Qwen Image 3 Edit API: Prompt-Guided Image Editing

Key Takeaways

• Qwen Image 3 ships three editing endpoints on Siray: alibaba/qwen-image-3-edit at $0.03 per image, plus Spicy and Pro Spicy variants.
• The edit request body is the text-to-image body plus one field. images accepts 1 to 3 inputs, so compositing is a single call.
• There is no strength or denoise parameter. Edit intensity is controlled entirely by how the prompt is worded.
• size and aspect_ratio are both required, and prompt expansion is on by default.

What the Edit Endpoints Are

Alibaba's Qwen Image 3 family, covered in full here, runs on Siray as six endpoints. Three of them edit an existing image rather than generating from scratch.

One naming detail catches people out: the editing variant is -edit, not -i2i. Other image models in the pool use i2i, so reaching for qwen-image-3-i2i out of habit calls an endpoint that does not exist. Spicy tier pricing is broken down here.

The Edit Body Is the Generation Body Plus One Field

Comparing the published specifications, qwen-image-3-edit and qwen-image-3-t2i differ by exactly one property: images. Same path, same enums, same optional parameters. All three edit endpoints are likewise identical to each other apart from the model string.

Required fields are model, prompt, images, size, and aspect_ratio. Submission and polling follow the standard asynchronous job pattern.

Up to Three Input Images

images is an array with a minimum of one entry and a maximum of three, each supplied as a URL or a data URL. That upper bound is the useful part: compositing two or three sources is a single request, not a chain of edits. Typical uses are placing a product into a scene or transferring a style from a reference frame.

There Is No Strength Slider

This is the difference that matters most for anyone arriving from a local diffusion workflow. The Qwen Image 3 edit endpoints expose no strength, no denoise, and no negative_prompt — nothing that governs how far the output may drift from the input.Edit intensity is therefore a writing problem. Scope has to live in the prompt itself:

• Light touch — "keep the composition and lighting unchanged; adjust only the shirt colour to navy"
• Local replacement — "replace the background with an overcast beach, keep the subject and pose identical"
• Heavy reinterpretation — "restage the scene as a night market, retaining only the subject's face and outfit"

Naming what must stay fixed does more work than naming what should change. A prompt that only describes the target tends to drift, because nothing else pins the rest.

Two Required Fields That Look Optional

Size accepts 1k or 2k and nothing else, and it is required. It does not match the input image's resolution automatically, so a 2K source passed with 1k comes back smaller than it went in.

Aspect_ratio is also required, with fifteen accepted values from 1:1 and 16:9 through to 9:21 and 21:9. It is required even though the endpoint already holds the source image, which does not imply it. A ratio that disagrees with the input reframes the result.

Prompt Expansion Defaults to On

Prompt_expansion_enable defaults to true, so the prompt is automatically rewritten unless the flag is set to false. For open-ended generation that usually helps. For precise editing it is a hazard: an expanded prompt can introduce detail that was never asked for, which is exactly what a careful edit avoids. Set it to false when the instruction is already specific. Alongside it, n controls how many outputs come back, and seed makes a run reproducible.

FAQ

Can Qwen Image 3 Edit combine multiple images?

Yes. The images array accepts one to three inputs per request.

How do I control how much the image changes?

Through prompt wording. There is no strength or denoise parameter, so state explicitly what should stay unchanged.

What resolutions does it output?

size accepts 1k or 2k. Both it and aspect_ratio are always required.

Summary

Qwen Image 3 Edit is a compact API with three defaults that reward attention: required sizing, required aspect ratio, and prompt expansion switched on. For the wider field, see the 2026 image generator roundup. Create your free Siray account and start editing with alibaba/qwen-image-3-edit today.

Try Qwen Image 3 Edit on Siray