Workflows overview

What a Wavemaker workflow is, how Discover, Compose, Nodes, and Training Studio fit, and how to launch a run.

Last updated August 12, 2026

A workflow is a production pipeline you own: a graph of typed blocks — research, scripting, image and video generation, review gates, audio, assembly — that compiles into a repeatable, priced recipe. Chat asks Wavemaker to make creative decisions for you; a workflow makes those decisions once, explicitly, then runs the same way every time.

The pieces

  • Chat (/platform/chat) — describe, Decide-for-me, iterate.
  • Discover (/platform/discover) — catalog of published workflows and styles. Try, Remix, Flow.
  • The builder — four doors (Describe / Template / Import / Blank). Compose Copilot drafts graphs; Save as workflow is deferred (no eager create). Publish is a separate verified-run step.
  • Training Studio (/platform/studio) — datasets, LoRAs, voice clones, Character adapter.
  • Runs & credits — preflight quote, reserve, settle to actual.

Four ways to launch

  1. From chat — unbound Send is Decide-for-me; or bind a Discover workflow and Send.
  2. From a Discover listing — Try fills the session; promoted inputs become the form.
  3. From a workflow you own — run the bound graph from chat / Nodes after Save.
  4. From API or MCPPOST /runs (and run_workflow / submit_run). See API & MCP.

What workflows guarantee

  • No invisible spend. Preflight quotes GPU + partner meters; reserve is the ceiling; settle refunds unused credits.
  • Nothing executes uncompiled. Typed ports make invalid wiring unbuildable.
  • Reproducibility. Published versions pin the graph and bound assets.
  • Safety is structural. Generative lanes always pass platform QC — you cannot wire around it.

Where to go next