Qwen-Image-Edit

Qwen-Image-Edit

Edits an existing picture from a written instruction, changing objects, style or the text inside the image in Chinese and English, for designers and developers working with the diffusers library.

Made by Qwen on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Qwen-Image-Edit is the image-editing version of the 20B Qwen-Image model. You give it a picture and an instruction, and it returns an edited picture. It extends Qwen-Image's skill with text inside images, so it can add, remove or change words while keeping the original font, size and style, in both Chinese and English.

Who it suits: It suits designers, content teams and developers who want to change images by describing the change, instead of editing by hand.

Title: Who uses this model
Industry | Who uses it | How they use it | Example
Design | Graphic designers | Change colors, objects or style in a picture | Turn a product shot purple with a new background
Marketing | Content teams | Edit text on posters and banners | Swap a headline in an image and keep its look
Software | Developers | Add image editing to an app | Call the pipeline from a Python script
Character and brand work | Creators | Produce new pictures of the same character | Rotate a mascot or restyle it

If you only need to make images from scratch, you can skip it, since this model starts from an existing picture.

Where to use it:

• Python scripts that use the diffusers library
• A machine with an Nvidia GPU, as in the card's sample
• Windows, macOS or Linux with Python

Title: Jobs it does
Use case | Who | What they get back
Appearance edits | Designers | A picture with one element added, removed or changed and the rest left alone
Semantic edits | Creators | A restyled or rotated version that keeps the character recognizable
Text edits | Marketers | An image with its words changed in the same font and style

How to set it up: You need Python, a recent diffusers and a GPU with enough memory for a 20B model. The weights are about 40.9 GB in BF16, or about 10.2 GB for a 4-bit copy by Hugging Face's estimate.

• Python with pip
• The latest diffusers, installed from its source repository as the card advises
• The PIL and torch packages used by the sample
• A GPU, since the sample moves the pipeline to cuda

1. Install the latest diffusers as the card describes.
2. Optionally save the model files first with the two Hugging Face commands.
3. Put the picture you want to edit next to the script, and save the sample script.
4. Run it, and the edited picture is saved as output_image_edit.png.

pip install -U huggingface_hub
hf download Qwen/Qwen-Image-Edit

How to use it: Load your picture, write the change you want as a sentence, and run the pipeline. The sample uses a fixed seed so you can repeat a result, 50 inference steps and a guidance value of 4.0.

import os
from PIL import Image
import torch

from diffusers import QwenImageEditPipeline

pipeline = QwenImageEditPipeline.from_pretrained("Qwen/Qwen-Image-Edit")
print("pipeline loaded")
pipeline.to(torch.bfloat16)
pipeline.to("cuda")
pipeline.set_progress_bar_config(disable=None)
image = Image.open("./input.png").convert("RGB")
prompt = "Change the rabbit's color to purple, with a flash light background."
inputs = {
    "image": image,
    "prompt": prompt,
    "generator": torch.manual_seed(0),
    "true_cfg_scale": 4.0,
    "negative_prompt": " ",
    "num_inference_steps": 50,
}

with torch.inference_mode():
    output = pipeline(**inputs)
    output_image = output.images[0]
    output_image.save("output_image_edit.png")
    print("image saved at", os.path.abspath("output_image_edit.png"))

Example: You load a photo of a rabbit and ask for the rabbit to turn purple against a flash-lit background. The script writes the edited picture to a new file, and everything else in the scene should stay as it was.

For newcomers and experts: Think of it as a photo editor you talk to. A first small test is the sample script with your own picture and a short instruction such as a color change.

The input picture goes to two places at once. Qwen2.5-VL reads it for meaning, so edits keep the sense of the scene, and the VAE encoder reads it for appearance, so untouched areas stay the same. That split is what lets one model do both a small local change and a full restyle.

Input picture and instruction
Qwen2.5-VL reads the meaning
VAE encoder keeps the appearance
Edited picture comes out

Good to know: The card says the model supports both low-level edits, where other regions stay unchanged, and high-level edits such as style transfer and object rotation. Other people have published 22 quantized and 29 fine-tuned versions.

Title: Estimated weight size
BF16: 40.9 GB
4-bit: 10.2 GB

• License: Apache 2.0.
• Cost to run: none named, apart from a GPU.
• Data it reaches: the files come from Hugging Face, and the pictures you edit stay on your own computer.
• Not for: making images from nothing, since it edits an existing picture.

On Hugging Face, 13 November 2026

• Task: image to image
• Library: diffusers
• Parameters: 20.4B
• Weights: about 40.9 GB in BF16; about 10.2 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 141,520
• Likes: 2,580
• License: apache-2.0
• Languages: en, zh
• Quantized versions: 22
• Fine-tunes: 29
• Tags: diffusers, safetensors, image-to-image

Links

• Model: https://huggingface.co/Qwen/Qwen-Image-Edit
• Files: https://huggingface.co/Qwen/Qwen-Image-Edit/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Qwen/Qwen-Image-Edit

Then follow the usage in its model card. Nothing in this kit runs by itself.