# Qwen-Image 2.1 Inpainting with LanPaint — alpha channel included — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/6ce7611eaeef9c9c

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-09-28 12:18 ingested 2026-09-29 02:40

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1wsd1xa/qwenimage_21_inpainting_with_lanpaint_alpha/)

## retained text



```
Hi everyone,

I wish to announce that LanPaint now supports Qwen-Image 2.1 — both the text-to-image model and the image edit model.

Qwen-Image 2.1 can generate transparent PNGs. With LanPaint you can inpaint the alpha channel too, not just the colours. So the transparent part can change shape, not only the object.

The image edit model works under a mask as well: describe the change, paint over the part it should touch, and everything else comes back exactly as it was.

LanPaint is a universal inpainting/outpainting tool that works with almost every diffusion model out there (SD, Flux, Wan, Qwen, and now Qwen 2.1)—it is especially useful for models without dedicated inpainting checkpoints.

Check it out on GitHub: [https://github.com/scraed/LanPaint](https://github.com/scraed/LanPaint) — feel free to drop a star if you find it useful! ⭐


```



document fetched 2026-09-29 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1wsd1xa/qwenimage_21_inpainting_with_lanpaint_alpha
