Image generation
Available now

Qwen Image 2.1

by Alibaba Qwen

Alibaba's open‑weight image model.
Sharp type, steady identity, edits from 10 references.

Qwen Image 2.1

Key features

Transparency, local mark-up edits and the portrait and texture gains are Alibaba's launch claims for Qwen Image 2.1 (September 20, 2026).

Technical specifications

7B

One 7-billion-parameter model handles both generation and editing.

Up to 10

Input images per edit, on Morphic and in Alibaba's release.

Up to 2K

Output on Morphic goes up to about 4 megapixels, keeping your aspect ratio.

8 ratios

From 21:9 wide to 9:16 tall on text to image, plus Auto on edits.

Use cases

A deep blue concert poster with the headline NIGHT TIDE and a smaller line of event details, lying on concrete

Posters with exact type

Set the headline and the small print in the prompt, and the layout comes back with every word spelled as written.

A sage green tea tin labeled Morning Field, jasmine green tea, 100 g, beside a ceramic cup in window light

Packaging and labels

Product labels carry a brand name and a line of detail in clean type, ready for a mockup or a listing image.

A woman with short dark curly hair in a rust sweater, lit by soft light from a tall window

Portraits in natural light

Soft window light, real skin texture and a calm expression, for editorial portraits that read as photographs.

The same woman from the window portrait, now in a navy raincoat on a rainy city street at dusk

Same person, new scene

Use one portrait as the reference and move the same face into a new setting, outfit and light.

Close-up of a stoneware bowl with a crackled celadon glaze and a raw clay foot on weathered oak

Product detail and texture

Close-ups hold crackle in a glaze, grain in wood and the grit of raw clay, for product pages that sell on material.

Three die-cut stickers in a row, a smiling fox head, a steaming coffee cup and a potted cactus

Stickers and cutout assets

Die-cut stickers, icons and isolated objects with clean edges, ready to drop onto a layout, a product or a social post.

Prompt examples

Poster with type

A minimalist concert poster, headline NIGHT TIDE in bold condensed type

Edit prompt

Product label

A sage green tea tin, label reading Morning Field in an elegant serif

Edit prompt

Window portrait

Editorial portrait by a tall window, soft light, natural skin texture

Edit prompt

Same face, new scene

Keep this exact woman, same face and hair, on a rainy street at dusk

Edit prompt

Texture close-up

Macro close-up of a crackled celadon glaze bowl, raking side light

Edit prompt

Sticker set

Three die-cut stickers, a fox, a coffee cup and a cactus, flat vector

Edit prompt

Overview

Qwen Image 2.1 is Alibaba's open-weight image model, released by the Qwen team on September 20, 2026. One 7-billion-parameter model handles both jobs, generating images from text and editing the ones you already have, so the same model that draws a poster can also change its colors, swap its subject or lift it onto a transparent background.

What Qwen Image 2.1 brings

The headline is control. Text renders as written, so posters, labels and packaging come back with the exact words in the prompt. Alibaba also built the model to output transparent images natively, rather than through a separate background removal step, and to keep a person or a product recognizable across edits. Portrait lighting and fine surface texture are the other two areas Alibaba lists among its gains.

How editing works

Editing takes up to 10 reference images at once, which covers jobs like building a group photo from separate portraits or dressing one model in several garments. For a change to one part of the frame, you can mark the area with a drawn circle, a painted note or a separate mask, and the model edits that region while the rest of the image stays put.

Open weights

Alibaba publishes the weights on Hugging Face under the Qwen Research License, and the model is sized to run on a single consumer graphics card. That makes it a common choice for teams that want to run image generation on their own hardware, as long as the license terms fit their use.

Qwen Image 2.1 and Morphic

Qwen Image 2.1 runs on Morphic for text to image and image editing, including on the free plan. Edits take up to 10 reference images, text to image offers eight aspect ratios from 21:9 to 9:16, and output goes up to 2K. It sits in one workspace with GPT Image 2.5, Nano Banana 2, Seedream 5.0 Pro and the rest of the image line, so you can compare the top AI image models on best use, text accuracy and resolution before you pick one.

Simple pricing

Get started for free today, with the option to upgrade or cancel anytime.

Basic

$19/ month
billed as $0 per year

2400 monthly credits

1 user only

All models

Workflows

Standard

$24/ month
billed as $0 per year

3625 monthly credits

1 user only

All models

Workflows

Pro

$45/ month
billed as $0 per year

6350 shared monthly credits

1 user

+ up to 4 more at extra cost

All models

Workflows

Max

$170/ month
billed as $0 per year

24650 shared monthly credits

1 user

+ up to 9 more at extra cost

All models

Workflows

Enterprise

For higher limits

Custom

pricing and billing terms

High-volume credits
Custom seat limits
All models
Workflows
Pricing Gradient

Free

For playing around

$0

forever free

Up to 20 credits
1 user only
Limited models
Workflows
Compare plan details

FAQs

What is Qwen Image 2.1?
Qwen Image 2.1 is an image model from Alibaba's Qwen team that generates images from text and edits existing ones in a single 7-billion-parameter model. Its headline strengths are accurate typography, native transparent output, identity that holds across edits, and editing from up to 10 reference images.
When was Qwen Image 2.1 released?
Alibaba released Qwen Image 2.1 on September 20, 2026, with open weights on Hugging Face and support in common local tools on the same day.
How good is Qwen Image 2.1 at text in images?
Typography is one of the improvements Alibaba highlights. Give the exact words in quotation marks and say where they go, for example a headline and one line of small print, and the model lays them out as written. Short, specific text works best.
Can Qwen Image 2.1 make transparent images?
Yes, as a model. Alibaba built it to return an image with a transparent background straight from a prompt, to extract a subject from a photo, and to edit transparent layers.
How many reference images does Qwen Image 2.1 accept?
Up to 10 in one edit, both in Alibaba's release and on Morphic. That is enough to combine several people into one group shot, dress a model in several garments, or merge a product, a setting and a style in one pass.
Is Qwen Image 2.1 open source?
Its weights are open and published under the Qwen Research License. Read the license on the model page before you build on the weights yourself, since it sets its own terms for use.
Can I use Qwen Image 2.1 for free on Morphic?
Yes. Qwen Image 2.1 runs on the free plan, for both text to image and edits from up to 10 reference images. Pick it in the model menu, choose one of eight aspect ratios from 21:9 to 9:16, and images come back at up to 2K.

Other models

FLUX Video Upscale

Black Forest Labs

Black Forest Labs' video upscaler. Up to 4K, with fine detail rebuilt and the sound kept.

FLUX 3 Max

fal.ai

fal's post-trained FLUX 3. Tuned to follow the script, built for speed.

Krea 3

Krea AI

Krea's fast image editing model. Combine up to 4 images in one edit.

OmniHuman 2.0

ByteDance

ByteDance's audio-driven human video model. One photo, one track, five minutes in one take.