Alibaba releases Qwen-Image-2.1 for generation and image editing
Alibaba’s Qwen team has released Qwen-Image-2.1, a unified model for text-to-image generation and image editing. Its visual generation component has 7B parameters and uses 32 single-stream diffusion-transformer layers. The same checkpoint supports local edits, multiple reference images and native RGBA transparency. The pipeline also loads an 8B Qwen3-VL encoder, so the 7B figure does not represent the full system.
The model can accept up to 10 reference images and supports region edits using painted annotations, circles or masks. It generates 2048 × 2048 images by default and supports seven aspect ratios, reaching 2752 × 1536. Qwen highlights applications including panoramas, infographics, storyboards and virtual try-ons. A prefix KV cache calculates text and reference-image information once, then reuses it during denoising.
Qwen-Image-2.1 has Day 0 support in Diffusers, ComfyUI, vLLM-Omni, SGLang and LightX2V, with additional support for AMD GPUs through ROCm and eight chip platforms via FlagOS. On Qwen’s Qwen-Image-Bench, it scores 60.28, above the listed open-weight models. The weights are available for research and evaluation; commercial use requires a separate license from Qwen.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.