Qwen-Image is the first image generation foundation model released by Alibaba’s Qwen team. It’s a 20B parameter MMDiT (Multimodal Diffusion Transformer) model open-sourced under the Apache 2.0 license. The model has made significant advances in complex text rendering and precise image editing, achieving high-fidelity output for multiple languages including English and Chinese. Model Highlights:
  • Excellent Multilingual Text Rendering: Supports high-precision text generation in multiple languages including English, Chinese, Korean, Japanese, maintaining font details and layout consistency
  • Diverse Artistic Styles: From photorealistic scenes to impressionist paintings, from anime aesthetics to minimalist design, fluidly adapting to various creative prompts
Related Links: Currently Qwen-Image has multiple ControlNet support options available:

Qwen-Image Native Workflow Example

Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated.(Desktop version’s update will delay sometime)If nodes are missing when loading a workflow, possible reasons:
  1. Not using the latest ComfyUI version(Nightly version)
  2. Using Stable or Desktop version (Latest changes may not be included)
  3. Some nodes failed to import at startup
There are three different models used in the workflow attached to this document:
  1. Qwen-Image original model fp8_e4m3fn
  2. 8-step accelerated version: Qwen-Image original model fp8_e4m3fn with lightx2v 8-step LoRA
  3. Distilled version: Qwen-Image distilled model fp8_e4m3fn
VRAM Usage Reference GPU: RTX4090D 24GB
Model UsedVRAM UsageFirst GenerationSecond Generation
fp8_e4m3fn86%≈ 94s≈ 71s
fp8_e4m3fn with lightx2v 8-step LoRA86%≈ 55s≈ 34s
Distilled fp8_e4m3fn86%≈ 69s≈ 36s

1. Workflow File

After updating ComfyUI, you can find the workflow file in the templates, or drag the workflow below into ComfyUI to load it. Qwen-image Text-to-Image Workflow

Download Workflow for Qwen-Image Official Model

Distilled version

Download Workflow for Distilled Model

2. Model Download

Available Models in ComfyUI
  • Qwen-Image_bf16 (40.9 GB)
  • Qwen-Image_fp8 (20.4 GB)
  • Distilled versions (non-official, requires only 15 steps)
All models are available at Huggingface and Modelscope Diffusion model Qwen_image_distill
  • The original author of the distilled version recommends using 15 steps with cfg 1.0.
  • According to tests, this distilled version also performs well at 10 steps with cfg 1.0. You can choose either euler or res_multistep based on the type of image you want.
LoRA Text encoder VAE qwen_image_vae.safetensors Model Storage Location
📂 ComfyUI/
├── 📂 models/
│   ├── 📂 diffusion_models/
│   │   ├── qwen_image_fp8_e4m3fn.safetensors
│   │   └── qwen_image_distill_full_fp8_e4m3fn.safetensors ## 蒸馏版
│   ├── 📂 loras/
│   │   └── Qwen-Image-Lightning-8steps-V1.0.safetensors   ## 8步加速 LoRA 模型
│   ├── 📂 vae/
│   │   └── qwen_image_vae.safetensors
│   └── 📂 text_encoders/
│       └── qwen_2.5_vl_7b_fp8_scaled.safetensors

3. Complete the Workflow Step by Step

Step Guide
  1. Make sure the Load Diffusion Model node has loaded qwen_image_fp8_e4m3fn.safetensors
  2. Make sure the Load CLIP node has loaded qwen_2.5_vl_7b_fp8_scaled.safetensors
  3. Make sure the Load VAE node has loaded qwen_image_vae.safetensors
  4. Make sure the EmptySD3LatentImage node is set with the correct image dimensions
  5. Set your prompt in the CLIP Text Encoder node; currently, it supports at least English, Chinese, Korean, Japanese, Italian, etc.
  6. If you want to enable the 8-step acceleration LoRA by lightx2v, select the node and use Ctrl + B to enable it, and modify the Ksampler settings as described in step 8
  7. Click the Queue button, or use the shortcut Ctrl(cmd) + Enter to run the workflow
  8. For different model versions and workflows, adjust the KSampler parameters accordingly
The distilled model and the 8-step acceleration LoRA by lightx2v do not seem to be compatible for simultaneous use. You can experiment with different combinations to verify if they can be used together.

Qwen Image ControlNet DiffSynth-ControlNets Model Patches Workflow

This model is actually not a ControlNet, but a Model patch that supports three different control modes: canny, depth, and inpaint. Original model address: DiffSynth-Studio/Qwen-Image ControlNet Comfy Org rehost address: Qwen-Image-DiffSynth-ControlNets/model_patches

1. Workflow and Input Images

Download the image below and drag it into ComfyUI to load the corresponding workflow workflow

Download JSON Format Workflow

Download the image below as input: input Other models are the same as the Qwen-Image basic workflow. You only need to download the models below and save them to the ComfyUI/models/model_patches folder

3. Workflow Usage Instructions

Currently, diffsynth has three patch models: Canny, Depth, and Inpaint. If you’re using ControlNet-related workflows for the first time, you need to understand that control images need to be preprocessed into supported image formats before they can be used and recognized by the model. Input Type Diagram
  • Canny: Processed canny edge, line art contours
  • Depth: Preprocessed depth map showing spatial relationships
  • Inpaint: Requires using Mask to mark areas that need to be repainted
Since this patch model is divided into three different models, you need to select the correct preprocessing type when inputting to ensure proper image preprocessing. Canny Model ControlNet Usage Instructions Canny Workflow
  1. Ensure that qwen_image_canny_diffsynth_controlnet.safetensors is loaded
  2. Upload input image for subsequent processing
  3. The Canny node is a native preprocessing node that will preprocess the input image according to your set parameters to control generation
  4. If needed, you can modify the strength in the QwenImageDiffsynthControlnet node to control the intensity of line art control
  5. Click the Run button, or use the shortcut Ctrl(cmd) + Enter to run the workflow
For using qwen_image_depth_diffsynth_controlnet.safetensors, you need to preprocess the image into a depth map and replace the image processing part. For this usage, please refer to the InstantX processing method in this document. Other parts are similar to using the Canny model.
Inpaint Model ControlNet Usage Instructions Inpaint Workflow For the Inpaint model, it requires using the Mask Editor to draw a mask and use it as input control condition.
  1. Ensure that ModelPatchLoader loads the qwen_image_inpaint_diffsynth_controlnet.safetensors model
  2. Upload image and use the Mask Editor to draw a mask. You need to connect the mask output of the corresponding Load Image node to the mask input of QwenImageDiffsynthControlnet to ensure the corresponding mask is loaded
  3. Use the Ctrl-B shortcut to set the original Canny in the workflow to bypass mode, making the corresponding Canny node processing ineffective
  4. In CLIP Text Encoder, input what you want to change the masked area to
  5. If needed, you can modify the strength in the QwenImageDiffsynthControlnet node to control the corresponding control intensity
  6. Click the Run button, or use the shortcut Ctrl(cmd) + Enter to run the workflow

Qwen Image Union ControlNet LoRA Workflow

Original model address: DiffSynth-Studio/Qwen-Image-In-Context-Control-Union Comfy Org rehost address: qwen_image_union_diffsynth_lora.safetensors: Image structure control LoRA supporting canny, depth, pose, lineart, softedge, normal, openpose

1. Workflow and Input Images

Download the image below and drag it into ComfyUI to load the workflow workflow

Download JSON Format Workflow

Download the image below as input workflow Download the model below. Since this is a LoRA model, it needs to be saved to the ComfyUI/models/loras/ folder

3. Workflow Instructions

This model is a unified control LoRA that supports canny, depth, pose, lineart, softedge, normal, openpose controls. Since many image preprocessing native nodes are not fully supported, you should use something like comfyui_controlnet_aux to complete other image preprocessing. Union Control LoRA
  1. Ensure that LoraLoaderModelOnly correctly loads the qwen_image_union_diffsynth_lora.safetensors model
  2. Upload input image
  3. If needed, you can adjust the Canny node parameters. Since different input images require different parameter settings to get better image preprocessing results, you can try adjusting the corresponding parameter values to get more/fewer details
  4. Click the Run button, or use the shortcut Ctrl(cmd) + Enter to run the workflow
For other types of control, you also need to replace the image processing part.

Qwen Image InstantX ControlNet Workflow

[To be updated]