LTX 2.5 Prompting Guide: Better AI Video on Mac (2026)

The best way to prompt LTX 2.5 on Mac is to write one present-tense paragraph that covers the shot, the scene, the action, the character, the camera move, and the sound. LTX 2.5 reads full sentences the way a film director reads a shot note. It does not read tag lists. A clear paragraph gives you steadier motion, better sound, and fewer wasted clips. The sections below show the formula, three original example prompts, and the mistakes that ruin most clips.

What Makes an LTX 2.5 Prompt Different?

An LTX 2.5 prompt is a short scene description written in plain prose. Lightricks built the model to make video and synchronized audio in one pass, so your words steer both the picture and the sound. A Gemma 4 12B text encoder reads the prompt, which is why full sentences with clear detail beat a pile of keywords.

Native multishot is the other big change. One prompt can now describe several connected shots that keep the same character, place, and light. Earlier LTX models made one continuous shot. Prompts also behave differently across local AI video generation tools, so a prompt that works in one tool may fail in another. Lekh AI Pro runs LTX 2.5 next to its other offline AI tools on a Mac.

The Six-Part LTX 2.5 Prompt Formula

A strong prompt covers six parts in one flowing paragraph. Each part answers a question the model would otherwise guess.

Shot and Scene

Name the shot size, such as wide, medium, or close-up. Then give one clear light source with a direction. "Warm window light from the left" tells the model where to put shadows. "Beautiful lighting" tells it nothing. Pick one light logic per shot, because mixed light sources muddy the result.

Action and Character

Open with a subject who is doing something. A subject with no clear movement often gives a nearly still clip. Use motion verbs and join steps with "then". Describe the person by age, hair, clothing, and one standout detail. Show feeling through the body. "Her shoulders drop" works. "She is sad" does not.

Camera Movement

Name one camera move per shot, such as a slow push-in, a pan, or a locked-off frame. Then say where the camera ends up. "The camera settles on her gloved hands" helps the model finish the move instead of dropping it halfway.

Audio

Name sounds that have a source in the frame, such as footsteps, wind, or a kettle. Say so when you want no music. If you skip the audio line, the model picks the soundtrack for you.

Here is a weak prompt and a strong prompt for the same idea.

Weak: beekeeper, meadow, sunrise, cinematic, 8k, masterpiece

Strong: A slow push-in on a beekeeper in a white veil lifting a wooden frame from a hive in a misty meadow at sunrise. Low golden light comes from the left and glows through the bees rising off the frame. She tilts the frame, brushes it once with a soft brush, and her shoulders relax as she slides it back. The camera settles on her gloved hands. Bees buzz softly, and a light breeze moves the grass; no music.

The strong version says what happens, where the camera sits, and what you hear. A chat model can turn rough notes into a paragraph like this, and Lekh AI runs many local AI models on a Mac for that job.

LTX 2.5 Multishot Prompts: How to Write Cuts

To write a multishot prompt, describe the whole scene as one chronological paragraph and name every cut in plain words. Keep it to two to four shots. A numbered shot list reads as a list, not as an edit, so prose with named cuts is the safest format.

Each cut needs four jobs done:

• Name the cut: Use words like "a hard cut moves to" or "a match cut connects".

• Reset the frame: State the shot size, the angle, and who is in view.

• Reuse the subject's look: Repeat "the man in the green wool coat" so identity carries over.

• Say what the sound does: It can continue, drop, or change.

An original three-shot example:

A wide shot shows a mountain train station at dawn, frost on the platform and one lamp glowing. A man in a green wool coat walks along the platform with a small suitcase, and wind hums over a distant bell. A hard cut moves to a close-up of the man in the green wool coat checking a pocket watch, his breath fogging, and the wind continues across the cut. Another hard cut shows a low-angle shot of a train lamp growing out of the mist as the bell rings louder.

LTX 2.5 Image-to-Video Prompts

For image-to-video, describe only what should move, because the picture already sets the look. Re-describing the photo can pull the result away from your image. Write three short parts: what the subject does, what the camera does, and what you hear. Give the motion an end point, such as "then holds still", so the clip has a place to land.

The cat on the windowsill stretches, turns toward the camera, and blinks slowly as the curtains stir in a light breeze. The camera pushes in gently and holds on its face. Soft purring and distant street sounds, no music.

Nothing in that prompt describes the cat, the window, or the room. The image already did that. Lekh AI Pro includes Flux image generation, which can make the starting still. Different image-to-video tools handle motion in their own ways, so keep the prompt simple whichever one you use.

Prompting Audio and Dialogue in LTX 2.5

Put spoken lines in quotation marks, name the language, and keep each line short. LTX 2.5 makes audio and video together, so the line, the room sound, and any music all come from your words. Short lines sound natural, and long lines get rushed.

A medium close-up of a market vendor in an apron holding up a warm loaf. He smiles and says in English, "Fresh bread, still warm." Crowd murmur and a paper bag rustling, no music.

Keep any on-screen text short, since readable text is still hard for video models.

Prompting LTX 2.5 on Mac: A Test-and-Refine Routine

On a Mac, the fastest route to a good clip is to start simple and change one thing at a time. Lekh AI Pro runs LTX 2.5 on Apple Silicon for text-to-video and image-to-video, with synchronized audio at up to 1024×768, and local workflows need no cloud account. The same routine works for LTX 2.5 on Mac with any prompt in this post.

1. Write the six-part paragraph with one action and one camera move.

2. Generate a first clip and judge the motion, the light, and the sound.

3. Change one part only, such as the camera move or the light direction.

4. Save the wording that works in a notes file.

5. Generate the clip you plan to keep from your best prompt.

Common LTX 2.5 Prompt Mistakes

Most weak LTX 2.5 clips come from six fixable habits.

LTX 2.5 Prompting FAQ

How long should an LTX 2.5 prompt be?

Long enough to cover the six parts, usually four to eight sentences for one shot. Add about one sentence per extra shot. Every sentence should add something you can see or hear.

Can LTX 2.5 write multiple shots from one prompt?

Yes. LTX 2.5 supports native multishot, so one prompt can describe two to four connected shots with named cuts. Earlier LTX models made one continuous shot.

Does LTX 2.5 use negative prompts?

Do not rely on them. Writing the result you want, such as "sharp focus on the cup", is more dependable than banning a flaw with "no blur".

How do you prompt dialogue in LTX 2.5?

Put the line in quotation marks, name the language, and keep it short enough to fit the clip. Add the room sound and say whether music plays.

Why does my LTX 2.5 clip look static?

The prompt usually has no clear action. Open with a subject doing something and use motion verbs.

Do these prompts work on a Mac?

Yes. The format is the same on any device. Lekh AI is a local AI app for Mac that runs LTX 2.5 in Pro, so you can test every example here on Apple Silicon.

Write Better LTX 2.5 Prompts on Mac with Lekh AI Pro

Better LTX 2.5 prompts come from one paragraph, one clear action per shot, named cuts, and a stated soundtrack. Start with the beekeeper prompt, change one part at a time, and keep the wording that works. Lekh AI Pro lets you test LTX 2.5 on Mac with no cloud account. Download Lekh AI and run your first clip today.