> ## Documentation Index
> Fetch the complete documentation index at: https://docs.qingbo.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# FLUX 3 Image Klein

> Compact image generation and editing with up to 10 references, five resolution tiers and prompt-based bounding-box layouts.

Compact image generation and editing with up to 10 references, five resolution tiers and prompt-based bounding-box layouts.

## Quick start

```bash theme={"system"}
curl -X POST https://api.qingbo.ai/v1/tasks \
  -H "Authorization: Bearer $WAVE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: flux-3-image-klein-example-001" \
  -d '{
  "model": "flux-3-image-klein",
  "action": "generate",
  "prompt": "A blue ceramic cup on a wooden table, soft morning light",
  "resolution": "1k",
  "aspect_ratio": "1:1",
  "steps": 16,
  "output_format": "png"
}'
```

Submission returns a task\_id. Use [`GET /v1/tasks/{task_id}`](https://docs.qingbo.dev/en/api-reference/task/status) to poll the result. Reference assets must use publicly downloadable URLs.

## Parameters

| Parameter | Type or values | Description |
| - | - | - |
| model | `flux-3-image-klein` | Exact model ID. |
| `action` | generate / edit | generate creates an image without references; edit requires references. |
| `prompt` | string | Required description or editing instruction. Layout tags and bbox JSON may be embedded in the prompt; boxes use normalized 0–1000 \[y1,x1,y2,x2] coordinates. |
| `image_urls` | string\[] | 1–10 reference images for edit; references have no extra fee. |
| `resolution` | 768sq / 1k / 1.5k / 2k / 4k | Defaults to 1k; selects the billing tier. |
| `aspect_ratio` | string | Defaults to auto. Supports 1:1, 16:9, 9:16, 21:9, 2:1, 1:2, 3:2, 2:3, 7:5, 5:7, 4:3, 3:4, 5:4, 4:5 and 9:21. |
| `steps` | integer | 1–16, default 16; does not affect pricing. |
| `seed` | integer | Optional, 0–4294967295. |
| `output_format` | jpeg / jpg / png | Defaults to jpeg; jpg is an alias for jpeg. |
| `safety_tolerance` | integer | 0–4, default 2; 0 is the strictest. |
| `n` | 1 | One image per task. |

grounding, guidance, negative\_prompt, prompt\_upsampling, mask\_url, width, height and exact-pixel size are not supported. Use aspect\_ratio for shape and resolution for output tier.

## Reference-image editing

```json theme={"system"}
{
  "model": "flux-3-image-klein",
  "action": "edit",
  "prompt": "Change the cup to red; preserve the background and lighting.",
  "image_urls": [
    "https://cdn.example.com/source.png"
  ],
  "resolution": "1k",
  "aspect_ratio": "auto"
}
```

## Billing and results

Successful generation and editing use the same fixed per-image rate for the resolution tier. Reference count, steps, seed and format add no fee.

Read current rates in the model market or price\_config from GET /v1/models. Images are in result.images and videos in result.videos; url is an array. Failed or cancelled tasks receive a full reservation refund. Read final cost in integer quota (500,000 quota = 1 USD) and confirm billing\_status is settled.

For callbacks, waiting, idempotency and spending limits, see [Submit a task](https://docs.qingbo.dev/en/api-reference/task/submit.md)。


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.