Skip to main content
Runflow is a platform for automating photo workflows with AI-powered pipelines. This guide explains how the pieces fit together so you can pick the right primitive for your integration.

​
The three layers

LayerWhat it isWhere you touch it
ModelsProduction-ready AI models (image scoring, headshot generation, background removal, face edits, …)Never directly — you use pipelines or endpoints
PipelinesComposed sequences of models wired together with pre- and post-processingVia a single API endpoint per pipeline
FlowsHigher-level orchestrations that chain pipelines with inputs, triggers, and outputsBuilt visually in the Runflow app (coming to the API)
For v1 of the public API, you interact with pipelines as endpoints — one HTTP call per pipeline. Flow-level orchestration still lives in the Runflow app; the API surface is deliberately endpoint-first so you can wire Runflow into your own backends, queues, and user flows.

​
Pipelines (what the API exposes today)

A pipeline is a production-ready AI workflow behind a single endpoint. You don’t think about the models, the GPU, the pre/post steps, retries, or scaling — you send inputs and receive a result. Currently available pipelines:
PipelineEndpointTypeTypical latency
Score ImagePOST /api/v1/images/scoreSync< 1s
Generate HeadshotsPOST /api/v1/images/generate-headshotsAsync (webhook)1–2 hours
Synchronous pipelines return in the same HTTP call. Asynchronous pipelines acknowledge immediately and push the final result to a webhook URL you control. See Webhooks for the async pattern.

​
Flows (app-level, today)

In the Runflow app, a flow is a visual composition that wires together:
  • Inputs (images, user-submitted data, URLs)
  • Pipelines (one or more of the endpoints above, plus internal-only ones)
  • Transforms (filters, sorting, gating logic — e.g. “only process photos scoring above 0.7”)
  • Outputs (delivered via webhook, stored in S3, surfaced in the dashboard)
Flows are how non-engineers at your company can stitch together “if a photo scores below 0.6, surface a retake message; otherwise run headshot generation” without writing code. For v1 of the developer API, that orchestration is something you build on your side using the endpoints as building blocks — the flow-level API is on the roadmap.

​
Mental model for picking a primitive

“Can I express my use case as a single HTTP call?” Use the endpoint directly. This is 90% of integrations. “Do I need to chain multiple endpoints with conditional logic based on results?” Wire them together in your own backend. Runflow’s sync endpoints are fast enough that you can run them inline; for async ones, chain from the webhook handler. “Does my non-technical team want to build these workflows?” Point them at the Runflow app and let them build flows visually. Your API integration still uses the same endpoint URLs.

​
Where to go next

Score Image reference

The sync pipeline — use it as a quality gate in front of other endpoints.

Generate Headshots reference

The async pipeline with webhook callbacks.

Webhooks

Async callback handling and 24-hour expiry rules.

Vibe Coder Integration

Paste-ready prompts to have an AI agent wire Runflow into your app.