Skip to main content
Leap runs the best image, video and audio models behind one request shape. Pick a model, send its input, get back a file. You pay from one balance, you can price any run before you start it, and a run that fails is never charged. Besides raw models, the catalog has presets: ready-made recipes such as leap/restore-photo@1 or leap/headshot@1 that pair a model with an expert prompt. You call a preset exactly like a model.

Quickstart

Create a key and get your first image in one request.

Generations

Start a run, then wait for it, poll it or cancel it.

Models and schemas

Find a model and read the JSON Schema of its input.

Use it from your agent

Install the Leap skill, or connect over MCP.

How a request works

  1. Find a model. GET /v1/models lists every model and preset with its price. GET /v1/models/{creator}/{name} returns one with its input_schema, the JSON Schema of the input it takes.
  2. Start a run: send POST /v1/generations with {"model": "...", "input": {...}}.
  3. Get the result. Add prefer: wait=60 to get an image back in the same request, or poll GET /v1/generations/{id} until the status is final. Each output is a signed link to the file.

At a glance

Last modified on October 4, 2026