> ## 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.

# GPT Image 2.6 Official

> Flare and Sunburst with six quality options, planning effort, masks and annotation editing, billed by actual tokens

Flare and Sunburst with six quality options, planning effort, masks and annotation editing, billed by actual tokens.

Model IDs: `gpt-image-2.6-flare` / `gpt-image-2.6-sunburst`.

## 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: image26-demo-001" \
  -d '{
  "model": "gpt-image-2.6-flare",
  "action": "generate",
  "prompt": "A sculptural glass vase on ivory stone, soft side light, refined editorial photography",
  "resolution": "1k",
  "aspect_ratio": "4:5",
  "quality": "medium",
  "complexity": "high",
  "n": 1
}'
```

Read `task_id` after submission and poll `GET /v1/tasks/{task_id}`. Download files from `result.images[].url[]` on success. Continue polling the original task after a timeout; do not submit a duplicate.

## Parameters

| Parameter | Values and rules |
| - | - |
| `action` | generate for text generation; edit requires image\_urls. |
| `prompt` | Required description or edit instruction. |
| `image_urls` | 1–16 publicly accessible reference images for editing. |
| `resolution` | 1k / 2k / 4k; defaults to 1k. |
| `aspect_ratio` | 1:1 / 3:2 / 2:3 / 4:3 / 3:4 / 5:4 / 4:5 / 16:9 / 9:16 / 2:1 / 1:2 / 21:9 / 9:21 / 3:1 / 1:3. Default: 1:1. |
| `quality` | low / medium / high / xhigh / max / auto; defaults to low. |
| `complexity` | low / medium / high, defaults to high. Planning effort is independent of quality and has no separate fee. |
| `n` | 1–4, defaults to 1. Reservation scales with image count. |
| `output_format` | png / jpeg / webp / gif. Default: png. |
| `mask_url` | Alpha PNG mask matching the first reference; transparent pixels mark the edit area. Requires image\_urls. |
| `markup_url` | Transparent PNG annotation layer matching the first reference. edit only, requires references, mutually exclusive with mask\_url. |
| `background` | auto / opaque / transparent. Transparent backgrounds require PNG or WebP. |
| `moderation` | auto / low; defaults to low. |
| `output_compression` | Integer 0–100 for JPEG/WebP only. |
| `safety_identifier` | Optional stable end-user identifier, up to 128 characters; use a hash. |

## Reference editing

```json theme={"system"}
{
  "model": "gpt-image-2.6-sunburst",
  "action": "edit",
  "prompt": "Preserve the subject and composition. Change the lighting to warm late-afternoon sunlight.",
  "image_urls": [
    "https://cdn.example.com/source.png"
  ],
  "resolution": "1k",
  "quality": "low",
  "complexity": "low",
  "n": 1
}
```

## GIF animation

Both official routes support `output_format: "gif"` for text or reference-image generation. Defaults are 512×512, 9 frames and 200ms per frame. `n` counts animations, not frames.

| Parameter | GIF rules |
| - | - |
| `size` | Per-frame pixels: auto / 512x512 / 768x768. auto means 512x512. |
| `frame_count` | Integer 2–16, default 9. 768x768 supports at most 12 frames. |
| `frame_duration_ms` | Default 200, minimum 20, in multiples of 10. Does not change the price. |
| `quality` | low / medium / high; defaults to low. |
| `resolution` / `aspect_ratio` | Keep the defaults 1k / 1:1; size determines per-frame pixels. |
| `n` | 1–4 animations; reserve one complete sprite sheet per animation. |

```json theme={"system"}
{
  "model": "gpt-image-2.6-flare",
  "action": "generate",
  "prompt": "An ivory paper lantern sways in a soft night breeze, warm amber light, deep navy background, seamless loop",
  "output_format": "gif",
  "size": "512x512",
  "frame_count": 9,
  "frame_duration_ms": 200,
  "quality": "low",
  "n": 1
}
```

Poll the same task endpoint and download the GIF from `result.images[].url[]`. Reservation uses the full sprite-sheet output token tier plus prompt and reference-input allowance for each animation. Settlement uses actual typed tokens within the hold and refunds the difference; it does not multiply a static-image price by frame count.

For example, 512x512 / 9 frames / low reserves 1172 output tokens ($0.031644), plus prompt allowance. A real receipt used 68 text input + 1172 image output tokens and settled at $0.03195. This is an example receipt; obtain a fresh quote before submission.

### Reference image to GIF

Both official variants accept up to 16 reference images. `action: "edit"` requires at least one `image_urls` entry; `n=2` requests two complete animations. Input fees are reserved for every animation, then settled from the reported aggregate usage. References are not charged once per frame.

GIF `size` describes one frame and does not follow the reference image aspect ratio.

```json theme={"system"}
{
  "model": "gpt-image-2.6-flare",
  "action": "edit",
  "prompt": "Keep the amber perfume bottle and fern. A warm ray of light glides gently through the glass, creating delicate caustics; the fern sways slightly. Refined product photography, stable camera, complete bottle visible on a square canvas, seamless loop, no text.",
  "image_urls": [
    "https://s3.qingbo.ai/qingbo-public/waveapi/site/assets/2026/11/428b6c75-654f-4cdc-8310-ba9ac545f2de.png"
  ],
  "output_format": "gif",
  "size": "512x512",
  "frame_count": 9,
  "frame_duration_ms": 200,
  "quality": "medium",
  "n": 1
}
```

Economy variants do not support reference-image GIFs.

### Flare: 16-frame GIF from a reference

![Flare reference GIF](https://s3.qingbo.ai/qingbo-public/waveapi/site/assets/2026/11/830db81e-5007-4508-a868-3254d06c2c17.gif)

### Sunburst: 9-frame GIF from a reference

![Sunburst reference GIF](https://s3.qingbo.ai/qingbo-public/waveapi/site/assets/2026/11/eaa118f2-d355-4907-8f58-ede024fc27c4.gif)

Native looping GIFs from the named models, verified for frame count, duration and downloadable output.

## Billing

Charged by actual typed token usage, in USD per million tokens:

| Text input | Cached text | Image input | Cached image | Image output |
| - | - | - | - | - |
| 4.5 | 1.125 | 7.2 | 1.8 | 27 |

Quotes reserve prompt, reference and output usage. `quality: "auto"` reserves max; successful tasks settle actual usage within the hold. Version 2.6 low/medium usage differs from 2.5, so request a new quote. Safety rejections may charge processed input, with no charge for ungenerated output. Unconfirmed cost or usage remains billing\_status pending until reconciliation.

## Limits

SVG and exact pixel sizes for static images are not exposed. GIF supports only the square frame sizes and qualities listed above. Reference-image GIFs are supported. Masks, annotations, compression and transparent backgrounds remain unavailable for GIF.


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