First Frame *

Upload First Frame

Last Frame *

Upload Last Frame

Prompt: 0 / 2048

Generate

Sample Video

Qwen Image 3.0 Image to Image for Detailed Photo Edits

Last verified: October 19, 2026

Qwen Image 3.0 image-to-image editing works best on edits that add information or restore detail, not quick one-line restyles. Qwen built the model to handle long, information-dense instructions, render very small text and reproduce fine textures like pores and hair strands. The prompt box here is big enough for a full brief, so in one pass you can describe the base image, what must stay, every change and the finish.

Qwen positions the model as a tool for work you can actually use, not just good-looking pictures. Its editing examples bring the same strengths to changes made on an existing image: fine, readable markings, restored texture that matches the original, and new content that draws on real-world knowledge. That makes it a good fit for work-ready visuals, as long as your instruction says exactly what to add and what to leave alone.

On the form, you upload the base image and any supporting references in Add Image, describe the edit in Prompt, and list anything you don't want in Negative Prompt. Aspect Ratio sets the shape of the result and Resolution Quality sets its size tier, so you can refine the wording on quick drafts and save the higher tier for the version you download.

Explore more image to image

Image To Image Seededit V3 Nano Banana 2 GPT Image 2 Seedream 4.5 Edit Qwen Image 3.0 Pro Recraft V3

Capability Snapshot

Qwen Image 3.0 Edit Specs at a Glance

The inputs, sizes and frames available for each edit run.

Input

Up to 3 images (JPG, JPEG, WEBP or PNG, up to 10 MB each) + a text prompt of up to 5,000 characters

Resolution

1K or 2K

Aspect Ratio

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

Output

1 image per run

Writing Multi-Reference Edits That Land

Number Each Reference and Give It One Job

Qwen's editing guide links image numbers in the prompt to the order the inputs arrive: the first file is image 1, the second is image 2, and markers like [image 1] also work. Its own multi-image example follows this pattern: the girl in image 1 wears the black dress from image 2 and sits in the pose from image 3. When you add more than one file in Add Image, write your instruction the same way, and pair each number with what that image shows, such as 'the woman in image 1, the black dress in image 2'. Then the edit stays clear even if you're unsure of the upload order. Give each reference a single role, which keeps the instruction short and easy to check against the result.

State What Stays Before Describing What Changes

Alibaba's own image-to-image example for the 3.0 series starts by telling the model to perfectly preserve the woman's facial features and long black hair. Only after that does it describe the new outfit, the coffee-shop setting, her pose, her makeup and the direction of the afternoon light. Use the same order in Prompt: first lock the elements that must survive (a face, a logo, a product's shape, printed text), then describe the change in concrete terms, then the light and finish. The long prompt box makes this practical, because you have room to describe the parts you're keeping as carefully as the new ones.

How Do You Edit a Photo With Qwen Image 3.0?

Five steps take you from uploading references to downloading the finished edit.

Should You Edit With Qwen Image 3.0, the Pro Model or Qwen Image 2.1?

How Qwen Image 3.0 compares with two other Qwen editing models that Alibaba documents.

Criterion	Qwen Image 3.0	Alternatives	Best For
Speed versus the most complex layouts	Alibaba lists Qwen Image 3.0 as the faster variant of Qwen Image 3.0 Pro and describes it as balancing quality and speed.	Qwen Image 3.0 Pro, on Alibaba Cloud Model Studio, is the model Alibaba recommends first for complex layouts such as newspapers, storyboards, menus and exam papers, as well as 10-pixel text and multilingual fonts.	Pick Qwen Image 3.0 for quick rounds of edits, and test the Pro model when an edit has the densest layout or the smallest type and speed matters less.
How many reference images one edit can use	The 3.0 series takes a compact set of references per edit, the same limit the Add Image upload enforces, which covers a subject plus a couple of supporting images.	Alibaba documents the Qwen Image 2.1 series as accepting up to ten input images in a single edit.	Stay with Qwen Image 3.0 when an edit centres on one subject and a couple of supporting references, and look at Qwen Image 2.1 when a composite genuinely needs more source images than that.

Qwen Image 3.0 is the practical choice for detailed, instruction-led edits on a small set of references when you want a fast turnaround. Qwen Image 3.0 Pro is Alibaba's first recommendation for the densest layouts, and Qwen Image 2.1 is the option when a composite needs many more source images.

Fixing Common Qwen Image 3.0 Edit Problems

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

My image won't upload.

The file is outside what Add Image accepts: it's in a format that isn't on the accepted list, it's too heavy, or it's smaller than the minimum dimensions (the upload FAQ below lists the exact limits). Re-export it in an accepted format, compress it below the size cap or use a larger original, then upload again. Your prompt and settings can stay as they are.

The edit looks soft or loses detail from my photo.

For best results, the model's documentation recommends input images between 384 and 2048 pixels in width and height. The upload also accepts smaller files than that, so a tiny or heavily compressed source leaves less detail to keep. Swap in a larger original within that range, set Resolution Quality to the higher tier for the final, and run the edit again.

My prompt is too long to fit in the box.

The Prompt box has a character limit, and a brief that lists every preserved element and every change can go over it. Cut repeated adjectives and filler, keep one sentence for each element you're locking and one for each change, and move unwanted elements into Negative Prompt instead of writing 'don't…' clauses in the main text. Then paste the shorter version back and generate.

I only got one image and want to compare options.

Each run returns one image, so you get variations by running again. Generate with the same uploads and prompt for another take, or change a single phrase between runs so you can see which wording changed the result. Once a version works, rerun that exact prompt at the higher Resolution Quality tier for your final.

Frequently Asked Questions

How many images can I upload to Qwen Image 3.0, and which files work?

Add Image takes 1 to 3 images per edit in JPG, JPEG, WEBP or PNG format. Each file can be up to 10 MB and must be at least 300×300 pixels, and at least one image is required. That matches the model itself: Alibaba documents the 3.0 series as accepting one to three input images in an image-to-image request.

How long can my edit instruction be?

Qwen Image 3.0 accepts up to 4.5k tokens of input and is built to interpret complex prompts that combine text and images. On the form, the Prompt box counts characters rather than tokens, so its limit (given in the steps above) is the ceiling you'll actually work against. Spend that room on the elements you're keeping and the exact changes, not on stacked adjectives.

What resolution does Qwen Image 3.0 output for edits?

For image-to-image, Alibaba documents a total output area between 512×512 and 2048×2048 pixels for Qwen Image 3.0. On the form you choose the size tier with Resolution Quality and the shape with Aspect Ratio. Use the lower tier while you test wording and the higher one for the version you keep.

What makes one generation cost more than another?

Once you're signed in, the credit cost of a run is calculated from the options you've chosen and shown above the Generate button. It changes with the number of images you upload, so check the figure again whenever you add or remove a reference.

What should I put in Negative Prompt when editing?

Put in the things you don't want to see in the result, and keep the main Prompt for the edit itself. Alibaba's editing guide gives blur and extra fingers as typical examples of content to exclude this way. It's also a good place for artifacts that keep turning up in your drafts, such as warped lettering or a stray object.