Qwen-Image-2.1: 7B Model Combines Generation, Editing, and Transparent Channel

Qwen-Image-2.1, a 7B parameter open-source model, unifies image generation, editing, and transparent channel support into a single Transformer architecture. It features block-causal attention for efficient caching, a 64-channel RGBA autoencoder for native transparency, and supports up to 10 reference images for high-fidelity editing. The model prioritizes engineering efficiency with support for FP8 quantization and multiple chip platforms, aiming to lower inference costs.