Qwen-Image-2.1 Brings Native 2K Generation and RGBA Transparency to ComfyUI

ComfyUI, the node-based open-source AI generation interface, added native support for Qwen-Image-2.1. The integration brings open-weight text-to-image synthesis, multi-image editing across up to 10 reference inputs, and direct RGBA alpha channel output to local ComfyUI workflows. This update gives creators a dedicated open-weights pipeline for producing production-ready transparent visual assets without secondary matting passes.

What's new

Qwen-Image-2.1 is a 7-billion parameter vision model developed by Alibaba Cloud, built specifically for high-resolution image generation and multi-turn instruction-based editing. Key technical capabilities integrated directly into ComfyUI include:

• Native RGBA transparency output: The model generates images directly with transparent alpha backgrounds on request, removing the need to run secondary rembg or segmentation nodes to isolate subjects.
• Native 2K resolution: Qwen-Image-2.1 synthesizes images up to 2048×2048 without intermediate upscalers or high-resolution fix passes.
• Multi-image reference editing: Users can feed up to 10 reference images into a single prompt node, enabling complex style blending, consistent subject modification, and multi-asset composition.
• Open-weight architecture: The weights are publicly accessible, allowing users to run the entire generation pipeline locally on consumer hardware with sufficient VRAM.

How it fits your workflow

For VFX artists, motion designers, and video editors working in DaVinci Resolve, Adobe After Effects, or Nuke, Qwen-Image-2.1 in ComfyUI streamlines asset generation. Traditional AI image generators like Midjourney v6 or FLUX.1 [dev] output RGB files with solid backgrounds, forcing artists to manually rotoscope or rely on background-removal models that often destroy fine hair, glass edges, or motion blur. Direct RGBA output bypasses this cleanup stage entirely, dropping transparent UI elements, prop cutouts, and character plates straight onto the editing timeline.

The 10-image context window also shifts how storyboard artists handle character consistency. Instead of juggling complex IP-Adapter setups or training dedicated LoRAs, creators can supply multiple angles, wardrobe variations, and background plates directly into the base prompt. Compared to single-reference systems in tools like Ideogram 2.0 or standard Stable Diffusion XL nodes, Qwen-Image-2.1 interprets spatial relationships across multiple visual references simultaneously.

What it costs / how to try it

ComfyUI and the Qwen-Image-2.1 model weights are free and open-source. Creators can run the model locally by updating ComfyUI to the latest release and loading the dedicated Qwen-Image-2.1 nodes through the standard node library.