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

# Pricing

> Credit costs per endpoint and how billing works.

## How credits work

The Dolphy API uses the same credit pool as dolphy.chat. Buy credits at
[dolphy.chat/credits](https://dolphy.chat/credits), and use them to pay for
both web and API usage. No separate API billing or invoicing.

| Tier       | Cost per credit | Notes                                 |
| ---------- | --------------- | ------------------------------------- |
| Member     | \$0.065         | Unlocks \~55% more credits per dollar |
| Non-member | \$0.0999        | Standard rate                         |

Members buy 153 credits for the same \$9.99 that non-members spend on 100.

## Chat completions

**1 credit per 10,000 tokens** (input + output combined). Minimum 1 credit
per call. Tracked via the `usage` block returned by the upstream model.

A typical short chat call (200 input + 300 output tokens = 500 total) costs
**1 credit**. A long-context conversation with 50,000 tokens costs **5
credits**.

## Image generation

Per-image cost depends on the model. From `GET /v1/models`:

| Model                                                                                                         | Credits/image | Quality tier    |
| ------------------------------------------------------------------------------------------------------------- | ------------- | --------------- |
| `lustify-v8`, `lustify-v7`, `chroma`, `z-image-turbo`, `wai-Illustrious`                                      | 1             | Specialty       |
| `hidream`, `qwen-image`                                                                                       | 1             | Standard        |
| `flux-2-pro`, `wan-2-7-text-to-image`, `recraft-v4`, `seedream-v4`, `seedream-v5-lite`                        | 1             | Standard        |
| `flux-2-max`, `hunyuan-image-v3`, `imagineart-1.5-pro`, `wan-2-7-pro-text-to-image`, `grok-imagine-image-pro` | 2             | Premium         |
| `nano-banana-2`, `qwen-image-2-pro`                                                                           | 3             | Pro tier        |
| `nano-banana-pro`                                                                                             | 5             | Pro+            |
| `gpt-image-2`, `gpt-image-1-5`                                                                                | 8             | OpenAI flagship |
| `recraft-v4-pro`                                                                                              | 9             | Top-tier        |

Multi-image requests (`n > 1`) bill linearly: `n × per-image cost`.

## Image editing

**2 credits per edit.** Single model — `seedream-4.5`. Specialty creative-mode.

## Video generation

Per-video cost depends on duration and resolution. Cheapest variant is 5s
480p at 10 credits; most expensive is 10s 1080p at 55 credits.

| Duration / Resolution | 480p | 720p | 1080p |
| --------------------- | ---- | ---- | ----- |
| 5s text-to-video      | 10   | 15   | 25    |
| 10s text-to-video     | 18   | 28   | 45    |
| 5s image-to-video     | 12   | 18   | 35    |
| 10s image-to-video    | 22   | 35   | 55    |

Use `POST /v1/video/quote` to get an exact credit cost before queueing.

## Motion control

**5 credits per second** of reference video. A 10-second clip costs 50
credits. Includes character swap + motion retargeting via Kling v3.

## What happens if you run out mid-stream

Chat: we still complete the response, deduct what we can, and your next
call returns `402` until you top up. We never cut off in the middle of a
streaming response.

Image / video / motion: pre-flight balance check. If you can't cover the
full cost, the call fails with `402` before we touch the upstream
provider. No partial billing.

## Failed requests

If a call fails on our side or upstream returns `5xx`, you don't get
charged. For video specifically, if a queued job times out or errors, we
**automatically refund** the credits you spent at queue time.