AI Providers

This page focuses on three core topics: how to configure models, how pricing is configured, and what dynamic pricing is currently supported.

Admin Configuration Screenshot

1. How to Configure Models (Fast Path)

In Admin -> AI Models, configure in this order:

1. Choose provider (fal / replicate / kie)
2. Set model (for example fal-ai/flux/dev)
3. Choose modality (image / video / audio / text)
4. Set enabled and fallback creditCost
5. Define paramSchema (allowed params + enum options)
6. Enable pricingRule when you need dynamic billing

2. How Pricing Works (Current Capabilities)

Pricing currently supports two layers:

• Fallback fixed price: creditCost
• Formula pricing: pricingRule (takes priority when present)

2.1 Fixed Price

Best for initial launch and low complexity.

2.2 Unit-Based Dynamic Pricing

pricingRule.unit currently supports:

• image: charged by output count
• second: charged by duration (video/audio)
• token: charged by text token usage (per 1k tokens)

2.3 Parameter-Based Dynamic Pricing

Yes, this is supported now.

• Define enum params in paramSchema (for example quality)
• Set per-attribute prices via pricing rule editor
• Or use tiers.match to apply conditional price tiers

So both "single-image pricing" and "dynamic pricing by model params" are supported.

3. Practical Pricing Rule Example

{
  "type": "formula",
  "billingMode": "charge_on_complete",
  "unit": "image",
  "price": { "perImage": 10 },
  "tiers": [
    { "match": { "quality": "high" }, "price": { "perImage": 16 } },
    { "match": { "quality": "ultra" }, "price": { "perImage": 24 } }
  ]
}

Meaning: default 10 credits per image, with automatic premium tiers for higher quality options.

4. Key Highlights

• Unified multi-provider management (fal/replicate/kie)
• Fixed + formula pricing can coexist
• Dynamic pricing by parameter enums and tier matching
• Separate estimate and final charge stages
• Price changes are operational, not code changes

5. Recommended Rollout Strategy

1. Start with creditCost for all models.
2. Move high-volume/high-cost models to pricingRule.
3. Add surcharges for expensive params (quality, duration, outputs).
4. Combine rate limits with pricing to cap risk.

Next Steps