What Is Qwen-Image-Lightning?

Qwen-Image-Lightning is a distilled version of the Qwen-Image model, fine-tuned to maintain image quality while drastically reducing the number of inference steps required.

The repo (https://github.com/ModelTC/Qwen-Image-Lightning) currently offers:

• Qwen-Image-Lightning-8steps-V1.1 (bf16)

• Qwen-Image-Lightning-4steps-V1.0

Both variants are built to save time and compute without compromising too much on fidelity — ideal for both quick drafts and production work.

Why distillation matters:
The base Qwen-Image model is already powerful, but it can take 20–30 steps for optimal results. Distillation compresses the generation process, giving you near-identical quality at over 50% faster speed.

Qwen-Image-Lightning’s GitHub repo includes ready-made ComfyUI workflow JSON files:

• workflows/qwen-image-8steps.json

• workflows/qwen-image-4steps.json

These are pre-configured to get you up and running with minimal setup.

How to Use Qwen-Image-Lightning in ComfyUI

1️⃣ Install or Update ComfyUI

Make sure you’re running a recent (preferably nightly) build for best compatibility. Updating ensures LoRA loading works properly — outdated versions may throw “lora key not loaded” errors.

2️⃣ Download the Loras

https://civarchive.com/models/1854805/qwen-lightning-lora?modelVersionId=2106476

3️⃣ Place the Files Correctly

In your ComfyUI directory:

• LoRA → models/loras/

4️⃣ Load the Workflow

1. Open ComfyUI.

2. Drag the JSON workflow file (qwen-image-8steps.json or qwen-image-4steps.json) into the canvas.

3. Double-check that the KSampler node matches the correct number of steps (8 or 4).

4. Adjust your prompt and settings.

5. Hit Queue Prompt — and watch the magic happen.

Tips & Notes

• Don’t combine the distilled model with the 8-step LoRA — they’re alternative speed-up methods, not additive.

• If you encounter “LoRA key not loaded” errors, update to ComfyUI’s nightly build.

• Keep your VRAM in mind — while steps are reduced, memory usage is similar to the base model.