GitHub radar

Qwen-Image-2.1 is now available for home use

The unsloth team has repackaged the recent Qwen-Image-2.1 released by Alibaba to a smaller local file that allows transparent backgrounds and up to ten reference images for editing.

01unsloth/Qwen-Image-2.1-GGUF 207130k downloads/mo7B paramstext-to-image

Qwen-Image-2.1 repackaged by unsloth as a small local file. This model was created by Alibaba’s Qwen team. It is an image generation and editing model based on textual input, generating images up to 2048x2048 pixels. Compared to the original, the repackaged file uses significantly less disk space and can easily be used in local applications like ComfyUI. The model generates images with transparent backgrounds, which can then be layered on top of other images without further editing in a photo editing program. You can also use up to ten reference images for editing, which will all be merged into a single scene. The original model was released a couple of days ago, and now it is available as a pre-packaged installable model for your local device.

Why a vibe-coder should care

Once upon a time, getting an image this good would have cost a monthly fee or per-cloud-call rate. Now, you can install it on your own machine at no charge per request. The advantage is that you can edit images with multiple reference images, and you can generate images with transparent backgrounds, which will significantly cut down on design time and content production. As always, the disadvantage of a local model is that it won’t reach the same level of quality and efficiency as the best commercial products (such as Midjourney or the built-in image generators in the top chatbots), but if you need to create rough drafts or frequently repeat tasks, it’s perfectly serviceable.

How to install

Copy this and send it to your agent — Claude Code, Codex, any of them:

Set up the image generation model Qwen-Image-2.1 in its compressed version from unsloth for me: https://huggingface.co/unsloth/Qwen-Image-2.1-GGUF — pick the right file for my laptop, run it locally (for example through ComfyUI), and generate a test image with a transparent background from my description.

A regular laptop with 16 GB of RAM or more — the compressed version of the model takes about 5 GB.

Open on Hugging Face