What is Qwen Image 2.1?
Released by Alibaba's Qwen team on September 20, 2026, Qwen Image 2.1 creates images from text and edits them with up to 10 reference images. It natively supports 2K output and is designed to render text within images more clearly, though you should still check every word. On this site, choose 1K for 10 credits or 2K for 20 credits.
The useful question is not whether a model can make an attractive first draft, but whether you can turn a clear brief into an image you can actually use. Begin with the subject, the setting, and the visual purpose. Then decide whether a reference image is necessary: a fresh composition does not need one, while a targeted edit usually does. The tool sits at the top of the page so you can try the workflow as you read, rather than following a tutorial before you can begin.
Choose the right starting point
For a new image, use text to image. Describe the subject and what it is doing, then add the environment, lighting, camera viewpoint, and style only where those details matter. A short, coherent brief usually gives you a clearer first result than a list of unrelated adjectives. For an edit, switch to image to image and upload the photograph or illustration you want to change. Tell the tool which part to revise and what must remain. These are different creative tasks: one asks the model to invent a scene; the other asks it to respect an existing composition. If you have several images, explain the role of each reference instead of expecting the tool to infer which one supplies the person, which one supplies the palette, and which one supplies the background. Start with a single important reference when possible, then add another only if it genuinely resolves ambiguity.
Generation and editing in one workspace
The workspace lets you write a prompt immediately. Select text to image for a new scene without uploads. Select image to image before adding references for a revision. The output controls let you choose an available aspect ratio, resolution, and file format. A square frame can suit a catalog tile; a horizontal frame leaves room for a landscape or banner; a vertical frame can suit a portrait composition. These settings affect framing, not the accuracy of every object or word inside the image. Before submitting, look at the credit amount displayed next to the action button. If you are exploring an idea, begin with a modest output size and save the higher-resolution choice for a concept worth keeping. When the result appears, inspect it at a useful size before deciding whether to download it or refine the prompt. Keep the source image handy if you plan several rounds of editing.
Write a brief that can be checked
A good prompt gives the model a visual hierarchy. Start with the main subject and its action. Add the setting, then specify the most important visual treatment, such as soft window light, a restrained color palette, or a low camera angle. For example, ask for a ceramic coffee cup on a dark wooden table, morning side light, and empty space on the right for a headline. That is more actionable than asking for a beautiful professional picture. If text must appear in the image, quote the exact wording and say where it belongs; keep the wording short, and verify the letters in the result. For a series, write down a few stable requirements such as crop, background, and brand colors. Change one variable at a time between attempts. This makes it easier to see whether an improvement came from the lighting instruction, the composition instruction, or a different reference. Avoid conflicting directions such as minimal background and densely decorated backdrop in the same request.
Make a controlled image edit
When editing an existing image, the most useful instruction names both the change and the boundary around it. Try: replace the blue packaging with a warm cream color, keep the logo position, camera angle, and shadows unchanged. That tells the model where to focus and what not to reinterpret. If the reference has a person, be especially precise about the face, clothing, pose, and background details that should stay recognizable. If you provide multiple references, identify them by their order in the upload area and describe their roles in ordinary language. Ask the tool to take the product shape from the first image and only the color palette from the second. After generating, check the image against the original at full size; small accessories and edges can change even when the main subject looks right. If too much moved, simplify the instruction and restate the elements to preserve. An edit is an iteration, not a guarantee of pixel-perfect preservation.
Plan for the final layout
Think about where the image will appear before choosing a frame. A thumbnail may need a simple focal point and generous contrast; a landing-page banner often needs negative space for separate interface text; a poster concept needs enough visual room around any headline. Describe these needs in the prompt, then choose the closest available aspect ratio in the workspace. Do not rely on generated lettering as the sole source of important information. For a campaign image, it is often safer to create the visual without a long block of copy and place the final approved text in your design tool. Review the result at the size at which people will see it, not only as a large preview. A detail that is striking on a monitor may disappear on a phone. If the composition is crowded, remove secondary props or ask for a simpler background before increasing the resolution. More pixels cannot repair unclear visual priorities.
Review before you publish
Treat each generated image as a draft requiring a human check. Compare it with your original brief: is the subject right, is the composition useful, and did the edit leave the important source details intact? Zoom in on hands, product labels, small typography, repeated objects, reflections, and the edges of edited regions. If you asked for exact writing, read every word rather than assuming it is correct because the overall layout looks plausible. Check that any reference image you uploaded is one you have permission to use. For customer-facing creative work, review brand requirements and any claims displayed in the image separately from visual quality. Save the successful prompt and the source references so a later revision is easier to reproduce. If a result misses the mark, change one specific instruction and try again. A focused second brief usually teaches you more than adding another paragraph of vague style words.
A practical first run
Start with a single goal: for example, a product photograph with a clean background, or a color adjustment to an existing poster. Enter one or two sentences that describe the desired outcome. For a new composition, leave the reference area empty. For an edit, choose image to image and upload the relevant image before writing the change. Select a frame that matches the intended destination and use the displayed settings rather than assuming every shape or size is available. Read the credit charge before you generate. When the result arrives, judge subject accuracy first, then framing, then finer details such as texture and lettering. If the subject is wrong, rewrite the main sentence; if the framing is wrong, change the aspect ratio or camera instruction; if only one detail is wrong, ask for that detail to change while preserving the rest. Download the version that passes your own review, and keep earlier drafts only when they help explain a useful creative direction.
Settings, credits, and expectations
The workspace displays the credit charge for the selected output before you submit. A larger output can cost more credits, so choose resolution according to how you intend to use the image. A quick composition test does not need the same settings as a final asset. Available choices can differ between models; use the controls on this page as the source of truth for this session. Generating an image is not the same as buying exclusive rights to every object, logo, or reference in it. Check the relevant terms and your own rights to any uploaded material before commercial publication. Nor should a model result be treated as a verified statement about a real product or person. If a result includes specifications, prices, names, or other factual text, verify those separately. The best workflow combines a clear prompt, an appropriate output size, and a final editorial review instead of treating the first render as automatically ready to ship.
Qwen Image 2.1: how to use the tool
How do I start with Qwen Image 2.1?
Type a clear description in the prompt field, choose text to image, review the available output settings and credit charge, then generate. For a first attempt, describe one subject, its setting, and the light rather than combining several unrelated scenes.
Can I upload a photo to edit?
Yes. Select image to image, upload a reference, and describe exactly what should change. State which parts should remain unchanged, then compare the result with the source at full size.
How should I ask for text in an image?
Put the exact words in quotation marks and specify their position. Keep the phrase short and check the final spelling yourself. For important production copy, consider placing text separately after generating the visual.
Which frame and resolution should I choose?
Choose a frame for the intended placement, such as square for a tile or wide for a banner. Start with a lower resolution while exploring, then choose a larger available output for a final candidate. The workspace shows the charge before submission.
What if an edit changes too much?
Rewrite the prompt to name the single change and the elements to preserve. Remove unnecessary references and compare the next result with the original. Some details may still need manual finishing.