# LTX-2.3 Prompt Guide: Tips For Prompting LTX-2.3

## Key Takeaways

- LTX-2.3 responds best to long, detailed prompts — the more specific you are about subject, action, lighting, camera movement, and audio, the closer the output matches your vision.
- Break dialogue into short phrases with acting directions between each line, and use physical cues rather than emotional labels to direct character performance.
- Match your prompt length to your video length — short prompts for long videos leave the model without enough direction to fill the duration.

Strong prompts produce better videos. [LTX models](https://ltx.io/model) respond best to detailed, descriptive prompts that paint a complete picture of the scene you're generating. Think of it as writing a shot description for a cinematographer.

If you're new to writing prompts for video generation, this guide will help you construct effective, production-ready prompts.# LTX-2.3 Prompt Guide: Tips For Prompting LTX-2.3

## Key Takeaways

- Specificity wins: LTX-2.3 responds best to long, detailed prompts. The more specific you are about subject, spatial layout, lighting, camera movement, and audio, the closer the output matches your vision.
- Direct the scene: Block your shots like a director. Use spatial relationships (left vs. right, foreground vs. background) to map out the scene.
- Motion is driven by verbs: Avoid static, photo-like prompts. If your prompt reads like a still image, the video will behave like one. Use strong action verbs to drive motion.
- Native Portrait Support: LTX-2.3 natively supports vertical video (up to 1080x1920). Don't treat vertical as a cropped landscape—compose and frame specifically for it.
- Match prompt length to video length: Short prompts for long videos leave the model without enough direction to fill the duration. 

Strong prompts produce better videos. [LTX models](https://ltx.io/model) respond best to detailed, descriptive prompts that paint a complete picture of the scene you're generating. Think of it as writing a shot description for a cinematographer.

## What Changed in LTX-2.3

[LTX-2.3](https://ltx.io/model/ltx-2-3) introduces major improvements to detail, motion, prompt understanding, audio reliability, and native portrait support. This changes how you should prompt:

#### 1. A Larger, More Capable Text Connector
Specific descriptions of facial expressions, spatial relationships, and emotional beats translate more reliably. Previously, simplifying prompts improved consistency. Now, specificity wins. You can confidently layer multiple actions, combine detailed environments with character performance, and introduce precise stylistic constraints.

#### 2. Sharper Details (Updated VAE & Latent Space)
With a rebuilt latent space and updated VAE, fine detail is sharper across resolutions. You can now explicitly describe micro-textures: fabric types, individual hair strands, surface finishes, and environmental wear.

#### 3. Native Portrait Support
LTX-2.3 is trained on vertical data and natively supports vertical video up to 1080x1920. When generating portrait content, compose for vertical intentionally (e.g., *"Influencer vlogging while on holiday"*). 

#### 4. Better Audio (New Vocoder)
The new vocoder improves audio reliability and alignment. Specific inputs produce more controlled outputs, so describe environmental audio, tone, intensity, and dialogue clarity in detail.

#### 5. Prompt length matters
Longer, more descriptive prompts consistently outperform short ones. If you're generating longer videos (8–10 seconds), make sure your prompt is detailed enough to fill the duration. 


## Core Principles

#### Be specific and descriptive
Instead of *"A woman in a café"*, try *"A woman in her 30s sits by the window of a small Parisian café. Rain runs down the glass behind her. Warm tungsten interior lighting. She slowly stirs her coffee while glancing at her phone. Background softly out of focus."*

#### Direct the Scene, Don't Just Describe It
LTX-2.3 is excellent at respecting spatial layout. Block the scene like a director. Be explicit about:
- Left vs. right
- Foreground vs. background
- Facing toward vs. away
- Distance between subjects

#### Avoid Static, Photo-Like Prompts
If your prompt reads like a still image, the output may behave like one. 
Instead of *"A dramatic portrait of a man standing,"* try *"A man stands on a windy rooftop. His coat flaps in the wind. He adjusts his collar and steps forward as the camera tracks right."*

#### Break dialogue into segments
When prompting for speaking characters, break long sentences into shorter phrases with acting directions between them. This gives the model explicit direction on pacing, emotion, and physical acting for each beat.


## Key Elements to Include

When writing a prompt, aim to include the following elements:

#### 1. Establish the Shot
Use cinematography terms that match your intended genre. Include shot scale, framing, or category-specific characteristics.

#### 2. Set the Scene & Textures
Describe lighting conditions, color palette, and atmosphere. Because of the updated VAE, be sure to describe surface textures, fabric types, and edge details to establish mood and realism.

#### 3. Describe the Action (Use Verbs!)
Write the core action as a natural sequence. Specify *who* moves, *what* moves, and *how* they move. Motion is driven by verbs.

#### 4. Define the Character(s)
Include age, hairstyle, clothing, and distinguishing features. Express emotion through physical cues, not abstract labels.

#### 5. Identify Camera Movement(s)
Specify how and when the camera moves. Describing how subjects appear after the movement helps the model complete the motion accurately.

#### 6. Describe the Audio
Clearly describe ambient sound, music, speech, or singing.
- Place spoken dialogue in quotation marks.
- Describe the tone and intensity of the sounds.


## Tips by Use Case

#### Text-to-Video
Start with a strong visual description. Include subject, spatial relationships, action, environment, lighting, camera movement, and audio. The engine holds structure under complexity, so don't be afraid to layer multiple actions into one shot.

#### Image-to-Video
One of the biggest upgrades in 2.3 is reduced freezing and more natural motion, but motion still needs clarity. Focus your prompt on the verbs. Avoid passive phrases like *"The scene comes alive."* Instead, describe exactly what happens next: *"The camera slowly pushes forward as the subject turns their head and begins walking toward the street. Cars pass."* 

#### Audio-to-Video
Your [audio input](https://ltx.io/model/capabilities/audio-to-video) anchors the temporal structure. Use the prompt to describe the visual interpretation of that audio — what scenes, subjects, and camera work should accompany the soundtrack.


## What Works Well

| Strength | Description |
|---|---|
| Cinematic compositions | Wide, medium, and close up shots with thoughtful lighting, shallow depth of field, and natural motion |
| Complex Spatial Layouts | Foreground/background layering, left/right positioning, clear distances between subjects |
| Micro-Textures | Individual hair strands, fabric weaves, surface wear, and edge details |
| Emotive human moments | Strong single subject emotional expressions, subtle gestures, and facial nuance |
| Clear camera language | Explicit instructions like "slow dolly in" or "handheld tracking" |
| Voice capabilities | Characters can talk and sing, with support for multiple languages |

## What to Avoid

| Avoid | Why It Doesn't Work |
|---|---|
| Static / Photo-like prompts | Describing a still scene results in frozen, static video outputs. Use verbs. |
| Internal emotional states | Use visual cues instead of labels like "sad" or "confused" |
| Passive I2V motion | "The image comes alive" is too vague. Specify *who* and *what* moves. |
| Text and logos | Readable text is not currently reliable |
| Conflicting lighting | Mixed light logic confuses scene interpretation |

## Common Mistakes

- Too vague: "A nice video of nature" — the model has too many options and picks arbitrarily. Be specific about what's in the frame.
- Treating Vertical as Cropped Landscape: If generating 1080x1920, compose the shot for a vertical frame (e.g., vlogging, full body portraits) rather than forcing wide cinematic scenes into a narrow box.
- Mismatched duration: A 10-word prompt for a 10-second video — the model doesn't have enough direction to fill the duration. Long videos need long prompts.
- Conflicting directions: "A still, peaceful lake with dramatic waves crashing" — contradictions confuse the model. Be internally consistent.

## Sample Prompts

##### Example 1: Spatial Blocking & Complex Action
> Two people stand facing each other on a quiet suburban sidewalk. The taller man stands on the left, hands in pockets, wearing a worn leather jacket. The woman stands on the right, holding a bicycle. The camera slowly pushes forward as the man pulls his hands from his pockets and gestures toward the street. A car passes in the blurred background. Soft afternoon backlight catches the edge detail of the woman's curly hair.

##### Example 2: Dialogue & Audio
> The shot opens on a news reporter standing in the foreground, yellow caution tape fluttering behind him. The light is warm, early sun reflecting off the camera lens. A low, pulsing energy hum radiates from the background. Reporter (live): "Thank you, Sylvia. And yes — this is a sentence I never thought I'd say..." He pauses, eyes widening as a sharp, intermittent alarm blares in the background, metallic and urgent. Reporter (shouting over the noise): "Something is happening behind me!" The camera pans right, revealing a glowing orb emerging from the ground.

## Helpful Terms

#### Categories
- Animation: Stop-motion, 2D / 3D animation, Claymation, Hand-drawn
- Stylized: Comic book, Cyberpunk, 8-bit pixel, Surreal, Minimalist, Painterly, Illustrated
- Cinematic: Period drama, Film noir, Fantasy, Epic space opera, Thriller, Modern romance, Documentary

#### Visual Details
- Lighting: Flickering candles, Neon glow, Natural sunlight, Dramatic shadows, Soft backlight
- Textures: Rough stone, Smooth metal, Worn fabric, Glossy surfaces, Fine hair strands
- Positioning: Foreground, Background, Left-frame, Right-frame, Facing away

#### Sound and Voice
- Ambient Settings: Coffeeshop noise, Wind and rain, Metallic alarms, Pulsing hums
- Dialogue Style: Energetic announcer, Resonant voice with gravitas, Distorted radio-style
- Volume: Whisper, Mutter, Shout, Scream

#### Technical Style Markers
- Camera Language: Tracks, Pans across, Tilts upward, Pushes in / pulls back, Handheld movement, Over-the-shoulder
- Visual Effects: Particle systems, Motion blur, Depth of field

---

# Super important info that you have to follow

The AI doesn't understand the word "no". So if you write your prompt like *"No extra text on screen, no subtitles, no logos, no background distractions, no exaggerated gestures, no camera shake,"* it will have the opposite impact. It will add text to the screen, subtitles to the screen, and more. 

Never ever mention a word that you do not want in the video. So how to avoid such words? 
1. Don't use them in the prompt. 
2. Don't use words that are related to those words. 
3. Try to use words opposite to them. 

For example, "no background music" will bring background music. Instead, you can use "clean audio". Think this way to construct your prompt.