LightX2V is an advanced lightweight image/video generation inference framework engineered to deliver efficient, high-performance image/video synthesis solutions. This unified platform integrates multiple state-of-the-art image/video generation techniques, supporting diverse generation tasks including text-to-video (T2V), image-to-video (I2V), text-to-image (T2I), image-editing (I2I). X2V represents the transformation of different input modalities (X, such as text or images) into vision output (Vision).
Quick Start Β· Models Β· Documentation & Deployment
π Try it online now! LightX2V Studio is a free, lightweight, and fast AI video generation platform. No installation is required to try MiniMax-H3 / Wan 2.2 / SekoTalk / Qwen-Image / SwiftVR and more.
π MacBook local inference app: LightX2V APP
π€ HuggingFace Model Repository: LightX2V HuggingFace
π Developer Quick Start: LightX2V Developer Quick Start Guide
π Technical Blog: LightX2V Blog
π Join our WeChat group! LightX2V Robot WeChat ID: random42seed
Before submitting, please ensure that the code format conforms to the project standard. You can use the following execution command to ensure the consistency of project code format.
pip install ruff pre-commit
pre-commit run --all-filesWe thank the following community developers for their contributions to LightX2V:
More community contributors
-
October 13, 2026: π We released the LightX2V APP, which supports local inference on MacBook with Qwen-Image-2.1 Turbo, PE-T2I GGUF Q8_0, and PE-I2I GGUF Q8_0 models.
-
September 22, 2026: π We added and optimized support for SwiftVR, covering image and video restoration and super-resolution. See the usage guide.
-
September 20, 2026: π We added day-0 support for Qwen-Image-2.1, covering text-to-image generation and prompt-based image editing. See the usage guide.
-
August 27, 2026: π We released the MiniMax-H3 Turbo 8-step 768p distilled LoRA for generating 768p video with synchronized audio in 8 inference steps using MiniMax-H3. See the MiniMax-H3 guide.
-
August 11, 2026: π We released the MiniMax-H3 Turbo 4-step 768p distilled LoRA and added support for generating 768p video with synchronized audio in 4 inference steps without CFG. See the MiniMax-H3 guide.
-
August 7, 2026: π We added support for synchronized video and audio generation with MiniMax-H3 and released a Prompt Rewriter LoRA that expands short prompts into structured descriptions of visuals, soundscapes, and music. See the MiniMax-H3 guide.
Earlier updates
-
July 28, 2026: π We release the LightLingBot-Video 4-step distilled LoRA for LingBot-Video. It supports T2V, T2I, and I2V generation in only 4 inference steps without CFG. See the LingBot-Video inference scripts for usage.
-
July 23, 2026: β‘οΈ We release the blog: LightX2V ROS: Closing the Loop for Action-Generating World Models
-
July 19, 2026: β‘οΈ We release the blog: Wan2.2-NVFP4-Sparse: Extremely Fast Wan 2.2 14B Inference
-
June 15, 2026: π Supported deployment on T-head PPU.
-
May 29, 2026: π We introduce an extremely efficient Wan 2.2 14B variant (T2V and I2V): NVFP4 Quantization-Aware Step Distillation with Sparse Attention for Blackwell Architecture. Under the test conditions in the model card, inference with 4 NFE achieves over 50Γ end-to-end speedup over the original model with 40 NFE on a single RTX 5090 GPU.
-
April 30, 2026: π We now support deployment on iluvatar. Thanks to the iluvatar team.
-
April 20, 2026: π We are excited to release the Wan2.2-I2V-A14B-4step-720p-high and Wan2.2-I2V-A14B-4step-720p-low models. Compared to previous iterations, this version was trained on a high-quality 720p dataset and features an optimized low-noise training algorithm. These enhancements significantly boost the model's performance in fine-grained detail rendering and visual texture.
-
April 17, 2026: π We support the WorldMirror 2.0 model. On a single H100 GPU, LightX2V delivers approximately 1.2Γ speedup. For detailed usage instructions, please refer to this guide.
-
April 10, 2026: π We released a technical blog on disaggregated deployment
-
March 5, 2026: π We now support deployment on Intel AIPC PTL. Thanks to the Intel team.
-
March 5, 2026: π We now support disaggregated deployment based on Mooncake. More improvements and documentation for disaggregated deployment are in progress. Thanks to the Mooncake team for their help!
-
February 27, 2026: π We now support FP8 and NVFP4 quantization for autoregressive video generation models (Self Forcing)! You can find the quantized model here: Self-Forcing-FP8, Self-Forcing-NVFP4.
-
February 11, 2026: π We are excited to announce GenRL - a scalable reinforcement learning framework for visual generation! GenRL enables training diffusion/flow models with multi-reward optimization (HPSv3, VideoAlign, etc.) using GRPO algorithm. We've released high-performance LoRA checkpoints trained with multi-node multi-GPU setup, demonstrating significant improvements in aesthetic quality, motion coherence, and text-video alignment. Check out our model collection on HuggingFace! Give us a β if you find it useful!
-
January 20, 2026: π We support the LTX-2 audio-video generation model, featuring CFG parallelism, block-level offload, and FP8 per-tensor quantization. Usage examples can be found in examples/ltx2 and scripts/ltx2.
-
January 6, 2026: π We updated the 8-step CFG/step-distilled models for Qwen-Image-2512 and Qwen/Qwen-Image-Edit-2511. You can download the corresponding weights from Qwen-Image-Edit-2511-Lightning and Qwen-Image-2512-Lightning for use. Usage tutorials can be found here.
-
January 6, 2026: π Supported deployment on Enflame S60 (GCU).
-
December 31, 2025: π We support the Qwen-Image-2512 text-to-image model since Day 0. Our HuggingFace has been updated with CFG / step-distilled LoRA. Usage examples can be found here.
-
December 27, 2025: π Supported deployment on MThreads MUSA.
-
December 25, 2025: π Supported deployment on AMD ROCm and Ascend 910B.
-
December 23, 2025: π We support the Qwen-Image-Edit-2511 image editing model since Day 0. On a single H100 GPU, LightX2V delivers approximately 1.4Γ speedup. We support for CFG parallelism, Ulysses parallelism, and efficient offloading technologies. Our HuggingFace has been updated with CFG / step-distilled LoRA and FP8 weights. Usage examples can be found here. Combined with LightX2V, 4-step CFG / step distillation, and the FP8 model, the maximum acceleration can reach up to approximately 42Γ. Feel free to try LightX2V Online Service with Image to Image and Qwen-Image-Edit-2511 model.
-
December 22, 2025: π Added Wan2.1 NVFP4 quantization-aware 4-step distilled models; weights are available on HuggingFace: Wan-NVFP4.
-
December 15, 2025: π Supported deployment on Hygon DCU.
-
December 4, 2025: π Supported GGUF format model inference & deployment on Cambricon MLU590/MetaX C500.
-
November 24, 2025: π We released 4-step distilled models for HunyuanVideo-1.5! These models enable ultra-fast 4-step inference without CFG requirements, achieving approximately 25x speedup compared to standard 50-step inference. Both base and FP8 quantized versions are now available: Hy1.5-Distill-Models.
-
November 21, 2025: π We support the HunyuanVideo-1.5 video generation model since Day 0. With the same number of GPUs, LightX2V can achieve a speed improvement of over 2 times and supports deployment on GPUs with lower memory (such as the 24GB RTX 4090). It also supports CFG/Ulysses parallelism, efficient offloading, TeaCache/MagCache technologies, and more. We will soon update more models on our HuggingFace page, including step distillation, VAE distillation, and other related models. Quantized models and lightweight VAE models are now available: Hy1.5-Quantized-Models for quantized inference, and LightTAE for HunyuanVideo-1.5 for fast VAE decoding. Refer to this for usage tutorials, or check out the examples directory for code examples.
We recommend using LightX2V through its service API. The example below uses MiniMax-H3 for text-to-audio-video (T2AV) generation.
We recommend the Docker environment. Clone the repository in your prepared environment:
git clone https://github.com/ModelTC/LightX2V.git
cd LightX2VFollow the MiniMax-H3 guide to prepare the weights and initial AdaLN cache, check the required kernels, and set the repository path, model path, and GPU in the server script. Run the commands below from the repository root.
Other installation options (Git / source)
Run these commands in a prepared Python environment. The service example uses scripts from the cloned repository.
pip install -v git+https://github.com/ModelTC/LightX2V.gituv pip install -v . # or: pip install -v .See the environment setup guide for attention and quantization kernel installation.
# Terminal 1: start the server
bash scripts/minimax_h3/server/start_server.shAfter the model loads, submit the example request from another terminal:
# Terminal 2: submit a T2AV request
python scripts/minimax_h3/server/post_t2av.pyEdit the prompt in the request script to generate your own content. Requests are asynchronous; results are saved by default to save_results/server_cache/outputs/ under the repository root.
Models and variants are grouped by their primary use.
- β MiniMax-H3 (including the causal variant)
- β LTX-2 / LTX-2.3 (including the autoregressive variant) / LTX-2.5 (distilled)
- β HunyuanVideo-1.5
- β Wan2.1 / Wan2.2
- β Wan2.1-VACE / Wan2.2-VACE-Fun-A14B
- β Wan2.2-S2V
- β Wan2.2-Animate / Wan2.2-Animate-2 (distilled)
- β LingBot-Video (dense / MoE)
- β SekoTalk (including the autoregressive variant)
- β InfiniteTalk
- β Wan-Dancer
- β Self-Forcing
- β Qwen-Image / Qwen-Image-2512
- β Qwen-Image-Edit / 2509 / 2511
- β Qwen-Image-2.1 / Turbo
- β Qwen-Image-Layered
- β FLUX.2 dev / klein (including base and distilled variants)
- β HunyuanImage-3.0-Instruct
- β Z-Image / Z-Image-Turbo
- β LongCat-Image / Edit / Edit-Turbo
- β BAGEL
- β HiDream-O1-Image / Dev-2604
- β ERNIE-Image / Turbo
- β SenseNova-U1 (NeoPP backend, used with LightLLM)
- β Cosmos3-Nano / Super (including Nano-Policy-DROID)
- β LingBot-World / Fast
- β Matrix-Game-2.0 / 3.0
- β WorldPlay (bidirectional, autoregressive, and distilled)
- β WorldMirror 2.0 (3D reconstruction)
- β DreamZero
- β LingBot-VA
- β Motus
- β Fast-WAM
- β RealtimeWAM (based on Fast-WAM / Faster-WAM)
- β Hunyuan3D-2.1 (image-to-3D shape generation)
- β MiniMax-H3 Turbo β 4-/8-step 768p distilled LoRAs for T2AV / FL2AV
- β LightLingBot-Video β 4-step distilled LoRA for T2V, T2I, and I2V
- β Wan2.1-Distill-Models / Wan2.2-Distill-Models
- β Wan2.1-Distill-Loras / Wan2.2-Distill-Loras
- β Wan-NVFP4 / LightWan2.2-A14B
- β Hy1.5-Distill-Models / Hy1.5-Quantized-Models
- β Qwen-Image-Lightning / Qwen-Image-2512-Lightning / Qwen-Image-2.1-Lightning
- β Qwen-Image-Edit-2511-Lightning
- β Z-Image-Turbo-Quantized
- β Self-Forcing-FP8 / Self-Forcing-NVFP4
- β Autoencoders (LightVAE / LightTAE)
- Few-Step Inference: Support distilled checkpoints and distillation LoRAs, including DMD2 and Wan2.1 MeanFlow inference.
- Quantization & Optimized Kernels: Support W8A8 (INT8 / FP8) and W4A4 (NVFP4) quantization, with optimized attention backends such as FlashAttention and SageAttention.
- Hierarchical Offloading: Manage weights across disk, CPU, and GPU with model-level and block-level offloading.
- Feature Caching: Support TeaCache, MagCache, and TaylorSeer to reuse intermediate results across denoising steps.
- Parallel Inference: Support tensor parallelism (TP), sequence parallelism (SP), CFG parallelism, and hybrid configurations.
- Variable-Resolution Inference: Schedule latent resolutions across denoising stages to control computation at each stage.
Available optimizations depend on the model, hardware, and selected configuration.
| Use Case | Documentation & Examples |
|---|---|
| Environment Setup & Configuration | Quick Start Guide |
| Model Preparation & Loading | MiniMax-H3 Β· Wan2.1 / VACE Β· Model Formats & Loading |
| Service Deployment & API Usage | Service Deployment Β· API Reference |
| Interactive Interfaces & Workflow Integration | Gradio-Lightx2v Β· ComfyUI |
| Local Deployment on Windows | Windows Deployment Guide |
| Resource-Constrained and Low-Latency Deployment | Low-Resource Deployment Β· Low-Latency Deployment |
| Model Quantization & LoRA Deployment | Quantization Β· LoRA Deployment |
| Inference Performance Optimization | Weight Offloading Β· Feature Caching Β· Attention Kernels Β· Parallel Inference Β· Variable-Resolution Inference Β· Step Distillation |
| Wan2.1 NVFP4 4-Step Inference | I2V Example Β· T2V Example Β· Quantization Kernel Installation |
| Code Examples & Development References | examples |
We thank the following open-source projects and communities for the models, tools, and research that have helped shape LightX2V:
More open-source projects
We also thank the cloud inference platforms supporting the wider ecosystem:
-
Atlas Cloud β a full-modal AI inference platform whose hosted API serves the Wan / Seedance / Kling model families, for teams that prefer a managed API over self-hosting.
-
Sensecore β Sensecore builds efficient, low-cost, scalable AI cloud infrastructure and develops deep learning platforms, algorithms, and models to support AI innovation in industry and academia.
If you find LightX2V useful in your research, please consider citing our work:
@misc{lightx2v,
author = {LightX2V Contributors},
title = {LightX2V: Light Video Generation Inference Framework},
year = {2025},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/ModelTC/lightx2v}},
}For questions, suggestions, or support, please feel free to reach out through:
- π GitHub Issues - Bug reports and feature requests