# How to Prompt LTX 2.3 for Video

LTX 2.3 doesn't read a prompt the way most image and video models do: short, comma-separated prompts actively hurt it. It was trained to read a full paragraph, so the way you write decides how closely the motion follows your idea. This lesson covers four rules for writing LTX 2.3 prompts and puts them together in one example.


## What you'll learn

- Why LTX 2.3 needs a full paragraph instead of a comma-separated tag list
- Why every action goes in the present tense
- How long a prompt should be: 4 to 8 descriptive sentences, up to about 200 words
- How to replace abstract words like 'sad' or 'dynamic' with concrete motion

- Duration: 2 min
- Updated: 2026-09-18
- Page: https://workroom.everypixel.com/tutorials/prompt-ltx-video

Why LTX 2.3 needs a different prompt. LTX 2.3 is built on a Diffusion Transformer, a different architecture from most image and video models. In practice, it was trained to read a full paragraph rather than a tag list, so a short, comma-separated prompt doesn't give it enough to work with. Rule 1: one continuous, cinematic paragraph. Write the prompt as one flowing paragraph, not a list of tags. If it reads like a shot description in a screenplay, you're doing it right. Rule 2: every action in the present tense. Describe everything as if it's happening right now: "the camera dollies forward," not "the camera will move forward." Rule 3: use the full length. This one is counterintuitive. LTX's own guide caps a prompt at around 200 words, roughly 4 to 8 descriptive sentences. A short prompt still gives a noticeably worse result than one that uses that full range. A long prompt isn't a problem for LTX 2.3. A short one is. Rule 4: no abstractions. Skip abstract words entirely. Not "sad", but "her eyes well up, her shoulders drop." Not "dynamic": say exactly what's moving, and how. All four rules in one prompt. Here's what the rules look like together. It's a lot of text, and that's the point: the motion follows the paragraph almost beat for beat. The camera opens close on the object resting on a wet dark rock at the edge of the sea, morning light spilling low across the surface. Water beads cling to its side, catching the light as they gather weight and begin sliding down in thin uneven trails. The camera holds steady for a beat, then dollies slowly forward, tightening the frame until the surface fills most of the shot and every ridge and drop of moisture is visible. Behind it, a wave rises in the middle distance, curls, and breaks against the rock, sending a fine mist upward that drifts across the frame and briefly softens the light. As the mist clears, the camera begins a slow pull back, widening the shot to reveal the full scene again — the object steady in the foreground, the water receding and gathering for the next wave, gulls crossing far in the background sky. The light shifts almost imperceptibly warmer as the shot continues, the shadow beneath the object lengthening slightly. In the final seconds, the camera settles into a static hold, the water continuing its rhythm behind an object that has not moved once throughout the entire sequence. Generate and adjust. Click Generate and review the result. If something doesn't match what you meant, change only that part of the prompt and regenerate. There's no need to rewrite the whole paragraph. On the Starter and Pro plans, LTX 2.3 runs unlimited, so you can refine the paragraph and regenerate as often as you need. See plans and pricing.
