Qwen-Image-3.0 от Alibaba: читаемый текст в 10 пикселей и инфографика за один проход

Qwen team (Alibaba) unveiled Qwen-Image-3.0 in July 2026 — an image generator that accepts prompts up to 4500 tokens long, renders readable text as small as 10 pixels, and natively supports 12 languages.

What Qwen-Image-3.0 can do

Qwen-Image-3.0 generates complex layouts in a single pass: infographics, pages styled like scientific LaTeX papers, and entire newspaper spreads. The main difference from previous generators is the readability of small type: the model outputs legible text as small as just 10 pixels tall, while most diffusion models turn such text into unreadable mush.

Key specs from the Qwen team:

• Prompt length — up to 4500 tokens
• Minimum readable text — from 10 pixels
• Native support for 12 languages
• In a single pass — infographics, LaTeX-style papers, newspaper spreads

The catch with complex layouts

The practical value of such layouts remains questionable. Qwen-Image-3.0 delivers a finished picture, not an editable document: a generated infographic or newspaper spread can't be opened and fixed — only redrawn with a new prompt. For real layout work, where text gets edited dozens of times, this limitation matters more than the mere fact that 10-pixel type is readable.

"The practical value of such layouts is questionable when the output is a pixel image rather than an editable format," notes

The Decoder.

That said, support for 4500-token prompts genuinely expands control: a single request can describe page structure, block text, and style all at once, without splitting the task into several steps.

How this looks against the market

Qwen-Image-3.0 continues Alibaba's line of open models, which compete with closed generators like Google Nano Banana and OpenAI's solutions. The bet on in-image text and multilingual support across 12 languages is a direct answer to the weak spot of most image models: until now, text on generated images had to be fixed by hand or pasted over it in an editor.

What this means

Image generators are edging into a niche once reserved for designers and layout artists: posters, infographics, covers dense with text. As long as the output stays a "flat" picture without layers, Qwen-Image-3.0 is more a tool for drafts and previews than a replacement for an editable layout — but it raises the bar for small-text readability noticeably.

Frequently asked questions

What can Qwen-Image-3.0 do out of the box?

In a single pass, the model generates infographics, pages styled like LaTeX papers, and newspaper spreads, accepts prompts up to 4500 tokens, and renders readable text from 10 pixels in 12 languages.

Why is complex layout generation criticized?

Because the result is a pixel image, not an editable file. A finished infographic or newspaper page can't be edited like a document — only regenerated entirely with a new prompt.