Overview
FLUX 3 Image generates images from text and edits or combines reference images. Each task returns one image. Choose a resolution and output format, submit a task, then poll for the result or receive a webhook callback.Endpoint and authentication
Base URL:
Include your APIXO API key in every request:
Content-Type: application/json.
Copy-paste async quickstart
Submit a text-to-image task:data.taskId and query the status endpoint. A successful submission means the task was accepted; check data.state to determine the final outcome.
Poll for result
data.state is processing. Continue polling until it becomes success or failed.
Success response:
resultJson is a JSON string. Parse it before reading the image URL:
code: 200 can still describe a failed task. Use data.state, failCode, and failMsg to handle task failures.
Image-to-image request
Provide 1-10 reference images inimage_urls. Their order is preserved, so refer to them as image 1, image 2, and so on in the prompt.
width * height).
When aspect_ratio is omitted or blank in image-to-image mode, the output follows the first reference image’s aspect ratio.
Parameters
Result delivery mode. Supported values:
async, callback. Use async for status polling or callback for webhook delivery.Required when
request_type is callback. Provide a public HTTPS endpoint that accepts POST requests. See Webhooks.Status response fields
All fields below are insidedata.
Webhook callback mode
Setrequest_type to callback and provide callback_url:
Billing
New non-LLM requests use 1 credit =$0.001. Calculate the full amount, including billable quantity and add-ons, apply the applicable account discount, then round the request total upward to the next whole credit. Agreed fixed unit prices retain their precision and replace the standard price and discount; final rounding still applies. See credit conversion and rounding.
Both modes generate one image per task and use the same resolution-based prices. The number of reference images does not add a separate charge.
See API pricing for current prices.
Latency and polling
Generation time varies with resolution, prompt complexity, reference images, and queue load. A 4K task can take several minutes.
If a request is rate-limited, reduce concurrency and retry with backoff. Keep the original
taskId when querying an accepted task.
Errors and troubleshooting
- Check that
modeand a non-emptypromptare present. - For image-to-image, provide 1-10 valid reference URLs and respect the per-image dimension and pixel limits.
- Use one of the documented aspect ratios, resolution tiers, and output formats.
- Inspect
data.stateeven when the status endpoint returnscode: 200. - When a task fails, use
failCodeandfailMsgto decide whether to change the input before submitting another task.
MCP integration
The complete model schema declares result decoding inmcp.result and supported file bindings in mcp.attachments. These are client integration metadata, not generation input parameters. Attachment paths: image_urls. Preserve the user’s file order. Model parameter validation remains on the API.