# I made a no-settings photo editor for Qwen-Image 2.1: brush an area, say what you want, it stays inside the brush — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/f1d5257081f9ef27

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-14 15:04 ingested 2026-10-15 02:40

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1x5tmlp/i_made_a_nosettings_photo_editor_for_qwenimage_21/)

## retained text



```
Not a replacement for node workflows: something for the times you just want to fix a photo, or for friends who will never build a workflow.

Kokon is a free, open-source desktop editor (Windows; Mac beta). Install it, it downloads the right model for your card, and from then on there's nothing to configure: it picks the resolution and memory split per job. 12 GB of VRAM is enough.

The part this sub may care about: Qwen-Image 2.1 has no mask input, so Kokon draws the brushed area on a second reference image: the model sees exactly where to work, so "remove the red scooter" removes the scooter and leaves the street alone. On 16 GB+ cards edits run natively up to 2K, no upscaling pass.

Every image here was made with it, prompts in the captions. The first one is a six-step project (generate, edit, combine two images, expand); the others are single edits.

Under the hood: Qwen-Image 2.1 Turbo (Unsloth GGUFs) + Qwen3-VL text encoder, running on stable-diffusion.cpp. No ComfyUI or Python needed.

[https://github.com/GabrieleLozupone/kokon](https://github.com/GabrieleLozupone/kokon)

Not there yet: Linux, LoRAs, other models. (Weights: Qwen Research License, non-commercial.)


```



document fetched 2026-10-15 00:34 · canonical https://reddit.com/r/StableDiffusion/comments/1x5tmlp/i_made_a_nosettings_photo_editor_for_qwenimage_21
