. This comparison is an author-recommended acceptance method; the node documentation does not specify a built-in identity report or a particular inspection tool.
Limits and recovery
The node replaces the audio stream supplied to it; it does not establish that the new audio is semantically appropriate, synchronized to an intended event, or intelligible. Listen to the resulting media through the downstream workflow when those qualities matter. The documentation describes trimming, zero-padding, fitting errors, and combined mask behavior, but does not promise preservation of an audio waveform beyond those latent-level operations.
If the node reports a fit error, first compare the input latent shapes and their batch and channel dimensions. If the workflow completes but the output video identity is unclear, trace the video_latent connection and inspect the result rather than concluding that preservation occurred from the output port alone. If the mask differs, record which input supplied a mask and treat the resulting mask as a separate review item; do not silently assume it was copied unchanged.
The official LTXVConcatAVLatent documentation defines the already-packed-input behavior, fitting rules, and output. Its input and noise-mask details are especially relevant when a short audio latent or a missing input mask affects your review.
Does the node keep the video if `video_latent` is already AV?
According to the documentation, yes: it keeps that input’s video stream and replaces its audio with the supplied audio latent.
What happens when the new audio is shorter?
The node zero-pads the shorter audio latent to fit. Its documentation says the padded region is left unmasked so the model can generate there; review the downstream result rather than assuming the padded portion will contain a particular sound.
Does a combined output mean the original mask was preserved?
No. The output may contain a combined mask, and a missing input mask is replaced with an all-ones mask matching its samples. Inspect and record the mask separately.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.