Print-grade layouts
Newspapers, magazines and posters with real column structure and hierarchy.
Alibaba's newest image model · announced July 21, 2026
Qwen Image 3.0
The “solid” generation: dense layouts, real detail, world knowledge.
Qwen-Image-3.0 pushes text-to-image from “looks good” to “actually usable” — rendering newspapers, storyboards, exam papers, and real UI at native 2K. Explore the official 3.0 examples below, or generate live with the Qwen Image tool right here.
The live tool runs Qwen Image (native 2K) through the KIE backend — the same Qwen image family. Qwen-Image-3.0 is a hosted preview with no public API yet, so the labeled examples below are official Qwen team outputs, not tool renders.
4.5K tokens
Prompt length it can render
10px text
Still legible on the page
12 languages
Rendered natively
100+ styles
Art directions covered
What makes Qwen Image 3.0 different
The Qwen team frames 3.0 around three kinds of “solid”: content that scales, detail that holds up close, and knowledge that spans languages and interfaces. Here is what each one looks like in the official launch images.
01 · Content that scales
One prompt, a full information-dense layout
Qwen-Image-3.0 can place many complex panels on one canvas without them bleeding into each other. The image below is a single generation — a nine-panel infographic covering tunnel safety, geometry, physics, biology, group theory and more, described in a 3.7K-token prompt.
Nested, picture-in-picture depth
A single instruction renders a VS Code window containing a chat app containing another chat containing a poster — each layer keeping its own real UI style.
A believable newspaper front page
Column grids, a masthead, headlines, body copy and captions — laid out with the texture and hierarchy of a printed paper.
02 · Detail that holds up close
Ten-pixel text, real skin, real paper
Where content richness answers ‘how much,’ detail realism answers ‘how fine.’ Qwen-Image-3.0 renders 10px type that stays sharp, LaTeX math with correct sub- and superscripts, and photographic textures down to pores and hair.
A full LaTeX paper page
Multi-line derivations, fractions, braces and theorem numbers, readable at small size.
A packed knowledge poster
Dozens of small captions and diagram labels, each rendered cleanly across the illustration.
Photographic micro-detail
Freckles, flyaway hair and soft skin texture that reads as a real camera capture.
Edit · Single-reference image-to-image
Editing that respects the original
Qwen image-to-image takes one reference and changes exactly what you ask. These official 3.0 edits restore a damaged ink painting and turn a plain photo into a publication-ready figure — the live tool above can do the same from a single upload.
Restore a damaged painting
Given a worn, spotted scroll, the model rebuilds the missing areas while keeping the ink gradients, feather texture and composition faithful to the original brushwork.
Instruction
Repair the damage and mold spots while preserving the original ink-wash style and brush strokes.
Before
After
Turn a photo into a labeled figure
Starting from a single insect photo, the model keeps the subject intact and adds taxonomy, morphology labels, a magnified inset and a scale bar — a figure ready for a paper.
Instruction
Keep the insect, add taxonomic classification, morphology labels, a zoomed detail view and a scale bar.
Before
After
03 · Knowledge across languages
Twelve languages, real interfaces, real references
The third strength is breadth. Qwen-Image-3.0 renders native type across many languages, reproduces convincing app and web layouts, and pulls in world knowledge — from a city weather board to specific artists — to ground an image.
Japanese
Manga-style poster with natural Japanese typography and layout.
Korean
A dress-collection poster set entirely in clean, correct Hangul.
Spanish
Handwritten-style whiteboard notes on a neural network, in fluent Spanish.
World knowledge
A weather-and-travel board that composes real places into one tidy infographic.
Named references
Two recognizable artists staged in a livestream studio, on brief.
Simulated, not screenshotted
A full mobile app feed rendered from one prompt — no templates, no stock assets, every pixel generated.
Legible small text
Type down to ~10px, LaTeX math and handwriting that stay readable.
Deep composition
Many parallel panels or nested interfaces held together in one image.
Reference editing
Restore, annotate or re-style a single source without losing the subject.
How to use the Qwen Image tool
1
Write a detailed prompt
Name the subject, spell out any on-image text and the language, and describe the layout. Long, structured prompts play to Qwen’s strengths.
2
Set ratio and generate
Pick an aspect ratio and generate at native 2K. Review the result in the history panel on the right.
3
Refine or edit
Change one detail at a time, or switch to Edit Image and upload a single reference to restyle or annotate it.
Qwen Image next to the other models here
Qwen Image is the value pick for text-heavy, multilingual and layout-driven work. Other models still make sense for polish or multi-reference editing.
| Capability | Qwen Image | Nano Banana Pro | GPT Image 2 | Z-Image Turbo |
|---|---|---|---|---|
| Dense text & layouts | Excellent | Strong | Excellent | Basic |
| Text-to-image | Yes | Yes | Yes | Yes |
| Image editing | Single reference | Yes | Up to 5 references | No |
| Native output | 2K | 2K | 1K–4K | 1K |
| Credits / image | 4 | 10 | 10 | 2 |
| Best fit | Newspapers, exams, multilingual UI | Premium image polish | Posters & text graphics | Fast idea tests |
Explore more image models
GPT Image 2
10 credits / image
Reach for GPT Image 2 when a poster or UI mock needs tight typography and multi-reference editing.
OpenNano Banana Pro
10 credits / image
The premium option when you want maximum polish and prompt adherence on a single hero image.
Try nowWan 2.7 Image
6 credits / image
Another native-2K model on the same KIE backend, tuned for clean photographic generation.
Try nowQwen Image 3.0 FAQ
What is Qwen Image 3.0?
Qwen-Image-3.0 is the third generation of Alibaba’s Qwen-Image foundation model, announced on July 21, 2026. Where 1.0 was about accuracy and 2.0 added variety and realism, 3.0 is built around one idea the team calls “solid”: dense, information-rich layouts, photoreal micro-detail, and broad world knowledge.
Can I use Qwen Image 3.0 online on this page?
You can generate live with the Qwen Image tool above, which runs the Qwen image family at native 2K. Qwen-Image-3.0 itself launched as a preview inside Qwen Chat without public weights or an API, so no site can serve 3.0 directly yet. Every example further down this page is an official Qwen-Image-3.0 result from the launch announcement.
What is new in Qwen Image 3.0 compared with 2.0?
Qwen-Image-3.0 accepts prompts up to about 4.5K tokens, so it can lay out a full newspaper, a nine-panel infographic, or an exam paper in a single pass. It renders text down to roughly 10px while staying legible, natively renders 12 languages, and can simulate real web, game, and app interfaces.
Is Qwen Image 3.0 good at rendering text?
Text is its headline strength. Qwen-Image-3.0 handles dense paragraphs, LaTeX formulas with superscripts and fractions, newspaper columns, and handwritten-style annotations, keeping small type readable instead of turning it into decorative squiggles.
How many languages does Qwen Image 3.0 support?
The launch describes native rendering across 12 languages plus more than 100 art styles, backed by enough world knowledge to reproduce believable posters, UI layouts, and infographics in each language.
Can the Qwen Image tool edit an existing image?
Yes. Switch the tool to Edit Image and upload a single reference. Qwen image-to-image is strong at restoration, adding realistic annotations, and turning a plain photo into a labeled figure while preserving the original subject.
How many credits does the Qwen Image tool use here?
Qwen Image costs 4 credits per image in the current Wan27Image pricing table, which makes it one of the cheaper high-detail options on the site.
Is Qwen Image open source?
Earlier Qwen-Image releases shipped open weights on GitHub and Hugging Face. Qwen-Image-3.0 was introduced as a hosted preview first, so the live tool here uses the available Qwen image model through the KIE backend rather than local weights.
Generate with Qwen Image
Test a prompt in the embedded tool above, or open the full workspace when you want history, downloads, and a larger editing surface.
Need tight typography and multi-reference edits? Try the GPT Image 2 generator