Image Generation

Mesh API exposes a standalone OpenAI-compatible image generation endpoint at POST /v1/images/generations, supporting models from OpenAI (gpt-image-*) and Vertex AI (Imagen).

Images can also be generated through POST /v1/chat/completions by setting the image modality on a chat request. The standalone /v1/images/generations endpoint below is the OpenAI-compatible path.

Endpoint

POST /v1/images/generations

Parameters

Field	Type	Description
`prompt`	`string`	Text description of the image. Required.
`model`	`string`	Model ID (e.g., `openai/gpt-image-1`, `vertex/imagen-3`). Required.
`n`	`integer`	Number of images to generate (1–10). Default: `1`.
`size`	`string`	Dimensions or aspect ratio (e.g., `1024x1024`, `1792x1024`). Default: `auto`.
`quality`	`string`	Rendering quality: `low`, `medium`, `high` (GPT image) or `hd`, `standard` (DALL·E), or `auto`. Default: `auto`.
`output_format`	`string`	Encoding for the returned image: `png`, `jpeg`, or `webp`. OpenAI only.
`response_format`	`string`	`url` or `b64_json`.
`background`	`string`	`transparent`, `opaque`, or `auto`.
`moderation`	`string`	Moderation level: `low` or `auto`.
`output_compression`	`integer`	Compression level 0–100 (for `jpeg`/`webp`).
`partial_images`	`integer`	Number of partial images to stream, 0–3.
`stream`	`boolean`	Enable SSE keep-alive / native streaming. Default: `false`.

Streaming

Setting stream: true returns a text/event-stream response that prevents the connection from closing during long-running generations.

The stream always ends with data: [DONE].

Keep-alive chunk sequence

data: {"id":"img-...","object":"image.chunk","created":...,"model":"...","data":[],"status":"processing"}
: ping
: ping
data: {"id":"...","object":"image.chunk","created":...,"data":[{"url":"..."}]}
data: [DONE]

Response

1 {
2   "created": 1715356800,
3   "data": [
4     {
5       "url": "https://...",
6       "revised_prompt": "..."
7     }
8   ],
9   "background": "opaque",
10   "output_format": "webp",
11   "quality": "high",
12   "size": "1024x1024",
13   "usage": {
14     "prompt_tokens": 100,
15     "completion_tokens": 1000,
16     "total_tokens": 1100
17   }
18 }

background, output_format, quality, size, and usage are only present when the upstream provider returns them. Vertex AI (Imagen) does not return token usage. Vertex AI always returns b64_json regardless of response_format.

Examples

Standard generation

$ curl -X POST https://api.meshapi.ai/v1/images/generations \
>   -H "Authorization: Bearer YOUR_MESH_API_KEY" \
>   -H "Content-Type: application/json" \
>   -d '{
>     "model": "openai/gpt-image-1",
>     "prompt": "A watercolor of a fox in a snowy forest",
>     "n": 1,
>     "size": "1024x1024",
>     "quality": "high",
>     "output_format": "webp"
>   }'

Streaming

$ curl -X POST https://api.meshapi.ai/v1/images/generations \
>   -H "Authorization: Bearer YOUR_MESH_API_KEY" \
>   -H "Content-Type: application/json" \
>   -d '{
>     "model": "openai/gpt-image-1",
>     "prompt": "A watercolor of a fox in a snowy forest",
>     "stream": true
>   }'

Image editing

POST /v1/images/edits

Edit or transform an existing image. This is a multipart/form-data request — send the source image file plus the fields below. The operation determines the transform:

`operation`	Description
`edit`	Prompt-guided edit of the input image.
`remove_background`	Remove the image background.
`upscale`	Increase resolution.
`outpaint`	Extend the image beyond its original borders.
`inpaint`	Fill a masked region (supply a `mask`).
`mix`	Blend multiple reference images.
`reframe`	Recompose to a new aspect ratio.

Common form fields: image (the source file), prompt, model, operation, mask, and reference_images. If the selected model does not support the requested operation, the endpoint returns 501 Not Implemented.