Midjourney v8.2 / Niji 7 image generation

Use the asynchronous image API at https://apimaster.ai/v1 with either midjourney-v8.2 or midjourney-niji-7. The model ID selects the version; do not add version override fields.

Create a task

POST https://apimaster.ai/v1/midjourney/generations

curl -s "https://apimaster.ai/v1/midjourney/generations" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "midjourney-v8.2",
    "prompt": "a quiet Japanese garden at dawn, cinematic composition",
    "size": "16:9",
    "speed": "relax"
  }'

For Niji 7, set model to midjourney-niji-7:

{"model":"midjourney-niji-7","prompt":"anime character under cherry blossoms","size":"9:16","speed":"fast"}

Field	Type	Required	Description
model	string	Yes	midjourney-v8.2 or midjourney-niji-7
prompt	string	Yes	Image description, including supported Midjourney parameters
size	string	No	Aspect ratio such as 1:1 or 16:9; defaults to 1:1
speed	string	No	relax, fast, or turbo; defaults to relax
image_urls	string[]	No	1–20 reference image URLs; supported image Data URLs are also accepted
iw	number	No	Reference image weight 0–3; requires image_urls
quality	string	No	0.25, 0.5, 1, or 2; not supported by Niji 7
stylize / chaos / weird	integer	No	Ranges: 0–1000, 0–100, and 0–3000, respectively
seed	integer	No	Random seed 0–4294967295
negative_prompt	string	No	Content to exclude
sref / sw	string / integer	No	Public style reference URL and weight 0–1000; sw requires sref
dref / dw	string or string[] / number	No	Reference URLs (up to 20) and weight 0–100; dw requires dref
raw / tile / hd	boolean	No	Raw style, tiling, and HD; tile and hd are v8.2 only
extra	string	No	Supported parameter string such as --ar 16:9 --s 750; structured fields take precedence
metadata	object	No	Custom business metadata
repeat	integer	No	2–40 to create multiple tasks

Niji 7 does not support quality, tile=true, or hd=true. Both models reject draft=true, cref, cw, stop, and request fields version and niji. style=raw is normalized to raw=true. Do not use --v or --niji in prompt or extra to override the model version.

The response contains a batch ID and task IDs:

{
  "id": "imagine_batch_xxxxxxxx",
  "request_id": "req_xxxxxxxx",
  "model": "midjourney-v8.2",
  "status": "queued",
  "data": [{"id": "imagine_xxxxxxxx", "task_id": "imagine_xxxxxxxx", "status": "submitted"}]
}

Check task status

curl -s "https://apimaster.ai/v1/tasks/imagine_batch_xxxxxxxx" \
  -H "Authorization: Bearer YOUR_API_KEY"

The completed response includes image URLs in data; tasks contains each child task:

{
  "id": "imagine_batch_xxxxxxxx",
  "status": "completed",
  "progress": 100,
  "data": [{"url": "https://example.com/generated.png"}],
  "tasks": [{"task_id": "imagine_xxxxxxxx", "status": "completed", "progress": 100, "data": [{"url": "https://example.com/generated.png"}]}]
}

Poll while the status is queued or processing. On failure, read tasks[].error and do not resubmit the same request. Download returned URLs promptly because they may expire.

If submission returns 502 with an X-Task-ID header, or a query returns pending_timeout, save the ID and keep checking its status instead of automatically submitting again. A batch can contain both successful and failed tasks; inspect tasks[] individually. Polling once every 15 seconds is recommended.

Errors and pricing

model and a non-empty prompt are required. Unsupported versions, invalid ratios, inaccessible references, and unsupported parameters return 400. 401 means an invalid key, 402 means insufficient balance, and 429 means rate limiting. See the Midjourney v8.2 card and Niji 7 card for current pricing and routes.