Qwen Image Edit lets you change an existing image with written instructions. Upload a photo, describe the change, and use an image-editing workflow to generate a revised image. Documented editing tasks include replacing a background, removing an object, changing colors, adjusting lighting, and editing text inside a picture.
The important first step is choosing a tool that returns an image. Uploading a photo to a conversation that only analyzes or describes it does not, by itself, produce an edited file. This guide walks through a practical editing workflow, with prompts you can adapt and checks to make before keeping the result. For creating a new picture from a text description, see our Qwen image generation guide.
Last reviewed: October 12, 2026. This guide is based on official Qwen documentation. The prompts are illustrative examples; available tools, model labels, and account allowances can change.
How to edit an image with Qwen
Start with the official Qwen assistant. The consumer service may be labeled Qwen Studio. Qwen’s official image-editing model documentation directs users to the Image Editing feature; follow the image tools available in your current interface. The consumer steps below are for personal, non-commercial use under the current service terms.
- Open an image-editing workflow. Sign in if requested, then look for Image Editing or an image tool that accepts a reference picture. Choose this before judging whether Qwen can carry out the edit.
- Upload the original image. Use a clear copy in a format accepted by the upload control. Wait for the upload to finish and confirm that the intended picture is attached. For a first attempt, work with one image.
- Describe one specific change. Name the object or area, explain the desired result, and identify the details that should stay the same. “Make this better” leaves much more room for interpretation than a defined background or color change.
- Request the edited image. Submit the instruction through the selected image tool. If the response only describes the photo or offers instructions for another editor, check the workflow before repeatedly changing your prompt.
- Compare the result with the original. Inspect both the requested change and the parts you wanted preserved. Check small text, faces, edges, reflections, and object proportions at a useful zoom level.
- Refine and save a separate copy. Make the next request about the most important remaining issue. Keep the original and each approved version so you can return to a good starting point if a later edit changes too much.
There is no need to choose a model merely because it can understand images. For example, the Qwen3.8-Max model page lists image input and text output. Image understanding can help explain a picture, while an image-output model or tool is needed to create a revised picture.
Write a prompt that protects the details you care about
A useful editing prompt has four parts: the source image, the change, the details to preserve, and the expected output. Put the main change first. Then name the two or three details that would make the result unusable if they changed.
Edit the attached image. Change: [describe the object or area and the exact change]. Preserve: [name the subject, text, colors, framing, or other details that must remain consistent with the original]. Avoid: [name the most important unwanted changes]. Return the edited image.
Treat preservation instructions as acceptance criteria. They tell you what to inspect afterward; they do not guarantee an exact match. If a logo, label, or face must remain unchanged, compare it directly with the source before accepting the edit.
Five practical Qwen image-editing prompts
Qwen’s official Image Edit introduction documents changes to image elements, backgrounds, clothing, and text. The examples below turn those capabilities into focused requests. Adapt the object names and instructions to your own source image.
1. Replace a background while keeping the subject
Imagine a photo of a blue ceramic mug on a cluttered desk. Your goal is to change the setting while keeping the mug recognizable. Start with a simple replacement:
Replace the cluttered desk background with a plain warm-gray studio background. Keep the blue mug's shape, handle, glaze, printed design, and position consistent with the original. Add a soft contact shadow beneath the mug. Keep the same framing. Return the edited image.
Check the handle opening, rim, printed design, and boundary between the mug and its new background. If the background works but the mug has changed, return to the original and make the preservation instruction more specific:
Change only the background in the original attachment. Preserve the mug's original handle shape and printed design. Do not restyle, reshape, or add decoration to the mug.
2. Remove an unwanted object
Remove the empty chair on the far left of the attached photo. Reconstruct the wall and floor behind it so their lines and textures remain consistent. Preserve the people, table, lighting, and camera framing. Return the edited image.
Inspect the area where the object stood, including its shadow or reflection. For floor tiles, furniture, fences, and walls, check that straight lines continue naturally through the edited area.
3. Change text inside an image
Put the existing and replacement wording in quotation marks, and identify where it appears. Qwen’s original Image Edit documentation specifically demonstrates English and Chinese text editing.
In the attached poster, replace the top headline "WINTER WALK" with "SPRING WALK". Match the existing letter style, color, alignment, and approximate size. Preserve the illustration, date, venue, and all other wording. Return the edited image.
Read every word afterward, including text outside the requested change. Compare spelling, punctuation, spacing, and numbers with the source. When exact typography is essential, adding the final wording with a conventional text layer gives you direct control over the characters and layout.
4. Change one object’s color
Change only the red backpack to dark forest green. Preserve its fabric texture, seams, zippers, shape, and shadows. Keep the person's clothing, skin, background, and composition consistent with the original. Return the edited image.
Check that the change stayed within the intended object. Inspect straps, small fasteners, overlapping edges, and nearby objects that share the original color.
5. Adjust portrait lighting
The Qwen-Image-Edit-2511 model card describes lighting control and improved character consistency. Keep your request focused on illumination when that is the change you want.
Adjust this portrait to soft, natural window lighting. Reduce the harsh shadow on the face while preserving the person's facial features, expression, skin texture, hairstyle, clothing, and pose. Keep the original background and framing. Return the edited image.
Compare the eyes, mouth, hairline, and skin details with the original. A more flattering picture is not automatically a faithful edit of the same person.
Use multiple reference images with clear roles
If the selected workflow accepts multiple images, say which one is the main image and what each additional reference contributes. Confirm their order before submitting the request. Avoid asking Qwen to combine several pictures without explaining which details belong in the result.
Image 1 is the photo to edit. Image 2 is a background reference. Place the subject from Image 1 in a setting inspired by Image 2. Preserve the subject's appearance, clothing, and pose from Image 1. Use only the environment and lighting direction from Image 2. Do not copy people, signs, or text from Image 2. Return one edited image.
Reference-image allowances depend on the model and service. Follow the limit displayed by your current consumer interface. A reference limit documented for an API model does not establish the upload allowance in Qwen Studio.
For a transparent background, check the actual file
The official Qwen-Image-2.1 model card documents transparent RGBA images and subject extraction. This is a capability of that model; your chosen service must also expose an appropriate workflow and preserve transparency in the download.
Extract the main subject from this image and remove the background. Preserve the subject's outline, fine edges, and original colors. Return a PNG with a transparent alpha channel. Do not draw a white background or a checkerboard pattern.
Open the downloaded file over a colored background to inspect the cutout. A white area or a checkerboard drawn into the picture is not transparency. Check for edge halos and missing fine details before using the image in another composition.
Check the whole image before accepting an edit
Use the original as your reference, and review the result at both normal viewing size and a closer zoom. Give particular attention to:
- The requested change: Did the correct object or area change, and was the instruction completed?
- The subject: Are facial features, proportions, clothing, and distinctive details still suitable?
- Text and markings: Are names, numbers, logos, labels, and small lettering still accurate?
- The surrounding image: Do shadows, reflections, perspective, and unedited objects remain coherent?
- The downloaded file: Are its dimensions, aspect ratio, format, and any required transparency appropriate for your use?
If an edit damages an important detail, work again from the original or the last approved version. Keep a short record of the source image and prompt for each version so you can identify which change produced the problem.
Troubleshooting Qwen image editing
| Problem | What to check or try |
|---|---|
| Qwen describes the picture but returns no edited image. | Confirm that you selected an image-output workflow and attached the source image. A text answer alone does not show that an image-editing task ran. |
| The main subject changes too much. | Restart from the original, request one smaller change, and name the subject details to preserve. Compare the next result with the original. |
| The edit affects the wrong object or area. | Identify it by position, color, and surrounding objects. If your specific tool supports visual annotations, use a clearly marked working copy and explain the mark. |
| Text is misspelled or other wording changes. | Specify the exact old and new wording, then inspect all visible text. Use a conventional text layer for wording that needs precise final control. |
| Reference images are mixed together. | Check the attachment order. Assign one role to each image and say which details must come from the main source. |
| The upload or edit is rejected. | Read the displayed message, check the accepted file format and account limit, and confirm the file opens normally. Follow any service restriction shown for that request. |
| The downloaded background is not transparent. | Check whether the selected model and export route support RGBA output. Inspect the downloaded file rather than relying on the preview background. |
Is Qwen image editing free?
Qwen’s public consumer overview describes the assistant as free to use. That statement does not establish unlimited image edits, a fixed daily allowance, or identical tools for every account. Check your current account for image-editing availability, usage limits, and any charges.
Developer API usage has separate billing. For example, the Model Studio Qwen-Image-3.0-Pro page lists charges for reference-image input and generated-image output. Repeated attempts can therefore add to the cost of an API editing workflow. See our Qwen pricing guide for the broader billing options, and confirm the current rate for your selected model and region.
Image-editing options for developers and local users
Alibaba Cloud Model Studio API
The current Qwen Image Generation and Editing API reference lists qwen-image-3.0-pro, qwen-image-3.0, qwen-image-2.1-pro, and qwen-image-2.1-turbo. In that API, the 3.0 series accepts 1–3 reference images and the 2.1 series accepts 1–10.
For its OpenAI-compatible route, editing uses an image field on /images/generations. The documentation does not support /images/edits, multipart uploads, or a mask parameter on that route. Use the documented protocol and a model, endpoint, and key from the same region.
Our Qwen API setup guide explains provider and credential choices. Use the dedicated image-editing reference above for the editing request itself; a text-generation quickstart is not an image-editing example.
Downloadable image-editing models
Qwen publishes image-model instructions for frameworks such as Diffusers. Follow the exact model card for the image pipeline and its requirements. Installing a general Qwen chat or vision model does not automatically install an image editor.
Licenses differ between releases: the Qwen-Image-Edit-2511 card lists Apache 2.0, while the downloadable Qwen-Image-2.1 weights use the Qwen Research License, which limits use to non-commercial research or evaluation unless a separate commercial license is obtained. Check the license of the exact files you plan to run.
Usage rights and privacy
The Qwen consumer Service Agreement, effective September 10, 2026, restricts the service to private, non-commercial use. For commercial editing, choose a service or model license that expressly permits your intended use. Hosted API terms and downloadable model licenses must be checked separately.
The Privacy Policy, effective September 7, 2026, says de-identified user content and feedback may be used for model training. The service agreement describes a training opt-out, with an exception for feedback. Review those provisions and confirm that you have permission to upload the images, especially when they contain other people’s personal information.
Frequently asked questions
Do I need an API key to edit a photo in Qwen?
You do not need to create a developer API key for the consumer image-editing workflow. Open the official assistant and follow its sign-in and access requirements. API keys apply when you integrate a developer service into your own software.
Can Qwen keep a face or logo exactly unchanged?
You can instruct Qwen to preserve it, but check the result against the original. A model’s documented consistency improvements do not guarantee an exact match for your image. If exact preservation is required, retain the original subject or logo in a conventional editing workflow and review the final composite.
Is Qwen Image Edit one specific model?
The phrase can refer to Qwen’s image-editing feature or to a named model release, such as Qwen-Image-Edit-2511. Hosted services also offer newer unified generation-and-editing models. Check the exact model and access route when comparing capabilities. Our Qwen AI overview explains the wider product family.
Official sources
- Qwen-Image-Edit introduction — documented editing tasks and text-editing examples.
- Qwen-Image-Edit-2511 model card — image-editing workflow, consistency improvements, and model license.
- Qwen-Image-2.1 model card — unified generation and editing, reference images, and transparency.
- Alibaba Cloud image-generation and editing API reference — hosted models, request formats, and route-specific limits.
ChinaAI.chat is an independent guide. See our review methodology for how we distinguish official documentation from hands-on testing.