ModelsMedia Generation 🇨🇳 27.07.2026 17:05

Qwen-Image-Edit: High-Quality and Efficient Image Editing

Alibaba/QwenAlibaba/Qwen
Alibaba's Qwen team releases Qwen-Image-Edit, an image editing model built on the 20B Qwen-Image model. It combines Qwen2.5-VL for visual semantic control and VAE encoder for appearance control, supporting precise text editing in Chinese and English, as well as semantic and appearance editing tasks.
Alibaba's Qwen team introduces Qwen-Image-Edit, an image editing model derived from the 20B Qwen-Image model. It integrates Qwen2.5-VL for visual semantic control and VAE encoder for visual appearance control, enabling both semantic editing (e.g., style transfer, object rotation) and appearance editing (e.g., adding/removing elements) with high precision. The model supports bilingual text editing (Chinese and English) while preserving original font, size, and style. Tests on multiple public benchmarks show state-of-the-art performance. Users can access it via Qwen Chat's 'Image Editing' feature. Examples include editing a Capybara mascot, rotating objects 180 degrees, transferring styles like Studio Ghibli, and correcting calligraphy errors step by step.
Сокращения
VAE = Variational Autoencoder — вариационный автоэнкодер
SOTA = state-of-the-art — передовой уровень
MBTI = Myers-Briggs Type Indicator — индикатор типов Майерс-Бриггс
Source: Alibaba Qwen — original
Our earlier posts on this topic ↓
Fresh news