LTX 2.3 is a cutting-edge, open-source AI video generator developed by Lightricks, built upon the advanced Diffusion Transformer (DiT) architecture with an impressive 22 billion parameters. This powerful model is designed to rapidly create cinematic AI videos from various inputs, including text, images, and audio, making it an indispensable tool for creators across different domains. It boasts an 18x faster throughput compared to models like WAN 2.2 on H100 GPUs, ensuring efficient and quick video generation.
Key Features:
- Multi-Modal Generation Pipeline: LTX 2.3 supports comprehensive input types, allowing users to generate videos from text descriptions (text-to-video), static images (image-to-video), and audio tracks (audio-to-video). It also offers video-to-video transformation and depth conditioning for advanced creative control.
- High-Fidelity Output: The 22-billion-parameter DiT engine delivers superior visual quality with sharper textures, finer edges, and enhanced detail. A rebuilt Variational Autoencoder (VAE) and optimized latent space further contribute to crisper hair, cleaner edges, and better texture preservation across all resolutions.
- Expanded Text Connector: Featuring a 4x-larger text connector, LTX 2.3 can interpret complex prompts with remarkable accuracy, understanding spatial layouts, character actions, and moods in a single pass.
- Face & Character Preservation: The model ensures consistent faces, expressions, and body proportions across video frames, which is crucial for compelling storytelling and multi-shot sequences.
- Native Portrait Video: LTX 2.3 is uniquely trained on real portrait data, enabling native vertical video generation at 1080x1920 resolution. This makes it ideal for platforms like Reels, Shorts, and TikTok, avoiding the quality compromises of cropped landscape videos.
- Audio Synchronization: When provided with an audio track, LTX 2.3 generates matching video content with tight lip-sync, beat-aligned motion, and spatial audio cues, perfect for music videos, voiceovers, and localized ads.
- Flexible Resolution & Aspect Ratios: Videos can be rendered at up to 1080p HD in various aspect ratios, including 16:9, 9:16, 1:1, and 4:3, with durations ranging from 4 to 20 seconds.
- Open Source & Commercial Use: The LTX 2.3 model weights are open-source and available on Hugging Face, free for personal and commercial use (under 10M annual revenue). Videos generated on ltx23.app also come with full commercial rights, free from watermarks and royalty fees.
Use Cases:
LTX 2.3 empowers a diverse range of creators, including filmmakers, marketers, developers, and social media managers. It can be used to:
- Generate cinematic b-roll and scenes with fluid motion and natural physics.
- Transform app mockups into polished walkthrough demos.
- Create product videos at scale for e-commerce and advertising.
- Produce localized video ads for multiple markets with accurate audio-to-video sync.
- Develop engaging vertical video clips for social media platforms like TikTok and Instagram Reels.
- Animate storyboard frames into cinema-grade motion sequences.
- Create game trailers that rival hand-crafted cinematics, saving weeks of animation work.
The platform offers a user-friendly experience, requiring no prior video editing or AI expertise. Users simply enter a text prompt, optionally upload reference media, set parameters like duration and aspect ratio, and generate high-definition AI videos ready for immediate publication. For those preferring local execution, LTX 2.3 supports ComfyUI workflows and quantized formats (GGUF/FP8) for lower hardware requirements, with a recommended setup of an NVIDIA GPU with 32 GB+ VRAM.
