AI models and providers

Adcraft does not include AI usage: you connect your own provider accounts with API keys, and each provider bills you directly. No key is required to start; every key switches on a set of capabilities.

What each key powers#

Admin → Providers shows which keys are set and how many calls to each provider succeeded or failed in the last 24 hours.

The built-in catalogue#

Text#

Text models cost no credits per call; a concept run is charged a flat 1 credit.

Image#

Video#

Video models list a per-second credit figure, but product and presenter videos are charged the flat amounts on Selling it as SaaS.

Other#

Model names and endpoints are those available when this version was built. Providers rename and retire models; if one stops working, update or disable it in the admin panel.

Managing models#

Admin → Models & pricing is the model catalogue. Without a redeploy, a platform admin can:

• enable or disable a model (disabled models disappear from pickers);
• make a model the default for its kind (text, image or video);
• change its credits per image, per second or per run;
• edit its endpoint, options and limits, or reset a built-in model to its shipped settings;
• add a custom model on fal.ai, Replicate, Runway, OpenAI, xAI or Anthropic, or any OpenAI-compatible text endpoint (Gemini, Groq, Together, Mistral, DeepSeek, OpenRouter, a self-hosted server) with a base URL and the name of the environment variable that holds its key;
• Test a model, which makes one real, small generation and shows the latency and cost. Testing a video model makes a five-second clip. Tests are billed by the provider.

If the default model's provider is not connected, Adcraft falls back to the first connected model of that kind. The catalogue page also shows measured cost and margin per model from real generations.

Estimating costs#

Adcraft has no markup on AI: your providers bill you directly. Prices change often, so this page links to the source rather than quoting numbers that would go stale.

A useful way to think about one ad:

cost of a static ad      ≈ image model price × sizes generated
cost of a product video  ≈ scene stills + video seconds per scene × scenes
cost of a presenter ad   ≈ HeyGen presenter render + voice-over + B-roll stills and clips

Video is the largest variable cost for most installs; a cheaper video model is the main lever. Measure a few real runs on your own accounts before you set credit prices for customers.

Tracking costs in the admin panel#

Every provider call is recorded as a generation event with its model, status, credits and an estimated cost in US dollars, shown in Admin → Generations & costs and on the overview. The estimate comes from each model's approximate price in the catalogue (and token usage for text). Models without an approximate price record no cost, so treat the figures as a guide and reconcile them with your provider invoices. The monthly provider cost guardrail uses the same estimates.