Qwen-Image-Edit-2511-Lightning

Introduction

Qwen-Image-Edit-2511-Lightning is a collection of optimized LoRA models designed for image editing. It uses step distillation and quantization to deliver high-efficiency inference, making it much faster and more resource-friendly for various tasks.

How to Use

• Trigger word: No trigger word needed.
• Recommended settings: This model suite supports two main frameworks. For detailed usage, environment setup, inference pipelines, and customization, refer to their specific documentation:
  • Qwen-Image-Lightning GitHub Repository
  • LightX2V Qwen Image Edit Documentation

FAQ

Q: What's the trigger word?

A: No trigger word required.

Q: What LoRA strength should I use?

A: Around 1 works well. Adjust based on how strong you want the effect.

Q: What are the main optimizations in these models?

A:

• Step Distillation: The LoRA models reduce inference steps to just 4, making them about 10x faster than standard 40-step inference while keeping image editing quality.
• FP8 Quantization: The quantized base model balances performance and resource efficiency. It cuts GPU memory usage by roughly 50% compared to FP32, maintaining editing fidelity.

Q: What different model files are available?

A: There are three core files:

• Qwen-Image-Edit-2511-Lightning-4steps-V1.0-bf16.safetensors: A lightweight, 4-step distilled LoRA in BF16 precision.
• Qwen-Image-Edit-2511-Lightning-4steps-V1.0-fp32.safetensors: A 4-step distilled LoRA in FP32 precision for higher accuracy.
• qwen_image_edit_2511_fp8_e4m3fn_scaled_lightning.safetensors: An FP8 quantized model, fused with the 4-step distilled LoRA, optimized for low-memory deployment.

Q: Where can I get support or ask questions?

A: For technical issues or feature requests, open an issue on the relevant GitHub repo:

• Qwen-Image-Lightning repo (for Qwen framework questions)
• LightX2V repo (for LightX2V integration questions)

Technical Details

• Base model: Qwen/Qwen-Image-Edit-2511
• Training info: Optimized using step distillation and quantization techniques for efficiency.
• Source: Hugging Face