LTX-2.3 Fast Text to Video: Prompts to Clips Up to 4K

Last verified: October 19, 2026

LTX-2.3 Fast renders portrait and landscape video at up to 4K and cinematic frame rates, and speed stays its priority. Working in LTX-2.3 Fast text-to-video comes down to three choices: the shot you describe in the Prompt box, the Resolution (4K on a short clip that has to hold up on a big screen) and a vertical or widescreen frame under Aspect Ratio.

You describe the scene, the camera motion and the mood, and the model returns a video with matching audio. Leave the Generate Audio switch on and write the sound into the prompt with the same care as the picture: rain on a window, a line of dialogue, a crowd in the distance. Switch it off when you only need the picture.

LTX-2.3 follows descriptions of timing, motion and expression closely. That makes the Fast tier a practical place to work out ideas, and Lightricks positions it for rapid iteration, storyboards and previews. Use it to test a concept, check framing and pacing, and build up a set of rough shots before you commit to a final cut.

Explore more text to video

Capability Snapshot

LTX-2.3 Fast at a Glance

What you put in, what you can set and what comes back from each text-to-video run.

Input

A text prompt of up to 2,048 characters

Video Duration

6, 8, 10, 16 or 20 seconds

Resolution

1080P, 2K or 4K, depending on the duration

Aspect Ratio

9:16, 16:9

Audio

Switch on or off, on by default

Output

1 video per run

How Do You Prompt LTX-2.3 Fast for Better Clips?

Write One Flowing Shot Description

Keep the prompt to one flowing paragraph of under 200 words. Open on the action, then cover movement, appearance, environment, camera, lighting and any sudden change. This beats a string of keywords because with LTX-2.3, how you phrase the prompt matters as much as which things you name. A paragraph that length fits comfortably in the Prompt box. If all you have is a rough idea, run Enhance Prompt first, then rework the suggestion you pick into that order and add the details only you know.

Spell Out Who Stands Where

LTX-2.3 respects spatial layout, so state positions outright: left or right, foreground or background, facing toward or away from the camera, and how far apart the subjects are. It reads multiple subjects, spatial relationships and detailed actions accurately when they are spelled out, and simplifying a scene no longer helps. This pays off most in scenes with two or more people, where a loose description lets the characters drift into the wrong part of the frame.

Name the Camera Move in Film Terms

Explicit camera instructions like a slow dolly in or handheld tracking are one of the things LTX-2.3 handles well. Terms such as tracks, pans across, circles around, tilts upward, over-the-shoulder or static frame give the model a precise target. When the shot needs unbroken camera motion or a performance held in one framing, describe it as a single continuous take.

Making a Clip with LTX-2.3 Fast

Four steps from a written idea to a downloaded clip.

LTX-2.3 Fast or LTX-2.3 Pro: Which Should You Use?

How Lightricks' Fast and Pro tiers of LTX-2.3 divide the work, according to the creator's own specs.

What it is tuned for It is built for high-speed generation and aimed at rapid iteration, storyboards and previews. LTX-2.3 Pro is the higher-fidelity tier, documented for higher detail and motion stability in final, commercial-grade renders. Work in LTX-2.3 Fast while you are still finding the shot, then render the version that ships with LTX-2.3 Pro when detail and steady motion matter most.
Longest single clip Its longest clips come at 1080p, twice the length Pro allows, while its 1440p and 4K output stops at ten seconds. LTX-2.3 Pro tops out at ten seconds at every resolution it offers. Pick LTX-2.3 Fast for any shot that has to run longer than ten seconds in one generation; at ten seconds or under, length is no reason to choose one over the other.
Resolution and frame-rate ceiling It is documented at up to 4K and up to 50 fps, in both portrait and landscape. LTX-2.3 Pro is documented with the same 4K and 50 fps ceiling in both orientations. Both tiers reach the same ceiling, so choose between them on clip length and finish quality, not on resolution.
Start and iterate on LTX-2.3 Fast, and stay with it for single shots longer than ten seconds. Move to LTX-2.3 Pro for final renders where detail and motion stability decide the result.

Fixing Common LTX-2.3 Fast Problems

What usually goes wrong with a clip made from text alone, and what to change before the next run.

The action feels rushed in a longer clip

A prompt that is short for the clip length leaves the model without enough direction to fill the time. Add beats: what happens first, what changes midway and how the shot ends. Or choose a shorter Video Duration that matches the amount of action you've written. Then run it again.

The video ignores part of my prompt

LTX-2.3 doesn't always match a prompt perfectly, and how well it follows depends heavily on how the prompt is written. Contradictions, such as a still lake with crashing waves, confuse it further. Remove any directions that pull against each other, rewrite keyword lists as plain descriptive sentences and generate again. Each run gives you one new take, so a second attempt with the cleaned-up prompt is a fair test.

I can't pick 2K or 4K for a long clip

The higher Resolution options only appear when Video Duration is set to one of the shorter lengths. Shorten the clip to ten seconds or less to unlock them, or keep the long clip at the lowest Resolution option. Nothing else needs to change before you run it.

The background sound is thin when nobody speaks

Audio generated without speech can come out at lower quality. Describe the acoustic space and name the ambient sounds you want, for example rain on a tin roof in a narrow alley or footsteps echoing in a station hall, then run it again. If the ambience still isn't usable, turn Generate Audio off and add your own sound in the edit.

Frequently Asked Questions

Does LTX-2.3 Fast generate audio with the video?

Yes. LTX-2.3 generates synchronized video and audio within a single model, and the Generate Audio switch decides whether your clip includes sound. Turn it off when you only need the picture, for example when the clip will be cut to your own music track.

Will Enhance Prompt overwrite what I typed?

No. Enhance Prompt suggests fuller prompts based on your rough idea, and the text in the Prompt box only changes when you pick one of the suggestions. Use it to turn a one-line concept into a first draft, then add what only you know, such as exact dialogue or the specific camera move.

What makes one generation cost more than another?

The credit cost of a run is worked out from the options you choose and shown above the Generate button once you're signed in. Video Duration and Resolution are the settings that change it, so you'll see the figure move as you adjust the clip length or resolution before you generate. Enhance Prompt suggestions don't use credits.

How do I write dialogue for a speaking character?

Break long lines into short phrases with acting directions between them. Direct the performance with physical cues, like a glance down, a pause or a hand tightening on a cup, rather than emotional labels. Describe the character's voice and the acoustic environment too, and keep Generate Audio on so the lines are spoken in the soundtrack.

References

Sources and citations used to support the content provided above.

Updated: 2026-10-19 23:25:19 6 Sources

docs.ltx.io

Source Link
https://docs.ltx.io/models/ltx-2-3

docs.ltx.io

Source Link
https://docs.ltx.io/api-documentation/api-reference/video-generation/text-to-video

ltx.io

Source Link
https://ltx.io/model/ltx-2-3

ltx.io

Source Link
https://ltx.io/blog/ltx-2-3-portrait-video

ltx.io

Source Link
https://ltx.io/blog/ltx-2-3-prompt-guide

huggingface.co

Source Link
https://huggingface.co/Lightricks/LTX-2.3