Submit Task (Vega Format)
Vega image generation runs as an async task: submit → poll → download the image. Submitting returns a task ID immediately without holding the connection open — ideal for batch submission, long-running generation, and centralized management in a task queue. Available models and current prices are subject to the Model Plaza.
All three models share the same endpoint and request fields. Text-to-image and image-to-image use the same endpoint: omit input for text-to-image; include input for image-to-image.
An async task pre-charges 0 at submission and settles against the usage actually returned by the upstream when the task completes (difference refunded or charged); failed tasks are not billed. Check the console's "Usage Logs" at any time.
Generation usually takes 20–100 seconds (Lite about 19 s and Pro about 98 s in our tests). Allow at least 5 minutes of polling, with a 30-second timeout per poll request — see Query & Download.
Endpoint
POST /v1/tasks/vega
Authentication
Pass the Bearer token in the request header (see Base URL & Authentication):
Request Body
This endpoint only processes the fields above; other image parameters (response_format, quality, user, etc.) are ignored. The image URL is given by the artifacts endpoint under Query & Download — response_format is not needed.
Text-to-Image
Pass only model + prompt (optionally size) for text-to-image.
Request Examples
Response Body
A successful submission returns the task object immediately, with status fixed at queued:
Response Example
Image-to-Image
Same endpoint with input carrying reference images — the model redraws the whole image based on them (change style, change scene, keep the composition and subject, etc.).
input Fields
Reference Image Limits
Request Examples
Notes
- The
promptmust include wording like "keep the original image / design" (保持原图 / 原设计), otherwise the model ignores the reference images and treats the task as plain text-to-image. - For local images or images you prefer not to expose, inline them as
data:image/png;base64,...— see the WARNING below. - Max 10MB per reference image; PNG / JPEG supported.
- Reference image resolution does not affect output size (determined by
size).
Reference images are downloaded by the upstream service, so they must be publicly reachable. In our tests an overseas image host (e.g. Wikimedia) failed to download and the task ended with 502001 上游模型服务异常或不可达 (upstream service error or unreachable). Use a publicly accessible URL, or inline the image as data:image/png;base64,... (recommended, most reliable).
Billing
Billing settles against the usage actually returned by the upstream: 0 pre-charged at submission, charged on completion at actual usage, and failed tasks are not billed. Consumption varies by model and reference image count — measured values (1024x1024, one reference image):
For wand-vega-image-pro image-to-image, the first 3 reference images are free and each image from the 4th on adds 10,000 tokens. Token prices and final amounts are subject to the Model Plaza; actual charges are visible in the console's "Usage Logs".
Errors
For more error codes see Error Codes & Troubleshooting.
The former synchronous endpoint POST /v1/images/generations no longer supports the Vega series — migrate to the async API on this page. The pre-upgrade synchronous integration is kept on the Legacy Guide for transition reference only.
After a successful submission, poll the task status and download the image — see Query & Download.
