# LTX 2.3

> LTX 2.3 is a browser-based AI video generator built on the Lightricks 22-billion-parameter open-source model for cinematic text-to-video, image-to-video, and audio-to-video creation with character consistency, realistic physics, multi-style control, and native portrait output.

- Brand: LTX 2.3
- Domain: https://ltx23.app/
- Contact: support@ltx23.app

## Pages

- [LTX 2.3 AI Video Generator](https://ltx23.app/): Browser-based AI video generator built on the Lightricks 22B-parameter open-source DiT model, creating cinematic video from text, reference images, or audio with native portrait output and full commercial rights.
- [LTX 2.3](https://ltx23.app/ltx-2-3): Overview of the open-source 22B LTX 2.3 video model — text-to-video, image-to-video, and audio-to-video with 4K output and native audio sync.
- [LTX 2.3 Distilled](https://ltx23.app/ltx-2-3-distilled): Fast 8-step, low-VRAM distilled build of the LTX 2.3 video model for rapid iteration on consumer GPUs; keeps the full text/image/audio-to-video pipeline.
- [LTX 2.5](https://ltx23.app/ltx-2-5): The new open-source 22B LTX 2.5 model by Lightricks — native multishot scenes with character, environment, lighting, voice, and style continuity across cuts, synced audio, and a fast 8-step distilled checkpoint.

## Hero

- Title: LTX 2.3 AI Video Generator
- Description: Generate cinematic AI videos from text, images, or audio — LTX 2.3 delivers 22B-parameter open-source video fast.
- Workflows: Text-to-Video, Image-to-Video, Audio-to-Video, Native Portrait Video.

## What Makes LTX 2.3 the Fastest Open-Source Video Model

- Built on DiT architecture with 22 billion parameters, LTX 2.3 runs 18x faster than WAN 2.2 on H100 GPUs.
- LTX 2.3 Text-to-Video Generation: describe any scene and the model generates cinematic video with fluid motion, accurate lighting, and natural physics.
- Image-to-Video with LTX 2.3: upload a reference image and transform it into a dynamic video clip with smooth camera motion and lifelike animation.
- Audio-to-Video Synchronization: feed an audio track and the AI generates matching video with tight lip-sync, beat-aligned motion, and spatial audio cues.
- Native Portrait Video at 1080x1920: create vertical video natively, trained on real portrait data, ideal for Reels, Shorts, and TikTok.

## Why Creators Choose LTX 2.3

- 22 Billion Parameter DiT Engine: Diffusion Transformer architecture delivers sharper textures, finer edges, and better detail than prior versions.
- Expanded Text Connector (4x): a 4x-larger text connector interprets complex prompts with spatial layout, character actions, and mood accurately in one pass.
- Face & Character Preservation: consistent faces, expressions, and body proportions across frames, ideal for storytelling and multi-shot sequences.
- Multi-Modal Generation Pipeline: text-to-video, image-to-video, audio-to-video, video-to-video, and depth conditioning from one unified platform.
- Rebuilt VAE for Sharper Output: the reworked VAE and latent space produce crisper hair, cleaner edges, and better texture preservation at every resolution.
- Open Source & Free to Use: LTX 2.3 weights are open-source on Hugging Face, free for personal and commercial use under 10M annual revenue.

## How to Use the LTX 2.3 Video Generator (Three Steps)

1. Enter Your Prompt - describe the video you want in natural language, or upload an image or video clip as a creative reference.
2. Set Your Parameters - choose video duration, aspect ratio, and output quality to match your creative vision and target platform.
3. Generate and Download - click generate, wait a moment, then download your high-definition AI video ready to publish anywhere.

## Use Cases

- Cinematic Video - professional AI-generated videos with film-grade lighting and realistic motion.
- Creative Artwork - generate unique video content in any visual style with character consistency.
- Product Content - create stunning product videos for e-commerce and advertising.
- Social Media Content - generate eye-catching vertical and landscape video for any platform.

## FAQ

- What is LTX 2.3? A 22-billion-parameter open-source AI video model by Lightricks on the Diffusion Transformer (DiT) architecture, supporting text-to-video, image-to-video, audio-to-video, and video-to-video generation with native portrait output and character consistency.
- Do I need a powerful GPU? No. On ltx23.app all rendering happens in the cloud. If you prefer running locally, LTX 2.3 supports ComfyUI workflows and GGUF/FP8 quantized formats for lower hardware requirements.
- How does LTX 2.3 compare to other video models like WAN 2.2? On H100 GPUs, the LTX 2 series achieves roughly 18x the throughput of WAN 2.2 14B, plus native 9:16 portrait video, a reworked audio vocoder, and sharper edge detail from its rebuilt latent space.
- What video specs does the model support? Videos render at up to 1080p HD in 16:9, 9:16, 1:1, and 4:3 aspect ratios with durations from 4 to 20 seconds including audio-synced output. Prompts accept up to 2,000 characters.
- Is LTX 2.3 free to use? Yes — free credits for new users. The model weights are open-source and free for personal and commercial use for organizations under 10M annual revenue. Subscription plans offer volume discounts.
- Can I use outputs commercially? Yes — full commercial rights with no royalty fees, no watermarks, and no additional licensing required.
- Which creative styles does LTX 2.3 support? Photorealism, cinematic film, anime, cyberpunk, watercolor, oil painting, 3D rendering, time-lapse, and more, controlled through text prompts or reference images.
- How do I get started? Create a free account on ltx23.app, enter a text prompt, optionally upload a reference image or audio, set parameters, then click generate. Your video is ready to download in moments.

## Call to Action

- Title: Start Generating with LTX 2.3 — Free AI Video Online
- Button: Try Free Now (https://ltx23.app/generate)

## Blog (/blog)
- LTX 2.5 ComfyUI: The Three Native Workflows Explained (`/blog/ltx-2-5-comfyui`) — LTX 2.5 in ComfyUI: the three official workflows (T2V, I2V, FLF2V), every model file and its folder, the gated-repo access step, and what int8 means for your VRAM.
- LTX 2.5 Distilled vs Dev: The Two Checkpoints (`/blog/ltx-2-5-distilled-vs-dev`) — LTX 2.5 distilled vs dev transformers: the fixed 8-step schedule of the distilled checkpoint, the trainable dev model, and which one each pipeline and workflow expects.
- LTX 2.5 Download: The Gated Repo, Split Files, and Where They Go (`/blog/ltx-2-5-download`) — LTX 2.5 download guide: access to the gated Hugging Face repo, every official file in the split pack, the ~66 GiB Quick Start set, and where each file goes.
- LTX 2.5 LoRA: What Carries Over from 2.3 and What's New (`/blog/ltx-2-5-lora`) — LTX 2.5 LoRA compatibility: which 2.3 LoRAs carry over, the distilled LoRA and detailing IC-LoRA that ship with 2.5, and what version-locking means for your stack.
- LTX 2.5 Multishot: How It Works and How to Use It (`/blog/ltx-2-5-multishot`) — LTX 2.5 multishot explained: what connected shots actually are, how identity and voice persist across cuts, and how to write prompts that produce usable multi-scene output.
- LTX 2.5 Prompt Guide: Multishot, Audio, and the Gemma 4 Encoder (`/blog/ltx-2-5-prompt-guide`) — LTX 2.5 prompt guide: how the Gemma 4 12B encoder changes prompting, when to use the prompt enhancer, and the shot-list structure that makes multishot and audio output work.
- LTX 2.5 System Requirements: VRAM, GPU, and the int8 8 GB Path (`/blog/ltx-2-5-system-requirements`) — LTX 2.5 system requirements: what the official docs actually require, how the int8 and NVFP4 builds change the VRAM math, and what runs on 8 GB-class consumer GPUs.
- LTX 2.5 VAE: The Diffusion Decoder, the Conv Fallback, and the Audio VAE (`/blog/ltx-2-5-vae`) — LTX 2.5 has three official VAEs: a diffusion video decoder for quality, a conv VAE for speed, and an audio VAE with vocoder. Which to use, when.
- LTX 2.5 vs LTX 2.3: Should You Upgrade? (`/blog/ltx-2-5-vs-ltx-2-3`) — LTX 2.5 vs LTX 2.3 compared on the three decisions that matter: multishot continuity, file and LoRA incompatibility, and the speed of the new distilled path.
- Is LTX 2.5 Free? License, the $10M Threshold, and API Costs (`/blog/is-ltx-2-5-free`) — LTX 2.5 licensing: the LTX-2.x Community License, the $10M revenue threshold, the fine-tune transfer clause, and what the official API costs per second.


- How to Use GGUF in LTX 2.3: High-Quality AI Video on Low VRAM (`/blog/how-to-use-gguf-in-ltx-23`) — Step-by-step guide to running LTX Video 2.3 with quantized GGUF models in ComfyUI, including the Kijai node setup, required downloads, and low-VRAM optimization tips.
- LTX23 ComfyUI: Run LTX 2.3 Locally (`/blog/ltx23-comfyui`) — Complete walkthrough for setting up LTX 2.3 in ComfyUI for local AI video generation.

- Is LTX 2.3 Free? License and Real Costs Explained (`/blog/is-ltx-2-3-free`) — LTX 2.3 is free to download, but the license has a revenue threshold. Here is what the LTX-2 Community License actually says and what running it really costs.
- LTX Desktop: Run LTX 2.3 Locally Without ComfyUI (`/blog/ltx-2-3-desktop`) — LTX Desktop runs LTX 2.3 locally on NVIDIA GPUs and Apple Silicon, with an API fallback. Hardware thresholds, the free key, LoRA rules and traps.
- LTX 2.3 Distilled vs Dev: Which Checkpoint to Use (`/blog/ltx-2-3-distilled-vs-dev`) — LTX 2.3 distilled vs dev explained: step counts, CFG, which pipelines require which checkpoint, the 1.0 vs 1.1 difference, and a workflow that uses both.
- LTX 2.3 Download: Every Official File Explained (`/blog/ltx-2-3-download`) — A complete LTX 2.3 download guide: every official safetensors file, exact sizes, which ones you actually need, and the folder each one belongs in.
- LTX 2.3 FP8: The Official Quantization Path (`/blog/ltx-2-3-fp8`) — How LTX 2.3 FP8 quantization works: fp8-cast vs fp8-scaled-mm, the allocator flag you must set, ComfyUI FP8 builds, and where INT8, NVFP4 and GGUF fit.
- LTX 2.3 GitHub: Every Official Repo Explained (`/blog/ltx-2-3-github`) — A map of the LTX 2.3 GitHub and Hugging Face repositories: the LTX-2 monorepo packages, LTX Desktop, the ComfyUI integration, and which one to open for what.
- LTX 2.3 Image to Video: Setup and Consistency Fixes (`/blog/ltx-2-3-image-to-video`) — How LTX 2.3 image to video works: the I2V and FLF2V workflows, keyframe conditioning, why faces drift, and the frame and resolution rules that break runs.
- LTX 2.3 Inpainting, Retake and Extend Explained (`/blog/ltx-2-3-inpaint`) — What LTX 2.3 can actually edit: temporal retakes with region masks, video and audio extension, and where inpainting really lives in the codebase.
- LTX 2.3 Lip Sync: Talking Video and Audio Sync (`/blog/ltx-2-3-lip-sync`) — Four ways to get lip-synced video from LTX 2.3, plus modality_scale, the guidance term built purely for audio-visual sync. Full setup and settings.
- LTX 2.3 LoRA: IC-LoRA, ID-LoRA, and Training Your Own (`/blog/ltx-2-3-lora`) — Every official LTX 2.3 LoRA, what IC-LoRA and ID-LoRA actually do, how to stack them at the right strength, and the real hardware cost of training your own.
- LTX 2.3 Motion Control: Camera LoRAs and IC-LoRA (`/blog/ltx-2-3-motion-control`) — Three layers of motion and camera control in LTX 2.3: prompt language, dedicated camera-move LoRAs, and IC-LoRA guidance, plus the STG parameter.
- LTX 2.3 Prompt Guide: Write Prompts That Work (`/blog/ltx-2-3-prompt-guide`) — The official LTX 2.3 prompt structure, why 200 words is the ceiling, how to prompt synchronized audio and dialogue, and the prompt enhancer explained.
- LTX 2.3 Resolutions and Frame Count Rules (`/blog/ltx-2-3-resolutions`) — Why LTX 2.3 needs resolutions divisible by 32 and frame counts on the 8k+1 pattern, how two-stage upscaling changes your target, and how to plan clip length.
- LTX 2.3 Spatial Upscaler: Two-Stage Rendering (`/blog/ltx-2-3-spatial-upscaler`) — What the LTX 2.3 spatial upscaler does, why it is not an ESRGAN-style post-process, which file to download (x2 1.1 vs x1.5), and where the temporal upscaler fits.
- LTX 2.3 System Requirements: VRAM and GPU Guide (`/blog/ltx-2-3-system-requirements`) — What LTX 2.3 really needs: VRAM by precision, the official FP8 and offload flags, Python and CUDA versions, training requirements, and where Mac and AMD stand.
- LTX 2.3 VAE, Vocoder and Text Projection Explained (`/blog/ltx-2-3-vae`) — LTX 2.3 has two VAEs — video and audio — plus a 24 kHz vocoder and a Gemma text encoder. Here is what each does and why the official repo ships no separate VAE file.
- LTX 2.3 Video Engine: Architecture Explained (`/blog/ltx-2-3-video-engine`) — Inside the LTX 2.3 video engine: a 48-layer dual-stream transformer with separate video and audio streams, two VAEs, a 24 kHz vocoder, and Gemma 3 text encoding.
- LTX 2.3 Video to Video: IC-LoRA and Retakes (`/blog/ltx-2-3-video-to-video`) — How LTX 2.3 video-to-video actually works: the IC-LoRA pipeline and its distilled-only rule, structural control from depth and pose, HDR linear output, and retakes.
- LTX 2.3 vs Wan 2.2: Which Should You Run? (`/blog/ltx-2-3-vs-wan-2-2`) — LTX 2.3 vs Wan 2.2 compared on what matters: native audio, licensing thresholds, single-GPU feasibility, and which one fits your hardware and your legal situation.
- LTX 2.3 Workflow: All Six ComfyUI Templates (`/blog/ltx-2-3-workflow`) — A breakdown of every official LTX 2.3 workflow in ComfyUI — T2V, I2V, FLF2V, IA2V, IC-LoRA and ID-LoRA — which checkpoint each one needs and when to use it.

## Optional

- [Privacy Policy](https://ltx23.app/privacy-policy)
- [Terms of Service](https://ltx23.app/terms-of-service)
- [Refund Policy](https://ltx23.app/refund-policy)

- [WAN 2.2 vs LTX 2.3](https://ltx23.app/blog/wan-2-2-vs-ltx-2-3) — WAN 2.2 vs LTX 2.3 — compare architecture, speed, native portrait video, audio, hardware needs, and licensing of both open-source video models.

- [Is LTX 2.3 Better Than WAN 2.2](https://ltx23.app/blog/is-ltx-2-3-better-than-wan-2-2) — Is LTX-2.3 better than Wan 2.2? We compared the 22B Lightricks model and Alibaba's Wan 2.2 on audio, licensing, VRAM, and speed — see which one to run.