Token-Based Billing

Some models (SeedDream, Bedrock LLMs) use token-based billing where cost scales with actual usage rather than fixed credits.

How It Works

Image Tokens (SeedDream)

SeedDream models generate at native resolution regardless of requested size. Cost is based on output pixels:

output_tokens = (native_width × native_height) / 256
cost_pln = (tokens / 1,000,000) × rate_usd × 4.0 × 1.5

Example: SeedDream 5.0 Lite

Request Size	Native Size	Tokens	Cost (PLN)
512×512	2048×2048	16,384	~0.20
1024×1024	2048×2048	16,384	~0.20
1024×1536	2048×2048	16,384	~0.20

INFO

SeedDream 5.0 always generates at 2048×2048 native resolution, then downscales to requested size. This means cost is constant regardless of requested dimensions.

Rate

Model	Rate (USD/1M tokens)
SeedDream 5.0 Lite	$2.00
SeedDream 5.0 Pro (Dola)	$2.70
SeedDream 4.5	$2.40
SeedDream 4.0	$1.80

LLM Tokens (Bedrock)

Bedrock models bill per input + output token separately:

cost = (input_tokens × input_rate) + (output_tokens × output_rate)
cost_pln = cost_usd × 4.0 × 1.5

Model	Input $/1M	Output $/1M
Claude Sonnet 4.6	$3.00	$15.00
Claude Haiku 4.5	$0.80	$4.00
Nova Pro	$0.80	$3.20
Nova Lite	$0.25	$1.00
Nova Micro	$0.10	$0.40

Billing Response

Token-based responses include detailed cost breakdown:

{
  "billing": {
    "method": "token",
    "credits_used": 2,
    "pln_charged": 0,
    "cost_breakdown": {
      "output_tokens": 16384,
      "cost_usd": 0.033,
      "cost_pln": 0.20,
      "rate_per_1m_tokens_usd": 2.00
    }
  }
}

Verifying Token Billing

Use the SDK to verify billing is correct:

from fotohub import FotohubClient

client = FotohubClient()

result = client.generate_image(
    prompt="Test image",
    model="seedream-5-0-260128"
)

# Verify token calculation
expected_tokens = (2048 * 2048) / 256  # = 16384
actual_tokens = result.billing.output_tokens

assert actual_tokens == expected_tokens, f"Expected {expected_tokens}, got {actual_tokens}"
print(f"Token billing verified: {actual_tokens} tokens = {result.billing.cost_pln} PLN")