Qwen - Alibaba AI Image Generator and Photo Editor

Qwen Multi-Mode Image Generator

Switch between prompt-driven generation, reference-guided variation, and precise image editing workflows powered by Qwen.

Each render costs 30 credits.Available balance: 0 credits

Model Type

Choose the Qwen workflow that fits your creative task.

Generate high-fidelity visuals directly from descriptive prompts.

Input

Configure prompts, guidance, and generation controls before submitting.

Prompt

Keep prompts under 5,000 characters. Use clear scene descriptions for best results.

Image size

Square HD works well for most prompts. Use portrait or landscape for specific framing.

Output format

Inference steps

Higher steps improve detail but take longer.30

Guidance scale

Balances prompt adherence versus creative freedom.2.5

Seed

Leave blank for random results. Set a number to reproduce outputs.

Acceleration

Negative prompt

Separate undesirable attributes with commas (e.g. blurry, low contrast, watermarks).

Safety checker

Disable only if your workflow requires unrestricted outputs.

Output

Monitor progress and download the generated image.

Your generated image will appear here.

Progress0%

About Qwen Image

Qwen Image is Alibaba's image model, and this page exposes three routes into it: text to image, image to image with a reference upload, and image edit. It is the most manual page on the site, with controls for steps, guidance, seed, strength, a negative prompt and acceleration, so a result can be tuned and then repeated.

Where most image pages hide the sampler, this one puts it on the panel. Inference steps run from 2 to 250 in text and image-to-image, and from 2 to 49 in edit mode. Guidance scale runs 0 to 20, and a fixed seed reproduces a result you liked instead of hoping for it again. The strength slider in image-to-image decides how far the output may travel from the reference, with 0 holding the original and 1 letting the prompt take over. A negative prompt accepts a comma-separated list of what should not appear.

Each mode asks for something different. Text to image needs only a prompt and one of six fixed sizes: 512 square, 1024 square HD, or a 3:4, 9:16, 4:3 or 16:9 frame. Image to image needs a reference as well as an instruction, and the reference can be uploaded or pasted as a hosted URL. Image edit is the corrective mode: it adds a size selector, can wait for the edit before returning when sync mode is on, and returns one, two, three or four images from a single request. Uploads accept PNG, JPG or WEBP files capped at 10 MB, one per run.

The trade is control for convenience. Nano Banana and GPT-4o finish faster because they expose fewer decisions, and Flux Kontext is better when the frame must barely move. This page earns its keep when the same prompt has to come back the same way next month, when a studio needs a seed written down, or when a negative prompt is doing real work. A request is charged once, even when an edit returns several images, and a failed job refunds that charge.

Start from what the modes prefill. Text to image opens on a square HD frame with the negative prompt empty, while both upload modes arrive with a short list already sitting in that field, so a first run has something to push against before you type anything. Edit mode also opens on a landscape frame rather than a square, which points at the work it is meant for: a correction lifted from an existing picture rather than a scene built from nothing. Change those defaults on purpose, not because a run came back wrong.

Work the controls in a fixed order or you will not know what changed. Set the size first, because it decides the composition before the sampler does anything. Then set guidance and leave it there while you adjust steps: steps add detail, guidance sets how tightly the model follows your wording, and moving both at once tells you nothing about either. Pin a seed once a picture is nearly right, not before, and only then experiment with strength in image to image. The negative field is the last place to look when one artifact keeps returning.

How Qwen Image works

1
Choose the workflow
Text to image starts from a prompt. Image to image needs a reference plus a strength value. Image edit handles corrections and can return more than one picture per request.
2
Set the sampler before anything else
Steps, guidance, seed and acceleration, plus a negative prompt and the safety checker. Higher step counts add detail and time, and guidance decides how literally the prompt is followed.
3
Pick a size and generate
Six preset sizes, PNG or JPEG output. Edit mode adds the image count and a sync switch. The exact cost sits above the generate button, and the finished image can be downloaded.

What Qwen Image is best for

Repeatable brand images
Write the seed, steps and guidance into a brief and the same run can be produced again next month, which is what a recurring asset series needs.
Reference-led variation
Image to image keeps a base photograph or illustration as the starting point, and the strength slider decides how much of it survives the new prompt.
Multi-image edit passes
Edit mode can return up to four images from one request, so comparing clean-up options does not mean submitting four separate jobs.
Clean-up with a negative prompt
Name the artifacts you keep getting, such as blur or extra fingers, in the negative field, then keep the run that stops producing them.

Qwen Image at a glance

Model: Alibaba Qwen Image, run through KIE
Credit cost: The exact cost is shown above the generate button, charged once per request in all three modes
Modes: Text to image, image to image, image edit
Image sizes: Square 512, square HD 1024, portrait 3:4, portrait 9:16, landscape 4:3, landscape 16:9
Sampler controls: 2 to 250 steps (2 to 49 in edit), guidance 0 to 20, strength 0 to 1, seed and acceleration
Output: PNG by default, JPEG optional; up to 4 images per edit request

Qwen Image FAQ

How much does a Qwen Image run cost?: The exact cost is shown above the generate button. A request is charged once in any of the three modes, even when image edit returns several images. The free plan opens with a 300-credit signup bonus and 50 claimable credits per daily login, both tied to a 30-day expiry; Starter is $4.90 a month for 300 credits.
What is the difference between image to image and image edit?: Image to image treats the upload as a starting point and uses the strength slider to decide how much of it survives. Image edit is aimed at corrections: it has its own step ceiling, a size selector, a sync switch and a choice of one to four output images.
How do I get the same result twice?: Write down the seed, the step count, the guidance scale and the negative prompt. If those match and the reference is the same, the sampler has what it needs to reproduce the run. Leaving the seed blank gives a random result each time.
Why use this page instead of Nano Banana or Flux Kontext?: Nano Banana and GPT-4o hide the sampler and finish faster. Flux Kontext is built to preserve a frame through one change. Qwen Image keeps steps, guidance, seed, strength and a negative prompt in view, which is what makes a result repeatable and a series consistent.
What happens when a generation fails?: A failed job returns its credits without a support ticket. The progress panel shows the task state while it runs, and the result can be downloaded once it completes.
Can image edit fix a photo without changing its composition?: It can come close. Keep the size on the same frame shape as the source, describe the fix as the only thing you want, and put what you do not want in the negative field. One edit submission may come back with as many as four images, so you can compare clean-up options before choosing one. If the composition still drifts, the correction is closer to what image to image does at a low strength; edit is for repairs inside the frame.

About Qwen Image

How Qwen Image works

Choose the workflow

Text to image starts from a prompt. Image to image needs a reference plus a strength value. Image edit handles corrections and can return more than one picture per request.

Set the sampler before anything else

Steps, guidance, seed and acceleration, plus a negative prompt and the safety checker. Higher step counts add detail and time, and guidance decides how literally the prompt is followed.

Pick a size and generate

Six preset sizes, PNG or JPEG output. Edit mode adds the image count and a sync switch. The exact cost sits above the generate button, and the finished image can be downloaded.

What Qwen Image is best for

Repeatable brand images

Write the seed, steps and guidance into a brief and the same run can be produced again next month, which is what a recurring asset series needs.

Reference-led variation

Image to image keeps a base photograph or illustration as the starting point, and the strength slider decides how much of it survives the new prompt.

Multi-image edit passes

Edit mode can return up to four images from one request, so comparing clean-up options does not mean submitting four separate jobs.

Clean-up with a negative prompt

Name the artifacts you keep getting, such as blur or extra fingers, in the negative field, then keep the run that stops producing them.

Qwen Image at a glance

Model

Alibaba Qwen Image, run through KIE

Credit cost

The exact cost is shown above the generate button, charged once per request in all three modes

Modes

Text to image, image to image, image edit

Image sizes

Square 512, square HD 1024, portrait 3:4, portrait 9:16, landscape 4:3, landscape 16:9

Sampler controls

2 to 250 steps (2 to 49 in edit), guidance 0 to 20, strength 0 to 1, seed and acceleration

Output

PNG by default, JPEG optional; up to 4 images per edit request

Qwen Image FAQ

How much does a Qwen Image run cost?

The exact cost is shown above the generate button. A request is charged once in any of the three modes, even when image edit returns several images. The free plan opens with a 300-credit signup bonus and 50 claimable credits per daily login, both tied to a 30-day expiry; Starter is $4.90 a month for 300 credits.

What is the difference between image to image and image edit?

Image to image treats the upload as a starting point and uses the strength slider to decide how much of it survives. Image edit is aimed at corrections: it has its own step ceiling, a size selector, a sync switch and a choice of one to four output images.

How do I get the same result twice?

Write down the seed, the step count, the guidance scale and the negative prompt. If those match and the reference is the same, the sampler has what it needs to reproduce the run. Leaving the seed blank gives a random result each time.

Why use this page instead of Nano Banana or Flux Kontext?

Nano Banana and GPT-4o hide the sampler and finish faster. Flux Kontext is built to preserve a frame through one change. Qwen Image keeps steps, guidance, seed, strength and a negative prompt in view, which is what makes a result repeatable and a series consistent.

What happens when a generation fails?

A failed job returns its credits without a support ticket. The progress panel shows the task state while it runs, and the result can be downloaded once it completes.

Can image edit fix a photo without changing its composition?

It can come close. Keep the size on the same frame shape as the source, describe the fix as the only thing you want, and put what you do not want in the negative field. One edit submission may come back with as many as four images, so you can compare clean-up options before choosing one. If the composition still drifts, the correction is closer to what image to image does at a low strength; edit is for repairs inside the frame.