Midjourney Diffusion

midjourney_diffusion

Midjourney Diffusion uses a diffusion model to generate high-quality images with artistic style control.

MidJourney interface notes

MidJourney endpoints are split into direct generation and secondary editing. Direct generation only requires a prompt and other required fields to create an image. Secondary editing requires an upstream task ID and an image number from a previous task. diffusion is a direct generation endpoint.

Prompt parameters

In the text field, official prompt parameters must be appended to the same string instead of being passed as separate JSON fields. Multiple parameters can be used in one prompt.

Prompt tips

Use short, clear, specific descriptions and prioritize what should appear in the image.
Use concrete numbers when possible, such as “three cats” instead of “cats”.
Use --no to exclude content instead of only describing what you do not want in natural language.
Add important details around subject, medium, environment, lighting, color, mood, and composition.

Common control parameters

Parameter	Meaning	Values and notes
`--ar` / `--aspect`	Aspect ratio	Sets the output image aspect ratio, for example `--ar 16:9`. Ratios greater than 2:1 are experimental and may produce unpredictable results.
`::` / `::<weight>`	Multi-prompt and weights	Splits prompt fragments with double colons and optionally adds weights, for example `space::2 ship`. The total weight must be positive.
`{option1,option2}`	Permutation prompts	Uses braces to generate multiple prompt variants, for example `a {red,green,yellow} bird`. Each generated prompt is billed separately.
`--v` / `--version`	Model version	Use `--v 8.1`. Examples and requests in these docs use `--v 8.1` consistently.
`--iw`	Image weight	Controls how strongly image prompts affect the result relative to text prompts. The default is `1`.
`--draft`	Draft mode	Enables faster draft generation. Availability depends on the current model capability.
`--no`	Negative prompt	Specifies content that should not appear, for example `--no plants`.
`--quality` / `--q`	Quality and render time	Controls render effort and GPU usage. Common values include `0.25`, `0.5`, and `1`; some versions support `2`.
`--repeat` / `--r`	Repeat generation	Creates multiple tasks from the same prompt. The official range is `1-40`.
`--seed`	Seed	Specifies an integer from `0-4294967295`. The same prompt and seed usually produce similar results.
`--stop`	Early stop	Stops the task at a percentage from `10-100`. Lower values usually create blurrier results with fewer details.
`--relax` / `--fast` / `--turbo`	Generation speed	Overrides the current setting and runs the current task in slow, fast, or turbo mode.
`--stylize` / `--s`	Stylization	Controls the strength of the default aesthetic algorithm. The default is usually `100`; common range is `0-1000`.
`--chaos` / `--c`	Diversity	Controls variation and unpredictability, from `0-100`. Higher values create more divergent results.
`--weird` / `--w`	Weirdness	Explores more unusual aesthetics. Common range is `0-3000`.
`--style`	Style mode	The main model commonly uses `--style raw` to reduce default beautification. Niji 5 can use `cute`, `expressive`, `original`, or `scenic`.
`--exp`	Creative style	Range `0-100`. Higher values usually make images more detailed, lively, and creative.
`--tile`	Tile	Generates seamless repeatable patterns.

Image and style reference parameters

Parameter	Meaning	Values and notes
`--sref <image URL\|style code\|random>`	Style reference	References the visual style of an image or style code. Multiple sources can be combined, for example `--sref URL1::2 URL2::1`.
`--sw`	Style reference weight	Controls style reference strength, from `0-1000`. The default is usually `100`. `--sw` is not compatible with Moodboards.
`--sv`	Style reference version	Selects the style reference algorithm version. Availability depends on the current model capability.
`--cref <image URL>`	Character reference	Keeps character features from the reference image. Availability depends on the current model capability.
`--cw`	Character reference weight	Controls how much character detail is preserved, from `0-100`. `--cw 0` focuses more on the face; `--cw 100` tries to preserve face, hair, clothing, and more details.
`--oref <image URL>`	Omni reference	Keeps consistency for the person, character, object, or vehicle in the reference image. Only one omni-reference image can be used at a time.
`--ow`	Omni reference weight	Controls omni-reference strength, from `1-1000`. Higher values increase consistency, but very high values may make results unstable.

Examples

A cinematic sunset landscape, ultra detailed --ar 16:9 --v 8.1 --s 250 --seed 1234
A clean product photo of a glass teapot --oref https://example.com/object.png --ow 300 --sref https://example.com/style.png --sw 120 --no text, watermark

Examples in this documentation use --v 8.1. If a parameter does not take effect, rely on the current model capability and platform response.

Endpoints

POST /v1/queue/midjourney_diffusion # Create task

GET /v1/queue/midjourney_diffusion/requests/{request_id}/status # Query status

GET /v1/queue/midjourney_diffusion/requests/{request_id}/response # Get result

PUT /v1/queue/midjourney_diffusion/requests/{request_id}/cancel # Cancel task

Request parameters

curl --location "https://cloud.seaart.ai/model/v1/queue/midjourney_diffusion" \
  --header "Authorization: Bearer YOUR_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "text": "A red apple on a white table, clean studio lighting, simple background."
  }'

text string required

Text description for image generation. Youchuan/Midjourney prompt parameters can be appended to the same string, such as --ar 16:9, --v 8.1, or --seed 1234.

Length range: 1 - 8192 characters

Response parameters

{
  "request_id": "mmsu_01hxyz...",
  "response_url": "https://cloud.seaart.ai/model/v1/queue/midjourney_diffusion/requests/mmsu_01hxyz.../response",
  "status_url": "https://cloud.seaart.ai/model/v1/queue/midjourney_diffusion/requests/mmsu_01hxyz.../status",
  "cancel_url": "https://cloud.seaart.ai/model/v1/queue/midjourney_diffusion/requests/mmsu_01hxyz.../cancel",
  "queue_position": 0
}

Field	Type	Description
`request_id`	string	Task ID used for all follow-up operations.
`response_url`	string	URL for retrieving the final result.
`status_url`	string	URL for querying task status.
`cancel_url`	string	URL for canceling the task.
`queue_position`	int	Queue position.

Follow-up operations

Query status

Get result

Cancel task

Error handling