Qwen Image 2 Pro Image to Image for Text and Style Edits

Last verified: October 19, 2026

Qwen Image 2 Pro carries its text rendering into editing. It can add lettering to a photo you already have, such as a title, a caption or a line of verse in brush script, while keeping textures realistic and sticking closely to what the instruction asks. To use it, add the picture in Image, describe the change in Prompt and set Aspect Ratio. Next to the usual shapes, the wide 2:1 and tall 1:2 frames cover banners, story covers and posters. In Qwen Image 2 Pro image-to-image work the written instruction is your main control: the more clearly it separates what changes from what stays, the closer the result lands.

Lettering is only half of a convincing edit. Qwen Image 2 Pro also renders fine, realistic textures and detailed photographic scenes, and it follows the meaning of a prompt closely. That matters in two places: the untouched parts of the picture still have to look like a photograph, and an instruction with several parts (the change, where it goes, its style) needs every part to show up in the result.

Explore more image to image

Capability Snapshot

Inputs, Aspect Ratios and Output at a Glance

The basics of the editing form, from what you upload to what each run gives back.

Input

Image (JPG, JPEG, PNG or WEBP, up to 10 MB) + a text prompt of up to 2,048 characters

Aspect Ratio

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

Output

1 image per run

Prompting Techniques for Cleaner Photo Edits

Saying What Must Stay Unchanged

In Prompt, spell out the parts of the photo you want protected as plainly as the change itself. One reference edit for Qwen Image 2 Pro adds illustrated figures to a city street. Before it does, the prompt rules out changes to the real buildings, streets, vehicles and people and asks for the photo's realism to be kept. That pattern is worth copying after any instruction. Name the actual things in your frame, such as the face, the logo or the tabletop, rather than a vague 'everything else'. If one unwanted detail keeps appearing anyway, put it in Negative Prompt instead of making the main instruction longer.

Placing Changes Relative to Objects in the Frame

Tell Qwen Image 2 Pro where an edit goes by pointing at something already in the picture, and give its size as a share of the frame. The model's reference prompts work this way. Illustrated figures are placed lying on top of a building, peeking out from its right side and sitting on the open ground in front, and added text is sized to about a tenth of the image height. These anchors keep their meaning whichever Aspect Ratio you pick, because they refer to the content, not to the edge of the canvas.

How to Edit a Photo with Qwen Image 2 Pro

Five steps from uploading your photo to downloading the edit.

Choosing Between Qwen Image 2 Pro and the Standard Tier

How the two Qwen Image 2 editing tiers differ on the point that matters most for photo edits.

Lettering and photographic finish This Pro tier is built for stronger text rendering, realistic textures and close semantic adherence, which are what a lettered or photo-real edit relies on. The standard Qwen Image 2 model is the faster version of the same family, built to balance quality against response speed. Choose Qwen Image 2 Pro when the edit adds text or has to pass as a photograph, and the standard tier when quick rough drafts matter more than the final finish.
For finished edits where lettering or photographic surfaces have to hold up, Qwen Image 2 Pro is the tier built for the job. The standard model suits fast exploratory passes.

Fixing Common Photo Editing Problems

What usually goes wrong with an edit, why it happens and what to change before the next run.

The edited picture comes out soft or blurry.

The usual cause is a small source image. The model edits best with inputs between 384 and 3072 pixels on each side, and a low-resolution input can produce a blurry output. Image accepts files a little below that range, so a thumbnail uploads fine but still comes back soft. Re-upload a full-size original instead of a screenshot or a copy from a messaging app, keep the same prompt and generate again.

My photo won't upload.

Image only accepts a few common web formats and has a file-size cap and a minimum resolution. That means HEIC files straight from a phone, TIFF or BMP scans, very large PNGs and tiny thumbnails are rejected. Export the picture to one of the formats listed in the upload FAQ, get it under the size cap and add it again. Your prompt and settings can stay as they are.

The Prompt box won't take any more text before I've finished the instruction.

The Prompt box has a character limit, and long edit briefs usually use it up describing what is already visible in the photo. Drop the scene description and keep three things: the change, where it goes and the style it should take. Then add a short list of what must not change and generate again with the shorter version.

The result is cropped or has new content around the edges.

Aspect Ratio starts on a square frame, so a landscape or portrait photo has to be recomposed to fill it. When the model is called directly on Alibaba Cloud without a size, it keeps the output close to the input's shape, which is why matching the frame to your photo gives the most faithful edit. Choose the Aspect Ratio nearest your original's proportions and run it again. Save the wide and tall frames for edits where a new composition is the point.

I want a few versions to choose from, but each run gives me just one.

You get variations by running the same edit several times. Keep the prompt as it is, enter a different number in Seed for each run, and note the number behind the version you like. Keeping that Seed fixed while you fine-tune the wording keeps your comparisons fair.

Frequently Asked Questions

What kinds of edits can Qwen Image 2 Pro make to a photo?

It works from a plain-language instruction. It can change text in the image, add, delete or move objects, change what a subject is doing, transfer a new style and enhance details. Changing a subject's action and enhancing details are the easiest to overlook. You can ask for a person to sit instead of stand, or for worn textures to look crisper. Describe the result you want, not the steps to get there.

Which image files can I upload?

Image accepts JPG, JPEG, PNG and WEBP files up to 10 MB, and the picture must be at least 300×300 pixels. Both the picture and a prompt are required. If you can, upload a large original rather than one near the minimum: small sources tend to come back soft, as the troubleshooting notes explain.

How is the cost of a run shown?

Once you're signed in, the credit cost of a run appears above the Generate button. It's calculated from the options you've chosen, so you can see it before you start.

Can it replace the text on a sign, label or poster in my photo?

Yes, changing text inside an image is one of the edits the model handles. In Prompt, quote the current wording and the new wording exactly, and say which sign or label you mean if there are several. Ask for the same lettering style, color and position so that only the words change.

Can Qwen Image 2 Pro write Chinese characters onto a photo?

Yes. One reference edit for the model adds a seven-character classical quatrain in semi-cursive script to an ink-wash scene, finished with a small vermilion seal that carries two characters in seal script. In Prompt, put the exact characters in quotation marks, then give the script style, ink color, size and position you want.

References

Sources and citations used to support the content provided above.

Updated: 2026-10-19 23:37:27 4 Sources

help.aliyun.com

Source Link
https://help.aliyun.com/en/model-studio/qwen-image-edit-api

www.alibabacloud.com

Source Link
https://www.alibabacloud.com/help/en/model-studio/qwen-image-2-0-pro

help.aliyun.com

Source Link
https://help.aliyun.com/en/model-studio/qwen-image-edit-guide

www.alibabacloud.com

Source Link
https://www.alibabacloud.com/help/en/model-studio/qwen-image-edit-guide