Qwen Image Edit 2511: what Sume offers for image edits

Sume lists no model named Qwen Image Edit 2511. It does take edits on qwen/qwen-image with up to 10 reference images. The request, the limits, and the gaps.

4 min readSume
All posts

Sume does not list a model named Qwen Image Edit 2511. What it does list is qwen/qwen-image, which accepts image edits: send your picture in input_references, up to 10, with an instruction as the prompt. Sume's docs do not say which Qwen release that id serves, so treat it as a different route to edits, not a stand-in for a named checkpoint.

Read from the Image API docs and the catalog code on 2026-09-29.

How do I edit an image with qwen/qwen-image?

Reference URLs must be public HTTPS; localhost, private-network and non-HTTPS URLs are rejected before submission. The docs suggest aspect_ratio: "auto" on edits to match the reference, but this model lists 13 named ratios and auto is not among them, so pick the ratio of your source.

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen-image",
    "prompt": "change the shirt to dark green, keep everything else",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/portrait.jpg"}}
    ],
    "aspect_ratio": "3:4"
  }'

What can and can't the edit route do?

The edit surface is the same reference-image mechanism every edit-capable Sume model shares.

Edit capabilities of qwen/qwen-image on Sume, read 2026-09-29.
NeedOn `qwen/qwen-image`
Reference imagesUp to 10, public HTTPS
Mask upload (mask_url)Not listed; 400 unsupported_parameter
Transparent background (background)Not listed; 400 unsupported_parameter
Edit with qwen/qwen-image-maxNot possible; text-to-image only
Images per request1 to 4

Is there a masked edit on Sume?

Yes, but not on Qwen. mask_url is documented for ChatGPT Image 2.5 edits only; see GPT Image 2.5 mask edit. On Qwen, describe the region in the prompt instead.

How do I know if an edit-named Qwen model is added?

Read GET /v1/images/models and look at architecture.input_modalities: a model that takes images lists image there, and input_references has a maximum above 0. The docs do not promise a new id, and neither does this post.

What does an edit return and cost?

A 200 returns Sume-hosted URLs with a media_type, and usage.cost in USD. Edits are billed like any generation: in full when completed, not at all when failed. The price is the model's pricing line, and references do not add a separate line in the docs.

A slow edit degrades to 202 with a job envelope after 30 seconds. Check the status code, then poll GET /v1/jobs/{id}/status and read GET /v1/jobs/{id}/result.

How do I chain several edits?

Each result is a Sume-hosted HTTPS URL, which is a valid reference for the next call. Take data[0].url from one response and pass it as the input_references entry of the next, one change per call. Keep the prompt to the change and name what must stay, as in the example above. Copy the final file to your own storage.

Which reference should I use when I have several photos?

Start with one reference and add more only when the first result misses something the extra photo can show, such as a product from a second angle. Every reference beyond your need is another URL that must be reachable and public when Sume fetches it, and an unreachable one fails the request.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume