# I tested Qwen-Image-2.1: it's amazing! — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/240edfe35289b071

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-09-20 13:17 ingested 2026-09-21 02:37

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1wlgv0l/i_tested_qwenimage21_its_amazing/)

## retained text



```
I got early access to Qwen-Image-2.1

My first impression: Fantastic editing, new standard for open models, support for a lot of references (I tested with 10), consistent, pixel precise at 2K, text rendering at closed models level

  
Edit prompts for image 1 and image 2:

1. make the horse blue, goat black and dress white

2. make the weather rainy at night

  
Support for AI png transparent generation! 

Prompt syntax for transparent image generation : 

This is an RGBA image with transparency. PROMPT The image has alpha channel and the background is transparent.

support local mask editing

Recognize a lot of IP characters

using Qwen3Vl-8b obviously supports very complex and long prompts

Focus between foreground/background can be deep (all image in focus, check the landscape image)

  
Facial expressions and emotions are full rendered! 

  
Things to improve:

1. sometimes images with too much contrast and deep blacks, obviously will be easily corrected using a good ComfyUI workflow

2. saturation on skin can be improved, maybe 10% less for reds

all images were generated with modelscope API 

  

```



document fetched 2026-09-21 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1wlgv0l/i_tested_qwenimage21_its_amazing
