Qwen-Image-2.1 Text to Image: Wide, Tall and Transparent

Last verified: November 1, 2026

Qwen-Image-2.1 turns a written description into a finished still image. Vidofy's form lets you frame that image as an ultrawide 21:9 scene or a tall 9:21 vertical, as well as the familiar ratios. You can also ask for a see-through background, which makes Qwen-Image-2.1 text-to-image work a practical fit for website headers, cinematic stills, phone-first story art and stickers that go straight into another layout.

Qwen-Image-2.1 focuses on how a picture is finished: typography, portrait lighting and fine detail. The same model can return a regular image or a transparent one, depending on what the prompt describes. For a working creator, that means one generator can make both the hero visual for a campaign and the cut-out asset that sits on top of it.

On Vidofy you control it through a short form: a prompt box with an optional enhancer, a frame shape, a resolution level, a file format, a background choice and a seed for repeatable runs. Decide on the frame and the background first, because those two choices shape how you write the rest of the prompt.

Explore more text to image

Text To Image About Qwen Image 2.1 Use in Image To Image GPT Image 2 Kling Image O3 Hunyuan 3.0 Ideogram V3 Imagen 3 Imagen 4
Capability Snapshot

Qwen-Image-2.1 at a Glance

What goes in, what comes out and which settings you can change.

Input

A text prompt of up to 5,000 characters

Resolution

1K or 2K

Aspect Ratio

1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, 9:21

Output Format

PNG, JPEG or WEBP

Background

Opaque or Transparent

Output

1 image per run

How Do You Get Better Results from Qwen-Image-2.1?

Expand a Rough Idea, Then Edit the Expansion

Qwen-Image-2.1's creators recommend expanding short prompts into detailed descriptions for the best results, and they publish a dedicated rewriting model for that job. Treat the expanded version as a draft. Take a suggestion from the Enhance Prompt button, cut any prop or style you did not intend, and add the details only you know: brand colors, exact lettering, the mood the image has to carry. You get the level of detail the creators recommend while keeping the creative decisions yourself.

Choose the Frame Together with the Composition

The official text-to-image prompt rewriter returns a recommended aspect ratio along with the expanded prompt, so the frame shape and the scene description are meant to be planned together. Pick the Aspect Ratio first, then write the layout for that canvas. For the ultrawide frame, spread a horizon line and subjects from left to right. For the tallest vertical, stack elements from foreground to sky. For square, center a single subject. That way the composition you describe matches the canvas the model fills.

Lock the Seed While You Refine the Wording

Fixing the seed makes Qwen-Image-2.1 outputs reproducible, which turns prompt editing into a controlled test. Enter a number in Seed before your first run and keep it while you change one phrase at a time, such as a color, a lighting note or a line of lettering. Any difference you see then comes from your words, not from a new random start. Change the number only when you want a completely new layout.

How to Make an Image with Qwen-Image-2.1

Four steps take you from the prompt box to a downloaded image.

1

Write Your Idea in the "Prompt" Box

Describe the subject, setting, light and style in plain sentences. The box holds up to 5,000 characters, which is plenty for a detailed scene brief. Under the box, the "Enhance Prompt" button offers fuller versions of a rough idea. Among the settings, the "Enhance Prompt" switch (marked Optional) is on by default. Leave the switch on if your idea is only a line or two.

2

Set "Aspect Ratio", "Resolution Quality", "Output Format" and "Background"

"Aspect Ratio" starts at 1:1 and also offers 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9 and 9:21. Use 21:9 for website headers and cinematic banners, and 9:21 for tall phone-screen art. "Resolution Quality" offers 1K (the default) and 2K. Draft at 1K while you settle the composition and wording, then switch to 2K for the version you keep, so small lettering and fine texture get more pixels. "Output Format" offers PNG, JPEG and WEBP. "Background" defaults to Opaque, and Transparent needs PNG or WEBP, so set the format first when you want a cutout.

3

Click "Generate"

Each click produces one image, so plan on a few runs while you explore. If you may want to repeat this exact result later, type a number into "Seed" before you click.

4

Download the Result

Before you save, view the image at full size and read every line of lettering. For a cutout, check the edges against both a dark and a light backdrop in your editor. The file comes in the format you set in the previous step, so if you need a different one, change "Output Format" and generate again.

Should You Use This Form or Run the Open Weights Yourself?

The model's weights are public, so the real choice is between this browser form and a setup you run yourself.

Getting started You write a prompt, choose a few settings and click Generate in the browser, then download the finished image. Running the open weights yourself means setting up the Diffusers pipeline the creators provide, which handles text-to-image in a single pipeline. Choose the form if you want images without managing software, and self-host only if you already maintain a Python image pipeline and want the model inside it.
Expanding a short prompt The Enhance Prompt button and switch expand a rough idea right on the form, and the button shows its suggestions before anything changes. When you self-host, you run the official prompt-rewriting model as a separate step, and it returns a detailed English prompt plus a recommended aspect ratio. Use the form when you want expansion as a one-click step, and the separate rewriter when you want to script expansion and ratio choice into a batch workflow.
Transparent assets You set Background to Transparent with an output format that supports it, and describe a subject with no scene in the prompt. The reference code requests transparency through a recommended prompt wording and saves the image as RGBA when the model generates it. Use the form for one-off stickers and product cutouts, and a script when you need many transparent assets generated with the same wording.
Resolution Resolution Quality gives you a lower level for drafts and a higher one for finals, chosen for each run. A self-hosted run defaults to the model's native 2K level, with a recommended width and height listed for each aspect ratio. Stay on the form if a draft level and a final level per image are enough, and self-host if your production already runs the model inside your own tools.
For one-off images, quick ratio changes and transparent assets with no setup, use the form. Move to the open weights only when the model has to live inside your own pipeline.

Fixing Common Qwen-Image-2.1 Problems

What usually goes wrong with text-to-image runs, and what to change before you generate again.

I picked a transparent background, but it won't work with my format or the cutout still has a backdrop

The Transparent background only works with two of the three output formats, so choose one of those in Output Format first. If the image still has scenery behind the subject, the prompt is probably describing a setting. Qwen-Image-2.1 decides from the wording whether to return a regular image or one with an alpha channel. Its creators recommend placing the subject description between "This is an RGBA image with transparency." and "The image has alpha channel and the background is transparent." Remove any mention of a room, floor or sky, add those two sentences around your subject, keep Background on Transparent and run it again.

The lettering in my image is misspelled or in the wrong place

Qwen-Image-2.1 treats text as part of the layout and weighs type style and placement against the whole composition, so vague wording leaves it guessing both. Put the exact words in double quotes, say what kind of type they use and where they sit (top banner, sign above the door, bottom caption), and keep each line short. Use the higher Resolution Quality level for the final so small letters get more pixels, then run it again. If one word still comes out wrong, shorten that line instead of rewriting the whole prompt.

The image has details I never asked for

The Enhance Prompt switch is on by default and lets the model expand your prompt before generating, which can add props, colors or a setting you did not write. If your prompt is already detailed, turn the switch off and run it again. If you still want help with the wording, use the Enhance Prompt button instead. It only offers suggestions and leaves your text alone until you pick one.

My long prompt won't fit in the box

The prompt box has a character limit, so a very long brief has to be trimmed. Cut repetition first: stacked adjectives, duplicate style tags, the same lighting described twice. Keep the subject, composition, lighting and any quoted lettering, because those are what shape the image. Then generate again with the shorter version.

I only get one image and want several options to choose from

Each run returns a single image, so variations come from repeat runs. Keep the prompt and settings the same and change the number in Seed between runs to get different takes on the same idea. Because the same seed reproduces a result, you can return to the number that gave the best take once you start refining the wording.

Frequently Asked Questions

Can Qwen-Image-2.1 make illustrations as well as photorealistic images?

Yes. The style comes from the prompt. The model's own quick-start example for transparent output asks for a cartoon dragon sticker, and for portraits the model brings improved lighting and fine detail for a realistic look. Name the medium early in your prompt (flat vector, gouache, 35mm photograph, 3D render) and add one or two details of that medium, such as brush texture, film grain or a lens, so the style is clear.

What's the difference between the Enhance Prompt button and the Enhance Prompt switch?

They expand your prompt at different moments. The button under the prompt box offers fuller versions of a rough idea and leaves your text alone until you pick one, so you can read, keep or trim what it adds. The switch works at generation time and lets the model expand the prompt itself before it renders. Use the button when exact wording matters, and rely on the switch when a quick idea is all you have.

What makes one Qwen-Image-2.1 image cost more than another?

The credit cost of a run is calculated from the options you choose and shown above the Generate button once you are signed in, so you see it before you generate. Of the settings here, Resolution Quality is the one that changes it. The Enhance Prompt button does not use credits.

Does Qwen-Image-2.1 render 2K natively or upscale it?

Qwen-Image-2.1 natively supports 2K resolution. Its recommended sizes include 2048 × 2048 for square and up to 2752 pixels on the long side for 16:9 and 9:16, and its default settings use that 2K level. On the form, the higher Resolution Quality level is the one to choose when small type or fine texture has to hold up at full size.

References

Sources and citations used to support the content provided above.

Updated: 2026-11-01 13:39:09 2 Sources

www.alibabacloud.com

Source Link
https://www.alibabacloud.com/blog/qwen-image-2-1-compact-efficient-and-unified-image-creation_603586

github.com

Source Link
https://github.com/QwenLM/Qwen-Image-2.1