Skip to content

Lightricks / video model

LTX 2.5

Explore LTX 2.5, its capabilities, use cases and ComfyUI workflows for video generation on HexGrid Cloud.

LTX-2.53 workflowsExplore workflows ↓

Deploy LTX 2.5 with ComfyUI

Choose a workflow to explore its setup, inputs, downloads and generation examples.

01Workflow guide

Text to Video

Create video from text prompts with LTX 2.5. Explore the ComfyUI setup, workflow instructions and generation examples.

Text → Video

Explore workflow
02Workflow guide

Image to Video

Animate an input image into video with LTX 2.5. Explore the ComfyUI setup, workflow instructions and generation examples.

Text + Image → Video

Explore workflow
03Workflow guide

First and Last Frame to Video

Create video between a starting image and an ending image with LTX 2.5. Explore the ComfyUI setup, workflow instructions and generation examples.

Text + Image → Video

Explore workflow

About LTX 2.5

LTX-2.5 is Lightricks' open-weight audio-video generation model, deployable on HexGrid.cloud with configurable inference workflows, custom model checkpoints and LoRA support.

It supports three primary generation modes: text-to-video, image-to-video and audio-to-video. Video and synchronized audio are generated together, with optional silent output. LTX-2.5 offers Fast and Pro variants, prioritizing inference speed and visual fidelity respectively.

LTX-2.5 features native multi-shot storytelling, consistent characters and environments across cuts, first-and-last-frame transitions, and audio-driven video generation. Both variants support resolutions up to 4K and frame rates up to 50 FPS. Fast supports clips up to 20 seconds at lower resolutions, while Pro supports up to 10 seconds.

What people use it for

Cinematic text-to-video

Generate cinematic scenes directly from detailed text prompts, controlling characters, environments, lighting, camera movement, dialogue and synchronized sound effects.

Image animation and transitions

Animate existing images into cinematic videos while preserving their composition and visual style. Optionally supply a final frame to create controlled transitions between two images.

Audio-driven video

Generate visuals synchronized to existing audio tracks, including speech, music and environmental sounds. Optionally provide a reference image to guide the opening frame and visual appearance.

Multi-shot storytelling

Generate multiple connected cinematic shots within a single video while maintaining character identity, environments, lighting, visual style and voice consistency across scene transitions.

Strengths

  • Video generation up to 4K resolution and 50 FPS, supporting landscape and portrait orientations
  • Native synchronized audio-video generation, including dialogue, ambient sounds and sound effects
  • Three primary generation modes: text-to-video, image-to-video and audio-to-video
  • Fast and Pro variants optimized for inference speed and higher visual fidelity respectively
  • Native multi-shot storytelling with consistent characters, environments, lighting and voice across cuts
  • First-frame and last-frame conditioning for controlled image-to-video transitions
  • Audio-conditioned video generation synchronized to supplied speech, music or soundtracks
  • Automatic duration selection based on prompt complexity and described actions
  • Eight camera-motion presets, including dolly movements, jib movements, static shots and focus shifts
  • Diffusion Video Decoder for improved texture detail, facial fidelity, typography and motion clarity
  • Diffusion Fidelity Rendering for improved visual quality in complex scenes
  • Custom Gemma 4 12B text encoder for stronger prompt understanding and adherence
  • Optional prompt enhancement for expanding short descriptions into detailed cinematic instructions
  • Open-weight architecture supporting LoRA fine-tuning, custom checkpoints and private deployment

Limitations

  • Maximum generation duration depends on resolution, frame rate and model variant. Fast supports up to 20 seconds at 720p or 1080p with 24 or 25 FPS, while Pro supports up to 10 seconds.
  • At 1440p and 4K, generation duration is limited to 10 seconds for both Fast and Pro variants.
  • Fast generation at 48 or 50 FPS is limited to 10 seconds, including at 720p and 1080p.
  • Standard generation supports 16:9 landscape and 9:16 portrait orientations. Other aspect ratios require custom processing or reframing workflows.
  • Audio-to-video duration follows the supplied audio track rather than an independently selected output duration.
  • Audio-to-video supports a maximum input audio duration of 20 seconds with Fast at 720p or 1080p, and 10 seconds for Pro or higher-resolution generation.
  • Automatic duration cannot be combined with last-frame conditioning in standard image-to-video generation; a fixed duration is required.
  • Retake, video extension and advanced video editing require specialized workflows rather than the standard generation modes.
  • Complex multi-character interactions, precise physical movements and extended scene transitions may still exhibit temporal inconsistencies or visual artifacts.
  • High-resolution generation and advanced workflows require substantial GPU memory and computational resources.
  • Fast prioritizes generation speed and efficiency, potentially sacrificing some fine detail and visual fidelity compared with Pro.

Specifications

Generation, input and serving details.

ProviderLightricksFamilyLTX-2.5CategoryvideoWorkflows3

Generation

Modes
Text-to-video, image-to-video, audio-to-video
Variants
Fast, Pro
Architecture
Diffusion Transformer with joint audio-video generation
Model parameters
22B diffusion transformer
Text encoder
Custom Gemma 4 12B
Resolution
720p, 1080p, 1440p, 4K
Default resolution
Configurable; 1080p commonly used
Duration
Fast: 6–20 seconds; Pro: 6–10 seconds
Duration options
Fast: 6, 8, 10, 12, 14, 16, 18, 20 seconds; Pro: 6, 8, 10 seconds; automatic duration supported
Frame rates
24, 25, 48, 50 FPS
Default frame rate
Configurable; 24 FPS commonly used
Aspect ratios
16:9, 9:16
Audio generation
Native synchronized audio, including dialogue, sound effects and environmental sounds
Multi-shot
Native multi-shot generation within a single video
Automatic duration
Supported for text-to-video and image-to-video
Video decoder
Diffusion Video Decoder
Rendering
Diffusion Fidelity Rendering

Inputs and Controls

Text prompt
Required for text-to-video; used for motion and scene guidance in image-to-video and audio-to-video
Start image
Required for image-to-video; optional for audio-to-video
End image
Optional for image-to-video and compatible audio-to-video workflows
Audio input
Required for audio-to-video
Audio input duration
Fast: up to 20 seconds at 720p/1080p and 10 seconds at 1440p/4K; Pro: up to 10 seconds
Audio-to-video resolution
720p, 1080p, 1440p, 4K
Camera motion
dolly_in, dolly_out, dolly_left, dolly_right, jib_up, jib_down, static, focus_shift
Automatic duration
Supported for text-to-video and image-to-video; duration determined from prompt content
Audio toggle
Optional synchronized audio generation; silent video output supported
Audio guidance
Configurable audio-conditioning guidance for compatible audio-to-video workflows
Prompt expansion
Optional prompt enhancement using a dedicated text model
Inference steps
Configurable through supported generation workflows
Guidance scale
Configurable for controlling prompt adherence
Random seed
Optional integer seed for reproducible generation
Frame interpolation
Optional through compatible interpolation workflows
LoRA support
Custom LoRAs supported through compatible inference workflows
Quantization
Optimized precision configurations supported for reduced memory requirements
Model offloading
Supported through compatible inference runtimes

Serving

Provider
Lightricks
API platform
HexGrid.cloud
Deployment
Private GPU deployment with configurable inference workflows
Generation workflows
Text-to-video, image-to-video and audio-to-video with Fast and Pro configurations
Output format
MP4 video with optional synchronized audio
Model availability
Open weights and private hosted deployment
Model license
LTX-2.x Community License; additional commercial licensing requirements may apply
Fine-tuning
LoRA training and custom model adaptation supported
Model checkpoints
Full-precision development and distilled checkpoints
Infrastructure
Dedicated GPU infrastructure with private model execution
Customization
Custom checkpoints, LoRAs, inference parameters, conditioning pipelines and post-processing workflows
GPU requirements
Dependent on model precision, resolution, frame count and optimization; high-memory GPUs recommended for full-precision inference

Explore related models