README.md

metadata

title: LTX-2.3 Image-to-Video
emoji: 🎞️
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 6.19.0
python_version: '3.12'
app_file: app.py
pinned: true
license: apache-2.0

LTX-2.3 Image-to-Video & Seamless Loops

This Space provides a robust interface for the LTX-2.3 video generation model. It specializes in high-quality Image-to-Video (I2V) generation with native support for Seamless Looping (cyclic motion) and customizable resolution tiers.

Key Features

• Seamless Looping: Dedicated mode to force the final frame to match the first, enabling perfectly infinite background loops.
• Resolution Presets: Easily toggle between "High" and "Low" resolution modes for different artistic needs and hardware constraints.
• Prompt Enhancement: Built-in capability to refine motion prompts for cinematic animation.
• Tiling Presets: Select from default, low-memory, and high-quality configurations to optimize VRAM usage during the diffusion process.

How it works

The pipeline uses the Gemma 3 12B text encoder to translate your motion descriptions into conditioning for the LTX-2 distilled video model. Our implementation includes custom optimizations to handle the complex memory requirements of large-scale video generation on GPU hardware.

Quick Start

1. Upload an image: Provide a starting frame for the animation.
2. Describe the motion: Use the "Motion Prompt" to define how the subject should behave.
3. Configure Settings:
   • Enable Seamless Loop if you want the output to be cyclic.
   • Choose your desired Duration (up to 10 seconds).
4. Generate: Click "Generate I2V" to begin the diffusion process.

Built With

• Lightricks LTX-2
• Gemma 3 (Google)
• Hugging Face Spaces (Gradio)

---

Note: This implementation includes custom device-mapping patches for transformers 5.x and meta-tensor handling to ensure compatibility.