What is Qwen Image Edit?
An instruction-based photo editor, not a blank-canvas generator. You start from a real file and ask for a specific change.
Searchers looking up this name usually want two things: a clear explanation of the model, and a place that actually runs it. This page is both. The form at the top is the product. The guide below covers what the editor is good at, how it differs from text-to-image, and how to write prompts that keep faces, labels, and product geometry stable.
Qwen Image Edit runs in the browser on Pixwit. There is no desktop install and no local GPU. You sign in, spend credits based on output size and how many variants you request, and export PNG or JPEG when a result is close enough to publish or to hand off to another tool.
The underlying job is image-to-image with language. The model reads your upload plus the prompt, then paints a new still that should follow the instruction without throwing away the original composition. That is why a sentence such as “change the sky to a golden sunset, keep the person and foreground unchanged” outperforms “make it cinematic.” Named regions and a keep-list give the editor something to hold.
Use this studio when the photograph already exists: a product packshot, a headshot, a listing photo, a fashion frame. If you do not have a source file and need a scene invented from words, switch to Qwen Image 2.0 text-to-image instead. Many teams generate a hero still there, then bring the file here for a last language-based tweak.
How the editor works on Pixwit
Upload one reference photo and describe the edit in your prompt. Choose the output aspect ratio, PNG or JPEG, and an optional seed.
How to edit images with Qwen Image Edit
Four steps from upload to download. Stay specific in the prompt and set the frame before you spend credits.
Upload reference images
Add one to three JPG, PNG, or WebP files (10MB each). The first file is the photo to change. Extra files are references — call them Image 1 / 2 / 3 in the prompt when you fuse style or objects.
Describe the change
Write what should move and what must stay. Background swaps, color and light, style passes, weather, and object recolors all work when you name the region. A keep-list (“keep the face, keep the label”) is as important as the change itself.
Choose output settings
Pick a ratio (1:1 through 21:9) and PNG or JPEG. Generate one to four variants per run. Use a seed when you already like a look and only want a small prompt edit. Acceleration trades a little quality for speed.
Compare, download, or continue
Use the slider to check identity and layout, download the still, edit again, or send the file to the AI video generator. If the pose is right but the sky is wrong, tighten the prompt rather than starting from a new photo.
What you can control before you run
Natural-language instructions
No masks or layer stacks. Describe the change in a sentence. Concrete nouns and a keep-list beat vague praise such as “make it better.”
Structure-preserving edits
The job is to revise a photo, not invent a new one. Composition, subject identity, and spatial layout should survive while the targeted region updates.
Flexible aspect ratios
Output in 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9, or 21:9 so social cuts, banners, and print frames are planned before you spend credits — not cropped later.
Reproducible seeds
Set an integer when you want the same prompt and model version to replay. Leave it empty for a fresh random sample. Useful when you A/B a caption around a locked visual.
Fast browser turnaround
Most jobs finish in under a minute. Upload, prompt, download — no desktop suite and no GPU setup on your machine.
PNG or JPEG export
PNG keeps edges and type cleaner for a further edit. JPEG is smaller for web and social. Choose per run; you do not have to reconvert unless a CMS requires it.
When this editor is the right tool
E-commerce product stills
Recolor a SKU, move a packshot onto white or a lifestyle counter, and refresh a catalog without a reshoot. Keep the label in the keep-list so type stays readable.
Social and channel cuts
Adapt one photo for Instagram, TikTok, or YouTube: seasonal skies, brand palettes, or a quieter background that leaves room for captions.
Real estate and interiors
Preview staging, wall color, or day-to-dusk light on listing photos. Describe the room region you want changed and name the architecture that must stay.
Portraits and profile frames
Adjust light, wardrobe, hair color, or backdrop for headshots without a second studio session. Ask the model to keep facial features when you apply a style pass.
Creative style experiments
Try anime, watercolor, vintage film, or a fantasy atmosphere on a photo you already like. Iterate with small prompt edits instead of restacking adjectives.
Campaign variants
Turn one hero still into several backgrounds, seasons, or colorways for A/B tests. Lock a seed when the composition is right and only the setting should move.
Prompt patterns that keep identity
Lead with the change, then add a keep-list. “Replace the office background with a sunlit kitchen, keep the product label sharp and the bottle outline unchanged” tells the editor both jobs. “Make it nicer” does not. Put the subject you care about in the first clause; early words carry more weight.
When you use extra references, name them. “Keep the person from Image 1; apply the jacket material from Image 2; use the pose from Image 3” is clearer than attaching three files and hoping the model guesses. If a still comes back with a warped face or a melted logo, shorten the style request and strengthen the keep-list rather than adding more adjectives.
Negative prompts are a veto list, not a second creative brief. “blurry, extra fingers, watermark, text artifacts” is useful. Repeating the whole positive prompt in the negative box is not. Generate two or three variants, pick the strongest identity, then run a tighter follow-up instead of stuffing the 2,000-character cap.
Qwen Image Edit vs Qwen Image 2.0
Qwen Image Edit needs an upload and applies your instructions to that file. Qwen Image 2.0 creates a new still from text with no source photo. Use the editor when the photograph exists and you only want a directed change. Use the generator when the photo does not exist yet.
A practical loop: generate two or three candidates in the text-to-image studio, download the strongest composition, then open this editor to fix a sky, outfit, or label. That is usually faster than hoping a brand-new text-to-image roll will nail every detail at once.
Related Pixwit tools
Stay on Pixwit if you need a broader image stack after you edit:
- Qwen Image 2.0
Text-to-image when you do not have a source photo yet.
- AI Image Generator
Multi-model studio for new stills from a written scene.
- AI Image Editor
Retouch, expand, and transform existing photos with a wider toolset.
- Pricing
See credit packs and what each megapixel of output costs.
Qwen is the model family from Alibaba Cloud’s Qwen team. Pixwit hosts this editor in the browser; model research and naming live on the official Qwen site. Official Qwen site
Common questions
What is Qwen Image Edit?
- Qwen Image Edit is Pixwit’s instruction-based photo editor powered by the Qwen2 image-edit model. You upload an existing still and describe the change in natural language. Unlike a text-to-image generator, it does not invent a scene from scratch — it revises the file you already have while trying to keep structure and identity.
What kinds of edits work well?
- Background replacement, color and lighting, style passes (anime, watercolor, vintage), weather, object recolor, and atmosphere. Be specific: “change the blue car to metallic red, keep the street and reflections” beats “make it look better.” If a region must not move, say so in the prompt.
How many reference images can I upload?
- Up to three per job. The first is the main photo. Extra files can supply style, an object, or a pose. Refer to them as Image 1, Image 2, and Image 3 in upload order. With several files, output ratio typically follows the last upload.
What formats and limits apply?
- Input: JPG, PNG, or WebP up to 10MB. Prompt up to 2,000 characters; negative prompt up to 500. Output: PNG or JPEG. Ratios from 1:1 through 21:9. One to four images per run.
How much does each edit cost?
- Pricing is $0.03 per output megapixel (12 credits per MP on Pixwit). Cost scales with ratio and how many variants you request — for example 16:9 (~1 MP) is about 12 credits per image; four images are roughly 48 credits. Minimum 5 credits per output image. The generate button shows the cost before you start.
What does the seed parameter do?
- An optional integer that controls randomness. The same seed, prompt, and model version should return the same still, which helps when you iterate copy around a locked visual. Leave it empty for a new random sample each click.
Can I use edited images commercially?
- Stills you edit on Pixwit are yours for personal or commercial projects. Review the current Pixwit terms for your account type, and do not generate content that violates those terms or applicable law.
Edit a photo with Qwen Image Edit
Upload a still, write the change, and download a structure-preserving result — the editor is at the top of this page.