Models, providers and cost control

How to choose which model does which step, keep generation costs visible, and switch providers without rewriting the project.

Structure and generation are separate decisions

The composition declares what should exist — an image, a video take, a music bed — and the provider layer decides who produces it. That separation is what lets you swap a model without touching the script.

Bring your own keys

You can route every generated asset through your own accounts. Nothing about the workflow requires the managed gateway, which is why the tooling can stay free.

Cost visibility before spend

Model rates are readable before you generate, and the API bills in credits with a per-model rate. Check the rate, then decide whether the step is worth it.

Step by step

1. 1. List what your key can reach

   Query the models endpoint to see which models your key routes to and what each one costs per unit.

2. 2. Assign models per step

   Cheap models for drafts and thumbnails, expensive ones only for the takes that will ship. Declare this in the project rather than per run.

3. 3. Pin the models that decide the format

   Whatever generates the presenter or the hero shot should be pinned; experimenting there changes the format itself.

4. 4. Fail over deliberately

   If a 503 or 504 repeats for one step, route that step to a second provider rather than retrying the same vendor indefinitely.

5. 5. Review the bill against the structure

   If cost per variant is not falling as you re-run, the problem is usually that generation, not structure, is doing the work.

After this

The healthy pattern: fixed structure, cheap drafts, expensive finals, and a per-model budget you can state out loud before you press run.

Keep reading

• All guides
• Examples
• Clone formats
• Glossary
• Quickstart
• Pricing

FAQ