Qwen Image 3 AI Generator & Text Editor

[data-cookie-consent-banner], [data-testid="mobile-preview-shortcut"], [data-tool-deferred-pricing] { display: none !important; } .mobile-generator-footer { position: static !important; inset: auto !important; }

You can read this page and view examples without JavaScript. Enable JavaScript to upload files, change settings, and generate.

Qwen Image 3 from Alibaba's Qwen team renders English and Chinese text, infographics and realistic scenes at 1K or 2K, and edits text in up to three reference images.

[data-tool-example-compare-switch] { display: none !important; } [data-example-comparison] { grid-template-columns: repeat(2, minmax(0, 1fr)); } [data-example-comparison] > [data-active="false"] { display: block !important; }

Example Image

Alibaba Qwen's image model for English and Chinese text, infographics and label edits, at 1K or 2K.

Real Qwen Image 3 output · 2K · 3:4

Proofread a Qwen Image 3 menu in two languages

Qwen Image 3, from Alibaba's Qwen team, is built for images that carry text. We rendered six quoted Chinese and English lines twice at 2K, prompt expansion off and on, and checked every line.

Expansion off 5 of 6 exactExpansion on 4 of 6 exact

Qwen Image 3 chalkboard menu rendered with prompt expansion off: Chinese title 山间茶室, Mountain Tea Room, and four tea rows with prices — 1536 × 2048 · same prompt, same size, one switch changed

Qwen Image 3 chalkboard menu rendered with prompt expansion on, with an added chalk teapot and cups below the tea list — 1536 × 2048 · same prompt, same size, one switch changed

Line-by-line proofread of both Qwen Image 3 takes
Quoted line	Off	On
山间茶室	exact	exact
Mountain Tea Room	exact	exact
龙井 Longjing 28	exact	exact
白毫银针 Silver Needle 36	exact	error针 drawn as 井
正山小种 Lapsang Souchong 30	exact	exact
桂花乌龙 Osmanthus Oolong 32	error乌 dropped	error乌 dropped

English and prices were exact in both takes; errors were single Chinese characters. Expansion on also added a teapot and cups.

Read the menu prompt

A hand-painted chalkboard menu hanging on a wooden wall in a small tea house. At the top, the title "山间茶室" in large Chinese brush-style characters, with the English line "Mountain Tea Room" beneath it. Below, four rows, each with a Chinese name, an English name and a price: "龙井 Longjing 28", "白毫银针 Silver Needle 36", "正山小种 Lapsang Souchong 30", "桂花乌龙 Osmanthus Oolong 32". Chalk texture, small drawn tea leaves in the corners, warm lamp light, photographed straight on.

One switch, two infographics

What Qwen Image 3 prompt expansion writes for you

Expansion is on by default. It rewrites the prompt before rendering, and can write copy you never approved. Same prompt, same 1536 × 2048 size.

Qwen Image 3 infographic How Pour-Over Coffee Works with four numbered steps and the footer Ratio 1:16 · Water 94°C — All seven quoted strings exact. The drawings follow loosely: steps 2 and 3 show the kettle without the dripper.

Qwen Image 3 infographic with prompt expansion on, adding a Brewing Guide badge, bullet lists and side labels to the four steps — Richer layout, every quoted string kept, plus a page of copy nobody wrote.

Lines the model added

Brewing Guidea badge under the title
Releases trapped CO2 gasone of twelve bullet lines added under the steps
100°C Watercontradicts the 94°C footer it also kept
Keep the keffle spoot close to the groundskettle spout, misspelled

Rule of thumb: exact copy, expansion off. Exploring a layout, expansion on, then proofread every line it added.

Read the infographic prompt

A clean educational infographic poster titled "How Pour-Over Coffee Works". Four numbered panels arranged from top to bottom: "1 Rinse the filter", "2 Bloom for 30 seconds", "3 Pour in slow circles", "4 Finish by 3:00". Each panel has a simple flat illustration of a glass dripper and a gooseneck kettle. Footer line: "Ratio 1:16 · Water 94°C". Off-white background, terracotta and charcoal palette, consistent line weight, generous spacing.

Qwen Image 3 source: kraft coffee bag labelled MORNING FIELD, Medium Roast · 250 g, on an oak table with beans and a wheat stalk

Qwen Image 3 edit: the same bag in matte black labelled NIGHT HARBOR, Dark Roast · 250 g, with beans and wheat unchanged

Before · MORNING FIELDAfter · NIGHT HARBOR

Text edit · drag the divider

Rewrite a label with Qwen Image 3 Edit

The source bag was made with Qwen Image 3 at 1K; one edit changed both lines of copy and the paper color.

Changed: “NIGHT HARBOR”, “Dark Roast · 250 g”, matte black paper
Kept: Fold and clip, label position, beans, wheat stalk, table, light
Output: 2048 × 2048 from a 1024 × 1024 source. The edit chooses its own size, so Voor prices every edit at the 2K rate.

Read the edit instruction

Change the label text to "NIGHT HARBOR" and "Dark Roast · 250 g". Make the bag matte black paper and the label off-white with black letters. Keep the bag shape, the fold at the top, the oak table, the coffee beans, the lighting and the camera angle unchanged.

Scene edit · toggle

Keep the face, change the season

We asked Qwen Image 3 Edit for snow on a 1K harbour scene while keeping the man's face, pose, hands and net.

Face, cap shape and grey hair held
Hands still on the netting needle
Orange net, crates, flask and boots in place
Snow on boards, crates, boots and cap
Output grew to 2720 × 1536 from 1024 × 576

Read the scene instruction

Turn the scene into a snowy winter morning: fresh snow on the pier boards, the crates and the boots, light snow falling, and a dusting of snow on the man's cap and shoulders. Keep the man's face, his pose, his hands, the orange net and the camera angle the same.

After · snowBefore · source

Qwen Image 3 edit: the elderly fisherman mending an orange net on a snow-covered pier with snow falling

Qwen Image 3 source: an elderly fisherman mending an orange net on a wooden pier at dawn

Pick a ratio

Qwen Image 3 sizes and credits

Text to image offers five aspect ratios at two tiers. The outer frame is 2K, the filled frame inside it is 1K, drawn to scale.

1K text to image: 14 credits
2K text to image: 27 credits
Edit, any reference count: 27 credits

Our runs took 21 seconds to 2.5 minutes; edits took 3 to 4.5 minutes.

1:14:33:416:99:16

2K · 2048 × 20481K · 1024 × 1024

2K · 2048 × 15361K · 1024 × 768

2K · 1536 × 20481K · 768 × 1024

2K · 2048 × 11521K · 1024 × 576

2K · 1152 × 20481K · 576 × 1024

Brief for text

Write a Qwen Image 3 brief for text-heavy images

1
Quote every visible string
Exact words in quotation marks, in their language. Prompts can be Chinese or English.
2
Give each string a job and a place
Title at the top, rows below, footer at the bottom.
3
Fill the space you create
A timetable sign with two quoted lines came back with ten rows of unreadable filler. Say how many lines a surface holds.
4
Turn expansion off for exact copy
Keep it on only to explore a layout.
5
Use 2K for small type
Dense characters need pixels. Draft at 1K, finish at 2K.
6
Keep live facts outside the pixels
Prices, schedules and legal lines belong in HTML, not pixels.

For a single quick change with a strict preserve list, the earlier Qwen Image Edit is still a good fit. To swap the language of text that is already in a photo, start with the AI image translator. To test the same brief on models from other labs, open the text to image workspace.

Qwen Image 3 FAQ

What is Qwen Image 3?

Qwen Image 3 is an image generation and editing model from Alibaba's Qwen team. It renders complex text at up to 2048 × 2048 pixels and edits with one to three references. Prompts can be Chinese or English.

How many credits does Qwen Image 3 cost on Voor?

Qwen Image 3 text to image costs 14 credits at 1K and 27 credits at 2K. Edits cost 27 credits. New accounts get free welcome credits.

Can Qwen Image 3 write Chinese and English text accurately?

Often, but not perfectly. In our Qwen Image 3 chalkboard test every English word and price was exact, while Chinese errors were single characters: five of six lines were exact with prompt expansion off and four of six with it on.

What does prompt expansion do in Qwen Image 3?

Prompt expansion lets the model rewrite your prompt before it renders. It is on by default. In our test it added unrequested labels, one contradicting the footer. Turn it off when every visible word must stay exactly as written.

Which sizes can Qwen Image 3 generate?

Voor offers Qwen Image 3 in 1:1, 4:3, 3:4, 16:9 and 9:16 at two tiers. 1K is up to 1024 px on the long side; 2K doubles it.

How do I change the text in an image with Qwen Image 3?

Choose Edit image, upload the picture, quote the new words exactly and list what must stay unchanged, such as shape, layout, lighting and camera angle. References: 384 to 2048 px per side, up to 10 MB.

Can Qwen Image 3 combine several reference images?

Yes. Qwen Image 3 Edit accepts up to three references. Refer to them in the prompt as image 1, image 2 and image 3.

Does Qwen Image 3 keep a person's face when editing?

In our snow edit, Qwen Image 3 kept the fisherman's face, cap, pose and hands, plus the net and props, while adding snow across the scene. Name the details you need kept.

How long does a Qwen Image 3 generation take?

Our Qwen Image 3 text-to-image runs took between 21 seconds and about two and a half minutes, at both 1K and 2K. The two edits took between three and four and a half minutes.