Qwen’s powerful, new image editing model

August 27, 2025

Alibaba’s Qwen team has launched Qwen-Image-Edit, an open-source model with 20 billion parameters, designed to combine pixel-perfect editing with stylistic transformation while keeping original characters and objects untouched.

The model splits editing into two categories: transformations like object rotation or style transfer, and targeted modifications on specific parts of an image while leaving the rest intact. Thanks to built-in bilingual capabilities, users can directly edit Chinese and English text within images without breaking existing fonts, sizes, or formatting.

Qwen-Image-Edit also allows layered editing, enabling users to refine visuals step by step instead of starting from scratch each time. In benchmark tests, the model delivered state-of-the-art (SOTA) results, outperforming rivals like Seedream, GPT Image, and FLUX.

While image generation models have surged ahead in the past two years, this marks a clear shift: AI image editing is entering a new era. By open-sourcing the model, Qwen puts natural-language-driven precision editing in the hands of the broader community—setting a new standard for how we’ll work with visual content going forward.

In Brief: Tech World Highlights

Anthropic expanded availability of its Claude Code programming agent to Enterprise and Team plans, adding new admin controls for cost and policy management.
MIT’s NANDA initiative revealed that only 5% of enterprise AI deployments generate revenue, with knowledge gaps and poor integrations slowing broader adoption.
Sebastien Bubeck (OpenAI) stated that GPT-5-pro can “prove new interesting mathematical theorems,” applying the model to tackle open complex problems.
Google’s Logan Kilpatrick teased a “nano-banana” model on LM Arena—likely a Google photo-editing model currently in testing.
OpenAI announced the launch of ChatGPT Go, a low-cost subscription for India priced under $5/month with local currency support.