Nodes/ComfyUI/LTX 2.5 Audio To Video
ComfyUI Node Runs on cloud

LTX 2.5 Audio To Video

Let the audio track drive the video

By Comfy-Org·Created 4 years ago·Updated about 5 hours ago· 128,055
LTX 2.5 Audio To Video
  • audio
  • image
  • VIDEO
model
prompt
seed42

Most video nodes start from words or pictures. This one starts from sound. Feed it an audio track and LTX 2.5 generates video that fits it - the clip's duration is literally set by the audio's length, and the model interprets the track as the scene's pulse. Music-video-style motion, a beat-synced abstract, a mood piece that matches a voiceover: this is the node for "the audio exists, make it visual".

It's the audio-to-video member of the LTX 2.5 API trio that landed in ComfyUI core in August 2026. Like its siblings it's a partner node - Lightricks' hosted model runs on their servers through Comfy's proxy, billed per second with Comfy account credits, and your local GPU only moves files around.

How it works

The audio is uploaded and the model generates footage conditioned on it, with the track's length (2 to 20 seconds - enforced) setting the output duration. You can anchor the scene with an optional first-frame image so the clip opens on something specific instead of letting the model improvise from scratch. A prompt steers the content on top of the audio's rhythm.

The model dropdown offers LTX-2.5 (Fast) and LTX-2.5 (Pro), each with a resolution choice (1920x1080 or 1080x1920 for audio-to-video). Pro is the higher-fidelity tier; Fast is the budget draft. The per-second price differs between them, shown as a badge on the node.

The inputs that matter

  • audio - the driving track. Its length sets the video's duration, so trim it to the clip you actually want before you pay for it.
  • model - Fast or Pro, plus the resolution inside the dropdown.
  • prompt - optional text guidance for what the video shows. It has real effect, so use it.
  • image - optional first frame, so the clip starts on a known subject.
  • seed - the familiar "force a re-run" control; actual results are nondeterministic regardless.

Output is a single VIDEO.

Gotchas

The 2-to-20-second audio window is the hard limit - a 30-second track errors out, so split long audio into chunks or extend it first. Audio-to-video also only offers the two 1080p resolutions, no vertical 9:16 unless you crop after. And while LTX's open-weights reputation is "fast but a tier below the big names," the hosted 2.5 tier is a genuine step up in quality - but it's still LTX, so give the prompt real content or you'll get ambient mush. When it works, though, this is the fastest road from "I have a track" to "I have a music video."

Categorypartner/video/LTXV

Inputs (5)

NameTypeDefaultDescription
audioAUDIOAudio track driving the video. Its length (2-20 seconds) sets the video duration.
modelCOMBO2 options: [object Object], [object Object]
promptSTRING
seedINT420–4294967295Seed to determine if node should re-run; actual results are nondeterministic regardless of seed.
imageoptIMAGEOptional first frame to be used for the video.

Outputs (1)

NameTypeDescription
VIDEOVIDEO