---
title: Qwen-Image-Lightning
canonical_url: "https://www.modelscope.cn/models/lightx2v/Qwen-Image-Lightning"
md_url: "https://www.modelscope.cn/models/lightx2v/Qwen-Image-Lightning.md"
repository: lightx2v/Qwen-Image-Lightning
last_updated: 2026-10-19
license: apache-2.0
pipeline_tag: text-to-image-synthesis
tasks:
  - text-to-image-synthesis
base_model:
  - Qwen/Qwen-Image
base_model_relation: adapter
parameters: 48.9B
tensor_type:
  - F8_E4M3
  - I64
  - F32
  - BF16
library_name:
  - pytorch
  - lora
  - safetensors
frameworks:
  - pytorch
language:
  - en
  - zh
supports_inference: txt2img
downloads: 45988
stars: 76
tags:
  - Qwen-Image
  - distillation
  - LoRA
  - lora
---

# Qwen-Image-Lightning

> Qwen-Image-Lightning - lightx2v 在 ModelScope 开源的模型。Qwen-Image-Lightning

lightx2v/Qwen-Image-Lightning 是 ModelScope 魔搭社区上的 48.9B 参数text-to-image-synthesis模型，采用 apache-2.0 许可，基于 Qwen/Qwen-Image 构建，并支持在线推理（txt2img）。

- **Repository**: lightx2v/Qwen-Image-Lightning
- **License**: apache-2.0
- **Tasks**: text-to-image-synthesis
- **Parameters**: 48.9B
- **Base model**: Qwen/Qwen-Image
- **Online inference**: txt2img
- **Tags**: Qwen-Image, distillation, LoRA, lora
- **Downloads**: 45988
- **Stars**: 76
- **Last updated**: 2026-10-19

Source: https://www.modelscope.cn/models/lightx2v/Qwen-Image-Lightning

---

# Qwen-Image-Lightning

Qwen-Image-Lightning provides distilled LoRA weights for Qwen-Image,
enabling image generation with fewer inference steps.

**[Create Online — LightX2V Studio](https://x2v.light-ai.top/)**
· **[GitHub — Qwen-Image-Lightning](https://github.com/ModelTC/Qwen-Image-Lightning/)**
· **[GitHub — LightX2V Framework](https://github.com/ModelTC/LightX2V)**

## Method 1: LightX2V Studio — Online Image Creation

[**LightX2V Studio**](https://x2v.light-ai.top/) is an online AI avatar
and multimodal content production platform built on the LightX2V video
generation inference framework. With AI avatar video creation at its
core, it supports video, image, and audio creation, plus workflow
orchestration and reuse.

Studio supports **text-to-image** and **image-to-image** creation
directly in your browser, without local installation or GPU setup.

### Quick Start

1. Open [LightX2V Studio](https://x2v.light-ai.top/), sign in, and
   choose text-to-image or image-to-image.
2. Enter a prompt, upload an image if required, and configure settings.
3. Review the credit cost, submit, then preview and download the result.

### Current Online Model and LoRA

- **Model:** Qwen-Image-2512
- **LoRA:** Lightning 8-step V1.0, FP32
- **Source:** [lightx2v/Qwen-Image-2512-Lightning](https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning)

> The online deployment differs from the original Qwen-Image setup
> shown below. Available models and versions may change over time.

[Open LightX2V Studio](https://x2v.light-ai.top/)
· [Pricing](https://x2v.light-ai.top/pricing)

## Method 2: Local Inference with Diffusers

For model versions, usage instructions, workflows, and evaluation
results, see the
[Qwen-Image-Lightning GitHub repository](https://github.com/ModelTC/Qwen-Image-Lightning/).

For inference and deployment with the LightX2V framework, see the
[LightX2V GitHub repository](https://github.com/ModelTC/LightX2V).

The following example uses
`Qwen-Image-Lightning-8steps-V1.0.safetensors`.

use with diffusers 🧨:

make sure to install diffusers from `main` (`pip install git+https://github.com/huggingface/diffusers.git`)
```
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
import torch 
import math

# From https://github.com/ModelTC/Qwen-Image-Lightning/blob/342260e8f5468d2f24d084ce04f55e101007118b/generate_with_diffusers.py#L82C9-L97C10
scheduler_config = {
    "base_image_seq_len": 256,
    "base_shift": math.log(3),  # We use shift=3 in distillation
    "invert_sigmas": False,
    "max_image_seq_len": 8192,
    "max_shift": math.log(3),  # We use shift=3 in distillation
    "num_train_timesteps": 1000,
    "shift": 1.0,
    "shift_terminal": None,  # set shift_terminal to None
    "stochastic_sampling": False,
    "time_shift_type": "exponential",
    "use_beta_sigmas": False,
    "use_dynamic_shifting": True,
    "use_exponential_sigmas": False,
    "use_karras_sigmas": False,
}
scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
pipe = DiffusionPipeline.from_pretrained(
    "Qwen/Qwen-Image", scheduler=scheduler, torch_dtype=torch.bfloat16
).to("cuda")
pipe.load_lora_weights(
    "lightx2v/Qwen-Image-Lightning", weight_name="Qwen-Image-Lightning-8steps-V1.0.safetensors"
)

prompt = "a tiny astronaut hatching from an egg on the moon, Ultra HD, 4K, cinematic composition."
negative_prompt = " "
image = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=1024,
    height=1024,
    num_inference_steps=8,
    true_cfg_scale=1.0,
    generator=torch.manual_seed(0),
).images[0]
image.save("qwen_fewsteps.png")
```

## Resources

| Resource | Purpose |
| --- | --- |
| [LightX2V Studio](https://x2v.light-ai.top/) | Online AI avatar and multimodal content production |
| [Qwen-Image-Lightning on GitHub](https://github.com/ModelTC/Qwen-Image-Lightning/) | Model usage, workflows, evaluation results, and issue tracking |
| [LightX2V on GitHub](https://github.com/ModelTC/LightX2V) | Open-source inference framework and deployment |
| [Studio API & Skills](https://x2v.light-ai.top/api-docs) | Programmatic access to supported online creation tasks |