For an LTXV audio round trip, encode an AUDIO input with an Audio VAE, then decode its LATENT output with an Audio VAE. Record the actual VAE selections at both ends, the decoded AUDIO sample rate, and waveform duration as acceptance evidence. The ComfyUI node documentation describes each node’s Audio VAE input but does not require the encoder and decoder to use the same configuration. Treat configuration differences as a diagnostic factor, or as an acceptance condition only when the project specifies one. A successful node run alone does not establish that the result meets your delivery needs.
Trace the audio path and decide what must match
ComfyUI documents LTXVAudioVAEEncode as accepting AUDIO and an audio_vae, and returning LATENT containing samples, the VAE model’s sample rate, and a type identifier. LTXVAudioVAEDecode accepts LATENT and an audio_vae, then returns an AUDIO waveform with an associated sample rate determined by that VAE. The two nodes therefore have distinct input and output data types; a latent is not a waveform. See the encode node documentation and decode node documentation.
Use this ordered check when an audio round trip is meant to preserve a defined VAE contract:
- Identify the source
AUDIOand note its stated sample rate and duration from the available media information. - Record the Audio VAE connected to the encode node. Do not identify it by node color alone: inspect the actual type label and model selection.
- Follow the encode output labeled
LATENTinto the decode node. Record the decoder’s selected Audio VAE alongside the encoder’s. Compare them if the project requires a consistent configuration; otherwise, keep the observed selections as trace information rather than treating a difference as an automatic failure. - Inspect the decoded
AUDIOoutput’s reported sample rate and measure its duration using the available audio inspection method. Record any resampling step separately, including where it occurs and the rate before and after it. - Compare the result with the project’s intended audio requirements. If the project requires matching VAE selections and they differ, or if the output rate or duration is unexpected, investigate before acceptance. A VAE-selection difference may be worth checking when output is unexpected, but the node documentation does not make it an automatic failure.
The documentation specifies the node types and the role of the VAE in the output sample rate. It does not define a universal duration tolerance, a waveform-quality score, or a guarantee that matching VAEs preserve every audible detail. Those acceptance thresholds must come from the project brief.
Use a trace sheet to make the handoff reviewable
The following is a hypothetical inspection record, not a measured ComfyUI result. Suppose a creator starts with a 48 kHz, 12.0-second narration, encodes and decodes it with the same named Audio VAE, and sees a decoded output reported at 48 kHz with a measured duration of 12.0 seconds. That observation supports the recorded configuration and timing check; it does not by itself prove perceptual equivalence or delivery quality.
| Check | Hypothetical observation | Decision |
|---|---|---|
Input AUDIO | Narration, 48 kHz, 12.0 s | Source reference recorded |
Encode audio_vae | Model selection A | Record the visible model/type label |
| Encode output | LATENT, type identifier present | Confirm the connected output is latent, not audio |
Decode audio_vae | Model selection A | Configuration matches the encoder |
Decoded AUDIO | 48 kHz, 12.0 s | Rate and duration meet this hypothetical brief |
| Resampling | None recorded | Do not infer that no resampling occurred unless the workflow confirms it |
For your own trace sheet, include the input, both VAE selections, latent type, decoded rate, measured duration, and resampling status. If the output rate differs from the intended delivery rate, distinguish the VAE-produced rate from any later resampled rate rather than replacing one value with the other. A side-by-side re-decode using a different VAE selection can be an author-suggested diagnostic comparison when the workflow permits it; it is not an official ComfyUI test or a documented requirement.
Interpret mismatches and know the limits
If the decoded rate or duration fails the project requirement, check the latent connection and any downstream resampling or timing operation; you may also investigate whether the recorded VAE selections are relevant to the issue. Change a selection and decode again only when that is part of the project’s requirements or a deliberate diagnostic test. The decode documentation notes that a nested latent input is handled by using the last latent in its sequence; if your workflow passes nested latents, verify that the intended latent is the one being decoded.
Matching VAE selections and a matching sample-rate label do not guarantee clean sound, correct channel layout, or a faithful reconstruction. The supplied node documentation does not specify those guarantees or prescribe a resampling method. Treat resampling as a separate processing stage: log its input and output rates, then inspect the resulting audio under the project’s own criteria. If the waveform sounds wrong despite a passing metadata check, investigate the audio path and processing rather than assuming the VAE pair is sufficient.
Does node success prove the audio is acceptable?
No. It indicates the node executed, not that its model pairing, timing, sound, or delivery rate satisfies your brief.
Should I resample just because the source and decoded rates differ?
Only when the project or receiving workflow requires a particular rate. Record the conversion as a separate step and inspect its result.
What if the latent is nested?
The decode documentation says the node uses the last latent in a nested sequence. Confirm that this is the intended item before judging the output.
Sources and next step
- LTXVAudioVAEEncode - ComfyUI Built-in Node Documentation
- LTXVAudioVAEDecode - ComfyUI Built-in Node Documentation
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.