[Workflow] LTX 2.3 and LTXDirector for Video Generation

Step 4. Add audio or a motion clip (optional). Drop a voice line or music on the audio track, or a video whose movement you want copied on the motion track. Leave both empty and the model writes its own sound.

Step 5. Hit run and download. A first pass draws picture and sound together at half size, then it scales up and a short second pass adds detail. You get one video file with the audio inside it. Ready for: Premiere · DaVinci Resolve · After Effects · CapCut

First time? Leave every setting as-is. The defaults (5 seconds · 24fps · 8 steps then 4) are the right starting point for almost everyone.

Overview

This workflow builds a video on a timeline using LTX Video 2.3, an open-weight 22B audio-video model from Lightricks, with the LTXDirector nodes for shot control. Two things set it apart. First, it writes picture and synced sound in a single pass, so footsteps, waves, and voices land on the frames that show them, and the clip comes back with audio already inside it. Second, the timeline is read as one continuous generation, not separate clips stitched together, so you split it into beats, give each stretch its own prompt, and the action changes partway through without the shot breaking or the subject drifting. This build is the distilled model, tuned to finish in 8 steps. You drop in an image, write your prompts, and get the clip back in about 5 minutes. No setup, no nodes to wire.

Who it's for: filmmakers, ad creators, and video teams who want shot-by-shot control with native sound in ComfyUI without wiring a timeline pipeline from scratch. Not for: a fast one-off clip or a story that spans many hard cuts. It is a heavy render, and it builds one continuous take, not an edit.

Why Floyo

Floyo is the only ComfyUI platform built for teams in the browser.

Made for teams. Share run history, files, and models across your whole team. A teammate opens your exact run and picks up where you left off. No file handoffs, no version confusion.

No install, no setup. Every workflow and model is preloaded. Open it in your browser and run. Nothing to download, nothing to configure.

No local hardware. Workflows run on H100 NVL GPUs, so heavy models run fast without a card of your own. Your VRAM stops being the limit.

Open and closed models in one place. Floyo runs open-source workflows and API models side by side.

How to use (in your browser on Floyo)

Drop a start image on the timeline and write a global prompt for the look.

Write a segment prompt for the action, split the timeline for more beats, add audio if you want, then run.

Download the video with embedded sound, ready for any editor.

Expectations The models are preloaded, so there is nothing to download. Running it needs a free Floyo account. This is a heavy render: each clip takes about 5 minutes, because it generates picture and sound, then upscales and refines in a second pass. A few things decide the result. Well-lit stills with one clear subject and prompts that name both a camera move and a sound give the cleanest clips. Keep guidance at 1, since the distilled build is tuned for it and raising it washes out contrast and motion. Rewrite the segment prompt before touching any number when the motion is wrong. On licensing: LTX 2.3 ships under the LTX-2 Community License, free for research and for commercial use by companies under 10 million dollars in annual recurring revenue, with larger organisations needing a license from Lightricks.

Use Cases

Previz & Short Film. Block out a scene with sound before you shoot, and show a director how a beat plays instead of describing it.

Social Video with Native Audio. Turn a still into a vertical clip that already has ambience or a voice on it, ready to post without an edit pass.

Dialogue & Voice-Led Scenes. Drop a recorded voice line on the audio track and let the picture follow the performance.

Product & Ad Spots. Animate a product still into a short spot with room tone and motion, then hand the file to an editor.

What can LTX 2.3 and LTXDirector direct?

Text encoder · Unknown