Create image edit
Last updated November 23, 2026
Edit and composite images with openai/gpt-image-2.6-flare using reference images, masks, or visual markup overlays.
Reference-based image editing with openai/gpt-image-2.6-flare. Send requests as multipart/form-data, passing reference images (image / image[]) and optional mask or 2.6 markup overlays as binary file parts, and remaining parameters as form fields.
https://gateway.amux.ai/v1/images/editsAuthorization
headerAuthorizationstringRequiredBearer <your Amux AI key> Amux AI API key (Bearer token).
Request
multipart/form-datamodelstringRequiredModel identifier (optionally suffixed with :amux-special or :adobe).
promptstringRequiredText instructions describing the desired edit (up to 32,000 characters).
imagearray<string>RequiredReference image files (up to 16 files, maximum 25 MB per file).
Both image and image[] field names are accepted (OpenAI SDKs use the former for single-image edits and the latter for multiple images).
maskstringOptional inpainting mask file whose transparent pixels (alpha = 0) mark the area to edit. Mutually exclusive with markup. Must satisfy:
1. PNG format with an alpha channel;
2. Exact same pixel dimensions as the source image, and file size under 4 MB;
3. Applies to the first reference image when multiple reference images are supplied.
Provider support: Unsupported on adobe; without :adobe pinned, routing automatically skips adobe in favor of amux-special, while pinning :adobe returns 400.
markupstringNew in GPT-Image 2.6: visual annotation/sketch overlay image (strokes, circles, arrows, or labels) indicating where and how to modify the source image. Mutually exclusive with mask; effective only alongside reference images.
Provider support: Unsupported on adobe (pinning :adobe returns 400); supported on amux-special.
nintegerDefault 1Number of images to generate (1–10, default 1); must be 1 when streaming.
sizestringDefault "1024x1024"Output image pixel dimensions (defaults to 1024x1024). adobe strictly enforces ≥ 655,360 total pixels and native 4K; amux-special scales non-4K sizes to ~1.57 MP and uses super-resolution for 4K.
aspect_ratiostringAspect ratio (1:1, 4:3, 3:4, 3:2, 2:3, 16:9, or 9:16) as an alternative to size. Prefer size; if both are supplied, size takes precedence and any conversion or omission is recorded in amux.notes.
image_sizestringResolution tier (1K, 2K, or 4K), used in combination with aspect_ratio under the same conversion rules.
quality"low" | "medium" | "high" | "xhigh" | "max" | "auto"Default "auto"Render quality tier (low, medium, high, xhigh, max, or auto).
complexity"low" | "medium" | "high"Default "high"New in GPT-Image 2.6: visual and compositional complexity (low, medium, or high, default high). Supported on both amux-special and adobe.
background"transparent" | "opaque" | "auto"Default "auto"Background mode (transparent, opaque, or auto). Transparency requires png or webp.
output_format"png" | "jpeg" | "webp"Default "png"Output image format (png, jpeg, or webp, default png). svg and gif are not supported on edits.
output_compressionstringCompression level (0–100, default 100); applies only to jpeg and webp.
userstringUnique end-user identifier for abuse detection and monitoring.
stream"true" | "false"Enable SSE streaming (pass 'true' or 'false' as text in multipart forms). Provider support: Unsupported on both amux-special and adobe until official openai / azure providers launch.
partial_imagesstringMaximum number of low-resolution preview frames during streaming (0–3, default 0). Provider support: Depends on streaming; unsupported on both amux-special and adobe.
Response
200responseImage edit succeeded. Returns the same payload structure as the generation endpoint.
400responseInvalid request parameters (such as combining mask and markup, requesting svg/gif on edit, pinning a provider that does not support mask/markup/stream, or n > 1 while streaming).
402responseInsufficient available workspace balance to cover the estimated reservation.
503responseConcurrent task limit reached or service temporarily unavailable. Retry later.
Providers & availability
| Provider | Status | Billing | Editing capabilities | Unsupported parameters |
|---|---|---|---|---|
amux-special | Live | Value per-image billing | Multi-image edits, mask, and markup visual annotations | stream, partial_images |
adobe | Live | Value per-image billing (slightly above amux-special) | Standard reference editing with native 4K output | mask, markup, stream, partial_images |
openai | Coming soon | Official token-based billing | Full official editing spec | — (activates upon launch) |
azure | Coming soon | Official Azure token-based billing | Full official editing spec | — (activates upon launch) |
Without an explicit :provider suffix, requests carrying mask or markup automatically skip adobe and route to amux-special. Pinning :adobe with mask or markup returns 400.
Reference images, mask, and markup
- Both
imageandimage[]field names are accepted, up to 16 files per request and 25 MB per file. maskandmarkupare mutually exclusive: passing both in one request returns400.maskmust be a PNG with an alpha channel (transparent pixels mark the region to edit), matching the first source image's pixel dimensions and ≤ 4 MB; unsupported onadobe.markup(new in 2.6) is a visual annotation/sketch overlay showing strokes, circles, arrows, or notes directly over the target image; unsupported onadobe, supported onamux-special.- The edit endpoint supports
png,jpeg, andwebpoutput formats only (svgandgifare generation-only).
input_fidelity
GPT-Image 2.6 does not accept input_fidelity. To preserve compatibility for callers migrating from gpt-image-1.5, the gateway strips input_fidelity before forwarding and records a note in amux.notes.
Asynchronous editing
Submit the same multipart/form-data fields to /v1/tasks to run edits asynchronously; see Create task.
Streaming
Neither live provider (amux-special or adobe) currently supports streaming edits (stream=true returns 400). Streaming will activate once official openai / azure providers launch.
Errors
Error responses follow OpenAI's standard schema. See Errors and retries for type codes and retry guidance.
curl https://gateway.amux.ai/v1/images/edits \
-H "Authorization: Bearer $AMUX_API_KEY" \
-F "model=openai/gpt-image-2.6-flare" \
-F "prompt=<string>" \
-F "image=@/path/to/image.png" \
-F "mask=@/path/to/mask.png" \
-F "markup=@/path/to/markup.png" \
-F "n=1" \
-F "size=1024x1024" \
-F "aspect_ratio=<string>" \
-F "image_size=<string>" \
-F "quality=auto" \
-F "complexity=high" \
-F "background=auto" \
-F "output_format=png" \
-F "output_compression=<string>" \
-F "user=<string>" \
-F "stream=true" \
-F "partial_images=<string>"{
"created": 1786000000,
"data": [
{
"b64_json": "<string>",
"revised_prompt": "<string>"
}
],
"usage": {
"input_tokens": 0,
"output_tokens": 0,
"total_tokens": 0
},
"amux": {
"task_id": "<string>",
"generation_id": "<string>",
"provider": "<string>",
"cost_nano": "<string>",
"images": [
{
"index": 0,
"url": "<string>"
}
],
"notes": [
"<string>"
]
}
}