Skip to content

endpoint-per-model

Serverless serving for community image models. Paste a Civitai link, get a dedicated API endpoint: the exact checkpoint, byte-verified, warm in seconds, priced at a deterministic compute floor. Stack LoRAs on any base through the same API, with every component owner paid per generation.

What makes it different:

  • Every model is its own endpoint. Not "SDXL with a LoRA URL parameter" — the specific community checkpoint, compiled once, swapped into resident GPU memory in ~300 ms, bitwise-identical on every render.
  • Provenance you can prove. Each endpoint carries a signed attestation binding what it serves to the exact Civitai source bytes and the verified owner. Verify it offline — no trust in our API required.
  • Composition with attribution. Stack up to 5 LoRAs on a base you don't own; the ledger splits every generation between the component owners in one atomic event. How composition works.
  • Deterministic pricing. floor = fixed + per-step·megapixel — quotable before the render. Owners price above floor with a multiplier; the spread is theirs. Serving model & pricing.

Start here: the Quickstart goes from API key to a rendered image in four commands. The API reference is generated from the live route code on every merge — it cannot drift.

Status

The API is in pre-launch staging. Base URL and auth are stable.