Alibaba Qwen-Image-3.0: 4500-Token Prompts, 10px Readable Text, 12 Languages

Qwen-Image-3.0, the third-generation image generation model from Alibaba's Qwen team, accepts up to 4500-token prompts and can generate complex content like multi-panel infographics or nested interfaces (e.g., VSCode with Qwen Chat open, embedded WeChat conversations, and a coffee poster). It renders readable text as small as 10 pixels, supports 12 languages including Japanese, Korean, and Spanish, and handles multi-line LaTeX formulas. Additionally, it can generate portraits with skin pore and hair details, and restore missing parts of traditional Chinese ink paintings. Currently available only via invite-only API, with plans to integrate into first-party apps like Qwen Chat. Unlike the original Qwen-Image, this model likely won't open-source weights. For design and content teams needing frequent infographics, paper covers, or UI mockups, this model reduces multi-image assembly and post-processing, though the limitation of pixel-based output (vs. editable formats like SVG) should be considered.