README_zh.md 15.1 KB
Newer Older
Harahan's avatar
Harahan committed
1
<div align="center" style="font-family: charter;">
gushiqiao's avatar
gushiqiao committed
2
  <h1>⚡️ LightX2V:<br> 轻量级视频生成推理框架</h1>
Harahan's avatar
Harahan committed
3
4
5
6
7
8
9

<img alt="logo" src="assets/img_lightx2v.png" width=75%></img>

[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/ModelTC/lightx2v)
[![Doc](https://img.shields.io/badge/docs-English-99cc2)](https://lightx2v-en.readthedocs.io/en/latest)
[![Doc](https://img.shields.io/badge/文档-中文-99cc2)](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest)
helloyongyang's avatar
helloyongyang committed
10
[![Papers](https://img.shields.io/badge/论文集-中文-99cc2)](https://lightx2v-papers-zhcn.readthedocs.io/zh-cn/latest)
Yang Yong (雍洋)'s avatar
Yang Yong (雍洋) committed
11
[![Docker](https://img.shields.io/badge/Docker-2496ED?style=flat&logo=docker&logoColor=white)](https://hub.docker.com/r/lightx2v/lightx2v/tags)
Harahan's avatar
Harahan committed
12

helloyongyang's avatar
helloyongyang committed
13
14
**\[ [English](README.md) | 中文 \]**

Harahan's avatar
Harahan committed
15
16
</div>

helloyongyang's avatar
helloyongyang committed
17
18
--------------------------------------------------------------------------------

gushiqiao's avatar
gushiqiao committed
19
20
**LightX2V** 是一个先进的轻量级视频生成推理框架,专为提供高效、高性能的视频合成解决方案而设计。该统一平台集成了多种前沿的视频生成技术,支持文本生成视频(T2V)和图像生成视频(I2V)等多样化生成任务。**X2V 表示将不同的输入模态(X,如文本或图像)转换为视频输出(V)**

Yang Yong (雍洋)'s avatar
Yang Yong (雍洋) committed
21
22
## :fire: 最新动态

23
24
- **2025年11月24日:** 🚀 我们发布了HunyuanVideo-1.5的4步蒸馏模型!这些模型支持**超快速4步推理**,无需CFG配置,相比标准50步推理可实现约**25倍加速**。现已提供基础版本和FP8量化版本:[Hy1.5-Distill-Models](https://huggingface.co/lightx2v/Hy1.5-Distill-Models)

Yang Yong (雍洋)'s avatar
Yang Yong (雍洋) committed
25
- **2025年11月21日:** 🚀 我们Day0支持了[HunyuanVideo-1.5](https://huggingface.co/tencent/HunyuanVideo-1.5)的视频生成模型,同样GPU数量,LightX2V可带来约2倍以上的速度提升,并支持更低显存GPU部署(如24G RTX4090)。支持CFG并行/Ulysses并行,高效Offload,TeaCache/MagCache等技术。同时支持沐曦,寒武纪等国产芯片部署。我们很快将在我们的[HuggingFace主页](https://huggingface.co/lightx2v)更新更多模型,包括步数蒸馏,VAE蒸馏等相关模型。量化模型和轻量VAE模型现已可用:[Hy1.5-Quantized-Models](https://huggingface.co/lightx2v/Hy1.5-Quantized-Models)用于量化推理,[HunyuanVideo-1.5轻量TAE](https://huggingface.co/lightx2v/Autoencoders/blob/main/lighttaehy1_5.safetensors)用于快速VAE解码。使用教程参考[这里](https://github.com/ModelTC/LightX2V/tree/main/scripts/hunyuan_video_15),或查看[示例目录](https://github.com/ModelTC/LightX2V/tree/main/examples)获取代码示例。
Yang Yong (雍洋)'s avatar
Yang Yong (雍洋) committed
26
27


helloyongyang's avatar
helloyongyang committed
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
## 🏆 性能测试数据 (更新于 2025.12.01)

### 📊 推理框架之间性能对比 (H100)

| Framework | Cards | Step Time | Speedup |
|-----------|---------|---------|---------|
| Diffusers | 1 | 9.77s/it | 1x |
| xDiT | 1 | 8.93s/it | 1.1x |
| FastVideo | 1 | 7.35s/it | 1.3x |
| SGL-Diffusion | 1 | 6.13s/it | 1.6x |
| **LightX2V** | 1 | **5.18s/it** | **1.9x** 🚀 |
| FastVideo | 8 | 2.94s/it | 1x |
| xDiT | 8 | 2.70s/it | 1.1x |
| SGL-Diffusion | 8 | 1.19s/it | 2.5x |
| **LightX2V** | 8 | **0.75s/it** | **3.9x** 🚀 |

### 📊 推理框架之间性能对比 (RTX 4090D)

| Framework | Cards | Step Time | Speedup |
|-----------|---------|---------|---------|
| Diffusers | 1 | 30.50s/it | 1x |
| xDiT | 1 | OOM | OOM |
| FastVideo | 1 | OOM | OOM |
| SGL-Diffusion | 1 | 22.66s/it | 1.3x |
| **LightX2V** | 1 | **20.26s/it** | **1.5x** 🚀 |
| FastVideo | 8 | 15.48s/it | 1x |
| xDiT | 8 | OOM | OOM |
| SGL-Diffusion | 8 | OOM | OOM |
| **LightX2V** | 8 | **4.75s/it** | **3.3x** 🚀 |

### 📊 LightX2V不同配置之间性能对比

| Framework | GPU | Configuration | Step Time | Speedup |
|-----------|-----|---------------|-----------|---------------|
| **LightX2V** | H100 | 8 cards + cfg | 0.75s/it | 1x |
| **LightX2V** | H100 | 8 cards + no cfg | 0.39s/it | 1.9x |
| **LightX2V** | H100 | **8 cards + no cfg + fp8** | **0.35s/it** | **2.1x** 🚀 |
| **LightX2V** | 4090D | 8 cards + cfg | 4.75s/it | 1x |
| **LightX2V** | 4090D | 8 cards + no cfg | 3.13s/it | 1.5x |
| **LightX2V** | 4090D | **8 cards + no cfg + fp8** | **2.35s/it** | **2.0x** 🚀 |

**Note**: All the above performance data were tested on Wan2.1-I2V-14B-480P(40 steps, 81 frames). In addition, we also provide a 4-step distilled model on the [HuggingFace page](https://huggingface.co/lightx2v).


helloyongyang's avatar
helloyongyang committed
72
73
## 💡 快速开始

Gu Shiqiao's avatar
Gu Shiqiao committed
74
75
> 🌐 **立即在线体验!** 无需安装即可体验 LightX2V:**[LightX2V 在线服务](https://x2v.light-ai.top/login)** - 免费、轻量、快速的AI数字人视频生成平台。

helloyongyang's avatar
helloyongyang committed
76
77
详细使用说明请参考我们的文档:**[英文文档](https://lightx2v-en.readthedocs.io/en/latest/) | [中文文档](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/)**

78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
### 从 Git 安装
```bash
pip install -v git+https://github.com/ModelTC/LightX2V.git
```

### 从源码构建
```bash
git clone https://github.com/ModelTC/LightX2V.git
cd LightX2V
uv pip install -v . # pip install -v .
```

### (可选)安装注意力/量化算子
注意力算子安装说明请参考我们的文档:**[英文文档](https://lightx2v-en.readthedocs.io/en/latest/getting_started/quickstart.html#step-4-install-attention-operators) | [中文文档](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/getting_started/quickstart.html#id9)**

Gu Shiqiao's avatar
Gu Shiqiao committed
93
### 使用示例
94
```python
gushiqiao's avatar
gushiqiao committed
95
96
97
98
99
100
# examples/wan/wan_i2v.py
"""
Wan2.2 image-to-video generation example.
This example demonstrates how to use LightX2V with Wan2.2 model for I2V generation.
"""

101
102
from lightx2v import LightX2VPipeline

gushiqiao's avatar
gushiqiao committed
103
104
# Initialize pipeline for Wan2.2 I2V task
# For wan2.1, use model_cls="wan2.1"
105
pipe = LightX2VPipeline(
gushiqiao's avatar
gushiqiao committed
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
    model_path="/path/to/Wan2.2-I2V-A14B",
    model_cls="wan2.2_moe",
    task="i2v",
)

# Alternative: create generator from config JSON file
# pipe.create_generator(
#     config_json="configs/wan22/wan_moe_i2v.json"
# )

# Enable offloading to significantly reduce VRAM usage with minimal speed impact
# Suitable for RTX 30/40/50 consumer GPUs
pipe.enable_offload(
    cpu_offload=True,
    offload_granularity="block",  # For Wan models, supports both "block" and "phase"
    text_encoder_offload=True,
    image_encoder_offload=False,
    vae_offload=False,
124
125
)

gushiqiao's avatar
gushiqiao committed
126
# Create generator manually with specified parameters
127
128
pipe.create_generator(
    attn_mode="sage_attn2",
gushiqiao's avatar
gushiqiao committed
129
130
131
132
133
134
    infer_steps=40,
    height=480,  # Can be set to 720 for higher resolution
    width=832,  # Can be set to 1280 for higher resolution
    num_frames=81,
    guidance_scale=[3.5, 3.5],  # For wan2.1, guidance_scale is a scalar (e.g., 5.0)
    sample_shift=5.0,
135
136
)

gushiqiao's avatar
gushiqiao committed
137
138
139
140
141
142
# Generation parameters
seed = 42
prompt = "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline's intricate details and the refreshing atmosphere of the seaside."
negative_prompt = "镜头晃动,色调艳丽,过曝,静态,细节模糊不清,字幕,风格,作品,画作,画面,静止,整体发灰,最差质量,低质量,JPEG压缩残留,丑陋的,残缺的,多余的手指,画得不好的手部,画得不好的脸部,畸形的,毁容的,形态畸形的肢体,手指融合,静止不动的画面,杂乱的背景,三条腿,背景人很多,倒着走"
image_path="/path/to/img_0.jpg"
save_result_path = "/path/to/save_results/output.mp4"
143

gushiqiao's avatar
gushiqiao committed
144
# Generate video
145
146
pipe.generate(
    seed=seed,
gushiqiao's avatar
gushiqiao committed
147
    image_path=image_path,
148
149
150
151
152
153
154
155
156
    prompt=prompt,
    negative_prompt=negative_prompt,
    save_result_path=save_result_path,
)

```

> 💡 **更多示例**: 更多使用案例,包括量化、卸载、缓存等进阶配置,请参考 [examples 目录](https://github.com/ModelTC/LightX2V/tree/main/examples)。

gushiqiao's avatar
gushiqiao committed
157
## 🤖 支持的模型生态
Harahan's avatar
Harahan committed
158

gushiqiao's avatar
gushiqiao committed
159
### 官方开源模型
160
-[HunyuanVideo-1.5](https://huggingface.co/tencent/HunyuanVideo-1.5)
helloyongyang's avatar
helloyongyang committed
161
-[Wan2.1 & Wan2.2](https://huggingface.co/Wan-AI/)
gushiqiao's avatar
gushiqiao committed
162
163
-[Qwen-Image](https://huggingface.co/Qwen/Qwen-Image)
-[Qwen-Image-Edit](https://huggingface.co/spaces/Qwen/Qwen-Image-Edit)
Watebear's avatar
Watebear committed
164
-[Qwen-Image-Edit-2509](https://huggingface.co/Qwen/Qwen-Image-Edit-2509)
Harahan's avatar
Harahan committed
165

gushiqiao's avatar
gushiqiao committed
166
167
168
169
170
### 量化模型和蒸馏模型/Lora (**🚀 推荐:4步推理**)
-[Wan2.1-Distill-Models](https://huggingface.co/lightx2v/Wan2.1-Distill-Models)
-[Wan2.2-Distill-Models](https://huggingface.co/lightx2v/Wan2.2-Distill-Models)
-[Wan2.1-Distill-Loras](https://huggingface.co/lightx2v/Wan2.1-Distill-Loras)
-[Wan2.2-Distill-Loras](https://huggingface.co/lightx2v/Wan2.2-Distill-Loras)
gushiqiao's avatar
gushiqiao committed
171

172
173
### 轻量级自编码器模型(**🚀 推荐:推理快速 + 内存占用低**)
-[Autoencoders](https://huggingface.co/lightx2v/Autoencoders)
helloyongyang's avatar
helloyongyang committed
174

gushiqiao's avatar
gushiqiao committed
175
176
### 自回归模型
-[Wan2.1-T2V-CausVid](https://huggingface.co/lightx2v/Wan2.1-T2V-14B-CausVid)
gushiqiao's avatar
gushiqiao committed
177
-[Self-Forcing](https://github.com/guandeh17/Self-Forcing)
Watebear's avatar
Watebear committed
178
-[Matrix-Game-2.0](https://huggingface.co/Skywork/Matrix-Game-2.0)
gushiqiao's avatar
gushiqiao committed
179

gushiqiao's avatar
gushiqiao committed
180
181
🔔 可以关注我们的[HuggingFace主页](https://huggingface.co/lightx2v),及时获取我们团队的模型。

gushiqiao's avatar
gushiqiao committed
182
💡 参考[模型结构文档](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/getting_started/model_structure.html)快速上手 LightX2V
gushiqiao's avatar
gushiqiao committed
183

gushiqiao's avatar
gushiqiao committed
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
## 🚀 前端展示

我们提供了多种前端界面部署方式:

- **🎨 Gradio界面**: 简洁易用的Web界面,适合快速体验和原型开发
  - 📖 [Gradio部署文档](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/deploy_gradio.html)
- **🎯 ComfyUI界面**: 强大的节点式工作流界面,支持复杂的视频生成任务
  - 📖 [ComfyUI部署文档](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/deploy_comfyui.html)
- **🚀 Windows一键部署**: 专为Windows用户设计的便捷部署方案,支持自动环境配置和智能参数优化
  - 📖 [Windows一键部署文档](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/deploy_local_windows.html)

**💡 推荐方案**:
- **首次使用**: 建议选择Windows一键部署方案
- **高级用户**: 推荐使用ComfyUI界面获得更多自定义选项
- **快速体验**: Gradio界面提供最直观的操作体验

gushiqiao's avatar
gushiqiao committed
200
201
202
## 🚀 核心特性

### 🎯 **极致性能优化**
gushiqiao's avatar
gushiqiao committed
203
- **🔥 SOTA推理速度**: 通过步数蒸馏和系统优化实现**20倍**极速加速(单GPU)
gushiqiao's avatar
gushiqiao committed
204
205
206
207
208
209
210
211
212
213
214
215
216
- **⚡️ 革命性4步蒸馏**: 将原始40-50步推理压缩至仅需4步,且无需CFG配置
- **🛠️ 先进算子支持**: 集成顶尖算子,包括[Sage Attention](https://github.com/thu-ml/SageAttention)[Flash Attention](https://github.com/Dao-AILab/flash-attention)[Radial Attention](https://github.com/mit-han-lab/radial-attention)[q8-kernel](https://github.com/KONAKONA666/q8_kernels)[sgl-kernel](https://github.com/sgl-project/sglang/tree/main/sgl-kernel)[vllm](https://github.com/vllm-project/vllm)

### 💾 **资源高效部署**
- **💡 突破硬件限制**: **仅需8GB显存 + 16GB内存**即可运行14B模型生成480P/720P视频
- **🔧 智能参数卸载**: 先进的磁盘-CPU-GPU三级卸载架构,支持阶段/块级别的精细化管理
- **⚙️ 全面量化支持**: 支持`w8a8-int8``w8a8-fp8``w4a4-nvfp4`等多种量化策略

### 🎨 **丰富功能生态**
- **📈 智能特征缓存**: 智能缓存机制,消除冗余计算,提升效率
- **🔄 并行推理加速**: 多GPU并行处理,显著提升性能表现
- **📱 灵活部署选择**: 支持Gradio、服务化部署、ComfyUI等多种部署方式
- **🎛️ 动态分辨率推理**: 自适应分辨率调整,优化生成质量
gushiqiao's avatar
gushiqiao committed
217
- **🎞️ 视频帧插值**: 基于RIFE的帧插值技术,实现流畅的帧率提升
gushiqiao's avatar
gushiqiao committed
218
219


gushiqiao's avatar
gushiqiao committed
220
## 📚 技术文档
Harahan's avatar
Harahan committed
221

gushiqiao's avatar
gushiqiao committed
222
223
224
225
226
227
### 📖 **方法教程**
- [模型量化](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/quantization.html) - 量化策略全面指南
- [特征缓存](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/cache.html) - 智能缓存机制详解
- [注意力机制](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/attention.html) - 前沿注意力算子
- [参数卸载](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/offload.html) - 三级存储架构
- [并行推理](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/parallel.html) - 多GPU加速策略
helloyongyang's avatar
helloyongyang committed
228
- [变分辨率推理](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/changing_resolution.html) - U型分辨率策略
gushiqiao's avatar
gushiqiao committed
229
- [步数蒸馏](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/step_distill.html) - 4步推理技术
PengGao's avatar
PengGao committed
230
- [视频帧插值](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/video_frame_interpolation.html) - 基于RIFE的帧插值技术
gushiqiao's avatar
gushiqiao committed
231
232
233
234
235
236

### 🛠️ **部署指南**
- [低资源场景部署](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/for_low_resource.html) - 优化的8GB显存解决方案
- [低延迟场景部署](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/for_low_latency.html) - 极速推理优化
- [Gradio部署](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/deploy_gradio.html) - Web界面搭建
- [服务化部署](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/deploy_service.html) - 生产级API服务部署
helloyongyang's avatar
helloyongyang committed
237
- [Lora模型部署](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/deploy_guides/lora_deploy.html) - Lora灵活部署
gushiqiao's avatar
gushiqiao committed
238
239
240
241
242
243
244

## 🧾 代码贡献指南

我们通过自动化的预提交钩子来保证代码质量,确保项目代码格式的一致性。

> [!TIP]
> **安装说明:**
Harahan's avatar
Harahan committed
245
>
gushiqiao's avatar
gushiqiao committed
246
> 1. 安装必要的依赖:
Harahan's avatar
Harahan committed
247
248
> ```shell
> pip install ruff pre-commit
gushiqiao's avatar
gushiqiao committed
249
> ```
Harahan's avatar
Harahan committed
250
>
gushiqiao's avatar
gushiqiao committed
251
> 2. 提交前运行:
Harahan's avatar
Harahan committed
252
253
> ```shell
> pre-commit run --all-files
gushiqiao's avatar
gushiqiao committed
254
> ```
Harahan's avatar
Harahan committed
255

gushiqiao's avatar
gushiqiao committed
256
感谢您为LightX2V的改进做出贡献!
Harahan's avatar
Harahan committed
257
258
259

## 🤝 致谢

gushiqiao's avatar
gushiqiao committed
260
我们向所有启发和促进LightX2V开发的模型仓库和研究社区表示诚挚的感谢。此框架基于开源社区的集体努力而构建。
Harahan's avatar
Harahan committed
261

gushiqiao's avatar
gushiqiao committed
262
## 🌟 Star 历史
Harahan's avatar
Harahan committed
263
264
265
266
267

[![Star History Chart](https://api.star-history.com/svg?repos=ModelTC/lightx2v&type=Timeline)](https://star-history.com/#ModelTC/lightx2v&Timeline)

## ✏️ 引用

gushiqiao's avatar
gushiqiao committed
268
如果您发现LightX2V对您的研究有用,请考虑引用我们的工作:
Harahan's avatar
Harahan committed
269
270
271

```bibtex
@misc{lightx2v,
gushiqiao's avatar
gushiqiao committed
272
 author = {LightX2V Contributors},
helloyongyang's avatar
helloyongyang committed
273
 title = {LightX2V: Light Video Generation Inference Framework},
gushiqiao's avatar
gushiqiao committed
274
275
276
277
 year = {2025},
 publisher = {GitHub},
 journal = {GitHub repository},
 howpublished = {\url{https://github.com/ModelTC/lightx2v}},
Harahan's avatar
Harahan committed
278
279
}
```
gushiqiao's avatar
gushiqiao committed
280
281
282
283
284
285
286
287
288
289
290

## 📞 联系与支持

如有任何问题、建议或需要支持,欢迎通过以下方式联系我们:
- 🐛 [GitHub Issues](https://github.com/ModelTC/lightx2v/issues) - 错误报告和功能请求

---

<div align="center">
由 LightX2V 团队用 ❤️ 构建
</div>