Create image edit

Image-to-image with x-ai/grok-imagine-image-quality. The body is multipart/form-data: reference images as files, everything else as ordinary fields.

This model supports neither streaming nor masks.

Authorization

header

AuthorizationstringRequired

Bearer <your Amux key> Your Amux API key.

Request

multipart/form-data

modelstringRequired

Amux model ID. Append :provider to pin a specific provider.

promptstringRequired

How the reference images should be changed.

imagearray<string>Required

Reference images, required. Both image and image[] are accepted — official SDKs

usually send the former for a single image and the latter for several.

At most 3, each up to 25 MiB, and at most 48 MB once encoded.

Exceeding either limit returns 400; images are never silently dropped.

nintegerDefault 1

How many images to produce (1–10, default 1). Billed per image; the hold scales with it too.

aspect_ratio"1:1" | "4:3" | "3:4" | "3:2" | "2:3" | "16:9" | "9:16" | "2:1" | "1:2" | "21:9" | "5:2" | "19.5:9" | "9:19.5" | "20:9" | "9:20" | "auto"Default "1:1"

Output aspect ratio, 1:1 by default. Identical to what the upstream model accepts — sixteen values:

1:1 4:3 3:4 3:2 2:3 16:9 9:16 2:1 1:2 21:9 5:2 19.5:9 9:19.5 20:9 9:20 auto

The ratio changes the shape of the output only and does not change the pricing tier — all sixteen were measured at the same price.

A ratio outside this set is dropped and explained in amux.notes.

image_size"1K" | "2K"Default "1K"

Resolution tier. 1K (default) or 2K. 4K is not supported.

It also determines the output format: 1K produces JPEG, 2K produces PNG.

Treat the extension of the returned URL as authoritative.

When omitted, 1K is sent explicitly — the tier is always written into the upstream

request rather than left to the upstream default.

moderation"auto" | "low" | "high"Default "auto"

Moderation strength. auto (default), low or high.

userstring

End-user identifier for abuse tracking. Omit the field entirely if you do not need it.

Response

200response

Edit succeeded. Same shape as the generation endpoint.

400response

Invalid request: too many or too large reference images, URL-form reference images, or a capability this model does not support.

402response

Available balance cannot cover the upper-bound hold.

503response

No available provider, or the in-flight queue is full.

Reference images#

The field name may be image or image[] (official SDKs send the former for a single image and the latter for several). Reference images are required.

Three limits; exceeding any of them returns 400 and never silently drops an image:

1. At most 3 — a hard upstream limit for this model;
2. 25 MiB per image;
3. 48 MB total once encoded.

Reference images are billed per image, independently of n: producing two images from one reference image still counts as one reference image.

⚠️ Masks are not supported (mask) — the upstream edit endpoint has no such concept. The field is dropped and explained in amux.notes.

⚠️ Reference images must be uploaded as files; URLs are not supported.

Dimensions and format#

Dimensions are expressed with two parameters: aspect_ratio and image_size. This model does not accept size (an exact pixel string); it is dropped and explained in amux.notes.

Other ratios scale within the same tier; the full list is in the next section. 4K is not supported.

The ratio changes the shape of the output only and does not change the pricing tier; the resolution tier does.

⚠️ The output format follows image_size and cannot be chosen — neither output_format nor output_compression has any effect on this model. Treat the extension of the returned URL as authoritative.

Aspect ratio#

Output aspect ratio, 1:1 by default. Identical to what the upstream model accepts — sixteen values:

The ratio changes the shape of the output only and does not change the pricing tier — all sixteen were measured at the same price.

A ratio outside this set is dropped and explained in amux.notes; we will not guess one for you.

quality is not supported#

The upstream model accepts this parameter, but on this model it changes neither the output resolution nor the pricing tier. It is therefore marked unsupported here: it is dropped and explained in amux.notes.

The pricing tier for this model depends on image_size alone.

Output files#

Both data[].url and amux.images[].url point at our stored copy, not at an upstream temporary URL.

Stored files are cleaned up — download anything you need to keep, or the URL will return 404.

Unsupported parameters#

Passing them is not an error: they are dropped and explained in amux.notes. If the result is not what you expected, check notes first.

⚠️ style is the exception: dropping it produces no amux.notes entry. Every other parameter above is marked unsupported in this model's parameter matrix, so there is something to report; style is not a parameter Amux models at all, so it is discarded while the request is parsed. In other words, sending style gives you no feedback whatsoever — write your code as if it did not exist.

Editing works asynchronously too#

Send the same parameters to /v1/tasks (also multipart) to run the edit in the background.

Errors#

The error body matches OpenAI's shape. See Errors and retries for the type values and retry semantics.

curl https://gateway.amux.ai/v1/images/edits \
  -H "Authorization: Bearer $AMUX_API_KEY" \
  -F "model=x-ai/grok-imagine-image-quality" \
  -F "prompt=<string>" \
  -F "image=@/path/to/image.png" \
  -F "n=1" \
  -F "aspect_ratio=1:1" \
  -F "image_size=1K" \
  -F "moderation=auto" \
  -F "user=<string>"

{
  "created": 1786000000,
  "data": [
    {
      "url": "<string>"
    }
  ],
  "usage": {
    "output_images": 0,
    "input_images": 0
  },
  "amux": {
    "task_id": "<string>",
    "generation_id": "<string>",
    "provider": "<string>",
    "cost_nano": "<string>",
    "images": [
      {
        "index": 0,
        "url": "<string>"
      }
    ],
    "notes": [
      "<string>"
    ]
  }
}