My Favorite Way to Use Qwen Image 2.1

I could write every prompt like a tiny novel—or let Qwen do the overthinking.

This is my favorite way to use Qwen Image 2.1. The Qwen 8B prompt enhancer expands simple ideas into detailed prompts, while the Viggle Turbo LoRA helps generate images quickly.

It supports both text-to-image (T2I) and image-to-image (I2I). In my experience, it produces great images and handles editing very well. I use it on an NVIDIA GeForce RTX 5080.

Workflow Features

Automatic prompt enhancement

Qwen Image 2.1 with Viggle Turbo LoRA

Support for T2I and I2I

Good image quality and editing results

Optional additional LoRAs

8 steps, CFG 2.0 - 3.0

res_2s sampler with beta scheduler

Viggle LoRA strength: 0.5

Required Models

Place qwen_image_2.1_bf16.safetensors in:

ComfyUI/models/diffusion_models/

Place qwen3vl_8b_bf16.safetensors and qwen3VL8BHereticINT8W4A8_int8.safetensors in:

ComfyUI/models/text_encoders/

The second model is used by the prompt enhancer.

Place qwen_image_2.1_vae_bf16.safetensors in:

ComfyUI/models/vae/

Place Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r256.safetensors in:

ComfyUI/models/loras/Qwen 2.1/

Viggle Turbo Loader

Download viggle_turbo.py from the comfyui folder on the official Viggle Hugging Face page.

Copy it to:

ComfyUI/custom_nodes/

Then restart ComfyUI.

If additional nodes are missing, use ComfyUI Manager → Install Missing Custom Nodes. The workflow also uses rgthree-comfy, Pixaroma nodes, and Qwen Image 2.1/Viggle custom nodes.

This is simply my preferred way to use Qwen Image 2.1—not the only correct setup. Let Qwen do the prompt engineering while you take the credit.