Download App

[Qwen-Image-2.1 Officially Open-Sourced: Lightweight Design, Outstanding Performance, and Integrated Creation and Editing Functions] According to official news from the Qwen large-scale model team, Qwen-Image-2.1 has recently been officially open-sourced. This new image large-scale model achieves an excellent balance between generation quality, inference efficiency, and usage cost. Qwen-Image-2.1 integrates text-to-image generation and image editing capabilities into the same model architecture. The visual generation module has only 7 billion parameters and natively supports the generation and editing of transparent images. The core highlights of this updated model are mainly reflected in four aspects: 1. Powerful small-parameter model with outstanding cost-effectiveness: Relying on a lightweight model architecture design and deep optimization of the inference process, the optimal balance between image generation quality and computational cost has been found, significantly lowering the barrier to entry. 2. Native support for transparency channels, integrated creation and modification: It can generate regular images or images with transparent backgrounds based on user-input prompts, while also supporting the editing of transparent layers and intelligent image cutout functions to meet diverse design needs. 3. Enhanced full-scene editing capabilities with multi-area detail preservation: Supports importing up to 10 reference images simultaneously, significantly improving the fidelity of local editing effects and the reproduction of portrait and product images, covering most everyday image editing tasks. 4. More realistic generated textures, beautiful text, and natural-looking portraits: Specific optimizations have been made to text layout, lighting and shadow rendering of portraits, and detail texture representation, resulting in a significantly improved visual appeal of the final generated content, better meeting practical usage needs.