Qwen Image 2.1 Turbo GGUF
Qwen Image 2.1 Turbo is an accelerated diffusion transformer for text-to-image generation and image editing. It uses the Qwen3-VL 8B text encoder and is designed for 8 denoising steps. ComfyUI workflows can also pass a reference image for image editing.
Qwen-Image-2.1-Turbo-GGUF on Hugging FaceExample output
Prompt: A snowy city street at night, with a person walking a dog, warm lights in the windows and on the road, and the words “Qwen Image 2.1 Turbo GGUF” in a speech bubble.
Available Quantizations
| Quant | Size | File |
|---|---|---|
| Q4_0 | 3.91 GB | qwen_image_2.1_turbo_Q4_0.gguf |
| Q4_K_M | 4.17 GB | qwen_image_2.1_turbo_Q4_K_M.gguf |
| Q6_K | 5.59 GB | qwen_image_2.1_turbo_Q6_K.gguf |
| Q8_0 | 7.16 GB | qwen_image_2.1_turbo_Q8_0.gguf |
| Q8_CR | 6.76 GB | qwen_image_2.1_turbo_Q8_CR.gguf |
Q4_0, Q4_K_M, Q6_K, and Q8_0 use standard GGML quantization. Q8_CR is a custom GGUF INT8 ConvRot format supported by comfyui-gguf-reboot. It uses the loader's native INT8 ConvRot execution path and is distinct from standard GGML Q8_0.
Model Input and Output
Inputs
| Input | Description |
|---|---|
| Text prompt | A prompt for image generation or editing instructions. |
| Reference image | Optional image input for editing and visual conditioning. |
Outputs
RGBA image output with transparency support.
ComfyUI Setup
Load GGUF files with the comfyui-gguf-reboot loader nodes. These nodes give the best performance and are required for the Q8_CR native INT8 ConvRot path.
Dependencies
The GGUF repository contains the diffusion model files. Get the matching text encoder and VAE from the upstream Qwen Image 2.1 release.
| Component | Example file | Folder |
|---|---|---|
| Qwen Image 2.1 diffusion model | One of the GGUF files above | models/diffusion_models/ or models/unet/ |
| Qwen3-VL 8B text encoder | qwen3vl_8b_fp8_scaled.safetensors | models/clip/ or text_encoders/ |
| Qwen Image VAE | qwen_image_2.1_vae_bf16.safetensors | models/vae/ |
This repository has no workflow JSON files. Start with the upstream Qwen Image 2.1 ComfyUI workflow and use the comfyui-gguf-reboot loader node for the diffusion model.