Create with Qwen Image 2.1
Turn a visual brief into an image, or bring reference pictures into the same workspace for a more directed edit.
Read the official model announcement
What is Qwen Image 2.1?
Released by Alibaba's Qwen team on September 20, 2026, Qwen Image 2.1 is an open-weight model for both image generation and editing.
Generation and editing are built into one model, rather than two unrelated engines. Its visual generation component has 7 billion parameters. Qwen also introduced native transparent (RGBA) images alongside detailed visual composition and typography. On Kuca, text-to-image starts with a prompt, while image-to-image can work with up to 10 uploaded references to guide an edit.
Qwen's official introduction highlights visual detail, typography and image editing. On Kuca, both text-to-image and image-to-image live in one generator. Start with a description for a new scene; attach references when the subject, composition or material should follow images you already have.
A useful result depends on the request as much as the model. Describe the subject, action, setting and visual treatment before adding minor adjectives. If text must appear, quote the exact words, describe their placement and inspect the final letters carefully. An image model can still misspell or distort text, especially when a design contains many small labels.
Editing is useful when a photograph or draft already carries most of the right information. Ask to change one thing at a time and name what should remain unchanged: crop, silhouette, face, lighting or background. Keep the untouched original to compare alternatives. Review every final image before publishing or handing it to a client.
Start with words or existing images
Choose the starting material that actually exists. Both variants share the Qwen model family and its creation history in the workspace.
Text to image
Write a visual brief without uploading a reference. Define the subject, composition, environment and style, then choose a ratio and 1K or 2K output. Use this mode to explore an initial direction for a poster, product concept or moodboard before committing to a source image.
Edit with reference images
Upload up to 10 source images to guide an edit. Explain which source supplies the subject, which supplies the look and which details must survive. Each file can be up to 30 MB in the Kuca uploader. The model may still reinterpret fine detail, so compare output with the originals.
Treat lettering as a design requirement
Qwen highlights image typography in its official introduction. Written content can be part of a prompt, but it is not a substitute for checking the output. Keep the message short, specify where it belongs and verify every visible character before using the image in an ad, menu or product listing.
Quote the exact copy
For a poster or packaging concept, put the required phrase in quotation marks. Include only the words that need to be visible and describe their hierarchy separately from the scene. Long paragraphs and tiny legal copy are better added in a design editor later.
Describe the surrounding layout
Say whether lettering sits above the subject, on a clean label or in the background. Reserve empty space if a designer will add the final words later. A usable composition matters more than forcing every possible word into one image.
Check at the intended size
Read the result at full size and at its actual display size. A small preview can hide crooked characters and invented labels. For precise copy, use the generated image as an art direction base and set final type in a dedicated editor.
A prompt that leaves less room for guesswork
Build your request from observable parts. The product still life and lettered poster above were made with Qwen Image 2.1 on Kuca. Try the sample prompt below, then adjust the subject, light or materials to make it your own.
- 1
Identify the main subject
Name the object or scene before naming a style. Include the action if people or moving elements are involved. A specific subject gives the model a clear center of attention and makes revisions easier to describe.
- 2
Set the composition
Add the crop, camera angle and relationship between foreground and background. Say where empty space should remain if text will be added later. For edits, identify which reference determines the layout.
- 3
Finish with light and materials
Choose a few compatible constraints such as overcast light, textured paper, polished metal or a limited palette. If the result misses one detail, change that instruction instead of rewriting the whole request.
Example prompt
A premium editorial product photograph for a contemporary botanical skincare campaign: one unbranded frosted-glass serum bottle with a simple cream label on a warm limestone plinth, a folded cobalt-blue translucent fabric draped behind it, a single glossy green leaf casting a crisp shadow. Soft late-afternoon window light from the left, refined tactile materials, warm ivory and muted cobalt palette, believable studio photography, generous negative space, no logos, no text, no watermark. Wide horizontal 3:2 composition.
Choose the shape and finish
Set a ratio and 1K or 2K resolution. Choose a solid background with PNG, WebP or JPEG, or a transparent background with PNG or WebP. The credit estimate updates as you change resolution.
- 1K or 2K
- Select the output tier before generation. A larger tier provides more pixels for inspection or layout work, but uses more credits. Exact pixel dimensions depend on the selected ratio, so inspect the downloaded file when a publisher requires a specific width.
- Shape and format
- Pick square, portrait or landscape to match the destination. For a cutout, select Transparent background and PNG or WebP, then describe the subject without asking for a scene behind it. JPEG is available for solid-background images.
- Reference inputs
- Image editing accepts up to 10 JPEG, PNG or WebP references, each no larger than 30 MB or 25 megapixels. Use a purposeful set, assign each reference a role and say which source must remain visually dominant.
- Credits and repeatability
- Kuca displays the credit estimate next to Generate and updates it when resolution changes. An optional seed can help repeat a setup, but the same seed cannot guarantee identical pixels across different prompts, inputs or model revisions.
From first draft to usable image
Treat the first output as a draft. A short review loop is more useful than an overloaded prompt full of competing corrections.
Draft
Start from words or references, choose the destination ratio and state one clear objective. Keep the prompt and source images together so the direction can be reconstructed.
Inspect
Open the result at full size. Check composition, text, hands, product labels and image edges. Compare an edited result with its untouched source before accepting a change.
Refine and reuse
Change the smallest failed instruction and run another version. Download the chosen image, keep the source and use workspace history to revisit earlier experiments.
Questions before you generate
Can I use Qwen Image 2.1 without an input image?
Yes. Select text-to-image, describe the subject and composition, then generate from the prompt. Use image editing only when an existing picture should guide a change. Both variants appear in the same Qwen family on Kuca.
How many reference pictures can I upload?
The Kuca editing form accepts up to 10 reference images, with a 30 MB limit per file. Explain the job of each image in the prompt. Several near-identical references may make the requested change less clear.
Does the model guarantee accurate text in an image?
No. The official model introduction highlights typography, but generated letters still need review. Quote the wording, describe placement and read the result at full size. Add legally required or precise copy in a design editor if needed.
Which sizes and file formats are available here?
Choose 1K or 2K, a common aspect ratio, and PNG, WebP or JPEG. Transparent backgrounds require PNG or WebP. Pixel dimensions vary with the ratio; review the displayed credits before generating.
Can I make a transparent image or edit with a mask?
Choose Transparent background and PNG or WebP to generate an image with an alpha channel. Describe the subject without a background scene, then inspect the download. Mask-based local editing is not yet available in this workspace.
Make a first image, then decide what to refine
Open the generator above, choose a mode and start with one clear prompt. Compare Qwen variants and review your results in the shared workspace history.