AI Video Generation Configuration

TinyShip supports AI video generation with multiple providers. This document explains how to configure AI video generation.

Related Pages

Supported Video Providers

Configuration

Video generation config is defined in config/aiVideo.ts:

// config/aiVideo.ts
export const aiVideoConfig = {
  defaultProvider: 'volcengine' as const, // default video generation provider

  defaultModels: {
    fal: 'fal-ai/kling-video/v3/pro/text-to-video',
    volcengine: 'doubao-seedance-2-5-260628',
    aliyun: 'wan3.0-video',
  },

  availableModels: {
    fal: [
      'fal-ai/kling-video/v3/pro/text-to-video',
      'fal-ai/kling-video/v3/pro/image-to-video',
      'fal-ai/veo3.1',
      'fal-ai/veo3.1/image-to-video',
    ],
    volcengine: ['doubao-seedance-2-5-260628', 'doubao-seedance-2-0-260128'],
    aliyun: ['wan3.0-video', 'wan3.0-video-prime'],
  },

  defaults: {
    duration: 5,        // default duration; users can change it in the UI
    promptExtend: true,
    watermark: false,
  },
}

Durations and resolutions are filtered by provider + model and exposed to all three front-end apps via getVideoDurationsForProvider() / getVideoSizesForProvider(). Users can pick what they want — nothing is locked down.

Duration and Resolution

The default duration is 5 seconds. Volcengine billing scales with duration and resolution, so keep the default at 5s + 720P to avoid someone accidentally picking 30 seconds.

Environment Variables

Add provider keys in .env:

# Volcengine
VOLCENGINE_API_KEY="your-volcengine-api-key"

# Aliyun
ALIYUN_ACCESS_KEY_ID="your-aliyun-access-key-id"
ALIYUN_ACCESS_KEY_SECRET="your-aliyun-access-key-secret"

# Fal.ai
FAL_API_KEY="your-fal-api-key"

Credits Configuration

AI video generation usually costs more credits than image generation. You can price per model in config/credits.ts:

// config/credits.ts
export const creditsConfig = {
  fixedConsumption: {
    // AI video generation - priced per model
    aiVideo: {
      default: 30,
      models: {
        'fal-ai/kling-video/v3/pro/text-to-video': 35,
        'fal-ai/kling-video/v3/pro/image-to-video': 35,
        'doubao-seedance-2-5-260628': 40,
        'doubao-seedance-2-0-260128': 35,
        'wan3.0-video': 25,
        'wan3.0-video-prime': 35,
      },
    },
  },
}

This is a fixed charge for end users: the same model costs the same whether it runs 5 or 30 seconds. Volcengine Ark bills differently — see the next section.

Volcengine Seedance Billing

The Seedance API is billed by tokens on the Volcengine side — a separate ledger from TinyShip credits. The token count roughly depends on output duration, resolution, frame rate, and whether a reference video is included; use the task's returned usage.completion_tokens as the source of truth.

2.5 costs about 50% more per token than 2.0:

Typical prices for 2.5 pure text-to-video (no reference video):

Image-to-video (first frame only) usually still falls into the "input without video" tier. Only inputs with a reference video get the cheaper token rate, but the input duration counts toward tokens.

The 2.5 page currently only exposes 720P; the adapter can already map 854x480, so you can add 480P to volcengineSizes720 if you want to stress-test costs.

Official pricing: Volcengine Ark model pricing

Local Testing Tips

Don't use the default 2.5 + long durations when you just want to verify "can it produce a video".

Fal / Veo run on overseas nodes. When testing in China, the dev server (Node) doesn't use the browser proxy — you need TUN mode or set HTTPS_PROXY for the process, otherwise requests will be slow or time out. Qwen and Volcengine use domestic routes and usually don't need this.

Related docs:

• AI Chat Configuration
• AI Image Generation Configuration
• Credits System Configuration