Model hub
How Qwen-Image-2.1 creates native transparent images
Qwen-Image-2.1 is an image generation and editing model designed for unified generation workflows, multi-reference edits, and native RGBA output. This page explains what that means in practice, embeds the Pellucen generator, and links to the model-specific ComfyUI and prompt guides.
Output
Transparent preview
Your image will appear here
The checkerboard reveals transparency. Download is always encoded as PNG.
What makes Qwen-Image-2.1 different?
Conventional background removal starts with an RGB image, estimates which pixels belong to the subject, and stores that decision as an alpha mask. Qwen-Image-2.1 can support transparency as an output property of generation or editing. That makes the transparent background part of the requested image, not an afterthought applied to a finished JPEG.
The distinction matters at difficult edges. Hair, fur, glass, foliage, and semi-transparent fabric are hard to recover after they have already been flattened onto a white or colored background. An RGBA workflow has a better opportunity to preserve soft coverage values while the subject is being generated or isolated.
Native support does not remove every practical limitation. The prompt still needs to describe the subject without requesting a scene, floor, wall, or cast shadow. The provider must enable transparent output, and the returned file must actually contain alpha data. Pellucen checks the PNG structure and rejects opaque files instead of displaying a misleading checkerboard preview.
For repeatable local results, follow the Qwen-Image-2.1 ComfyUI workflow. For better prompts, use the model-specific prompt guide.
Generate a transparent image with Qwen-Image-2.1
The generator above supports two paths. Text-to-image starts with a subject description and creates a new image. Image-to-image accepts a JPG, PNG, or WebP upload and asks the model to isolate the primary subject. Both paths send the editable prompt to the image provider from the server, so no image API key is exposed in the browser.
Pellucen begins the final model prompt with “This is an RGBA image with transparency.” It then describes the subject, framing, lighting, edge quality, and forbidden additions. DeepSeek is used only to rewrite text; it does not generate the image. The image request is handled by the configured Qwen-Image-2.1 provider.
If your final goal is a portable asset, continue to the transparent PNG maker or the remove background from image workflow.
Native RGBA versus traditional background removal
Traditional cutout software is often strongest when the subject is already in focus and cleanly separated from a plain background. Its workflow is inspect, refine mask, then export. Qwen-Image-2.1 can invert that order for generated assets: transparency can be requested from the beginning, which reduces the need to invent missing edge information later.
The practical result is not a guarantee of perfection. It is a faster first pass for products, character assets, game UI elements, stickers, and ecommerce drafts. Review the downloaded PNG at full resolution before publishing. If an edge needs more work, edit the alpha channel in a raster editor rather than placing the image over another background and flattening it again.
What does native RGBA mean for Qwen-Image-2.1?
RGBA adds an alpha channel to the red, green, and blue color channels. A native RGBA model can produce transparent pixels as part of the generation process instead of estimating a mask after an opaque image has already been created.
Is Qwen-Image-2.1 the same as Qwen-Image-3.0?
No. They are different model families. Pellucen uses provider identifiers that explicitly match Qwen-Image-2.1 so the model named on this page is the model used for the task.
Can Qwen-Image-2.1 remove the background from an uploaded photo?
Yes. The image-to-image route accepts a source image and a transparency instruction, then returns an isolated subject on a transparent background when the configured upstream model supports that task.