Qwen Image 2.1: Image Edit
Edit images with Qwen-Image-2.1 by feeding two reference images through a single checkpoint that handles both generation and editing, then comparing the result against the original. Ideal for character-consistent edits, product cutout compositing with real transparency, and style-transfer passes at native 2K.
例を試す
このワークフローについて
This workflow demonstrates image-to-image editing with Qwen-Image-2.1 by feeding two reference images into a single checkpoint that handles both generation and editing. The first image (image_1) is your base subject, and the second image (image_2) is the visual guide for what you want to change—such as clothing, style, or material. A ResolutionSelector node controls canvas size: by default it preserves image_1’s aspect ratio at approximately resolution×resolution (default 1024, rounded to a multiple of 32).
Qwen-Image-2.1 supports native 2K (2048×2048) output, or you can set resolution to 0 to keep the original pixel dimensions (again rounded to a multiple of 32).
Technically, the LoadImage nodes feed both references into the Qwen-Image-2.1 checkpoint node (bc1c967a-7f6a-4be9-a372-dad16e4f28e3), which fuses them to produce an edited image anchored to the content of image_1 and guided by the appearance, style, or cutout mask in image_2. This makes it well-suited for character-consistent edits, product cutout compositing with real transparency (when your guide image has an alpha channel), and style-transfer passes at high resolution. The ImageCompare node shows before/after frames side-by-side for quick review, and SaveImageAdvanced lets you export clean PNGs to preserve transparency.
MarkdownNote and Note nodes in the graph provide inline tips, and the Qwen edit block is organized as a Subgraph for easy navigation.
よくある質問
ResolutionSelector preserves the aspect ratio of image_1 and scales to about resolution×resolution (default 1024), rounded to a multiple of 32. Set resolution to 2048 for native 2K, or 0 to keep the original pixel dimensions (also rounded to a multiple of 32).
No. This workflow uses two reference images. The base image (image_1) anchors content and composition, while the guide image (image_2) drives the visual change, such as clothing or style.
Yes. If your guide image (image_2) includes an alpha channel, the workflow can maintain real transparency in the result. Export with SaveImageAdvanced set to PNG to preserve the alpha channel.
Reduce the ResolutionSelector value (e.g., to 1024) or keep original size with resolution=0 if it’s smaller. Native 2K (2048×2048) is supported by Qwen-Image-2.1 but requires more GPU memory.