FlowRunner
PricingContact
Theme
Start Free

Luma AI

AI

Generate video and images with Luma AI's Dream Machine API. Agents turn prompts and keyframes into video clips with the Ray models, create images with Photon, extend, upscale, reframe, or add audio to results, and persist ephemeral assets to durable FlowRunner storage.

15 actions API key available
Luma AI website ↗ Platform Documentation ↗ Capability data verified 2026-07-27
A campaign brief marks three product clips as needed for launch week
Agent assembles prompts, keyframes, aspect ratios, and durations from the brief
Get Credits confirms the balance covers the renders before anything is queued
Generate Video renders each clip, and the finished assets land in durable FlowRunner file storage
Add Audio to Video scores the strongest cut with a synchronized track
The review channel receives the durable URLs with prompt and settings per clip
The marketing lead picks the cut that ships and decides whether a 4k upscale is worth the credits

What This Integration Enables

Luma's Dream Machine is a generation engine, not an asset library. Its asset URLs expire by design, and its generations run asynchronously. Both facts break naive automation: flows that treat a render like a synchronous API call fail, and flows that store Luma URLs end up linking to nothing. FlowRunner's connector is built around both. Generation actions poll to completion by default, then download the result and persist it to FlowRunner file storage, so every clip and image your flows produce has a URL that still works next quarter. From there, agents run full production chains: text-to-video with the Ray models under keyframe, camera-motion, looping, and resolution control, image generation with Photon steered by style, character, and modify references, and post-production passes that extend, upscale, reframe, or score existing generations. Long renders can be fired asynchronously and collected later with Wait for Generation, which is how a flow renders overnight and assembles results in the morning.

Without FlowRunner

Expiring asset links Finished renders live on ephemeral URLs that die before anyone downloads them
Blind credit burn Nobody knows what a render batch will cost until the balance is gone
One-off generations Every variation means re-prompting by hand and losing the settings that worked

With FlowRunner

Durable by default Completed generations are downloaded and saved to FlowRunner file storage automatically
Spend checked first Get Credits runs before expensive renders are queued, and a human approves the draw
Iteration as a pipeline Extend Video, Reframe Video, and Upscale Video chain from any prior generation id

Use Case Scenarios

Brief to Published Clip

A content agent takes a campaign brief, renders a hero clip with Generate Video using a product still as the start keyframe, and saves the result to durable storage. It then uploads the file to [WordPress](/integrations/wordpress) and drafts the post with the clip embedded. The draft, not the publish, is the agent's finish line; an editor reviews before anything goes live.

Consistent Product Imagery at Catalog Scale

An agent generates listing imagery with Generate Image, passing a Character Reference to keep the same model across every shot and Style References to hold the brand's art direction. Each finished image is pushed to [Dropbox](/integrations/dropbox) with a shared link handed to the team. When a seasonal campaign needs the same images in a new format, Reframe Image outpaints them to the new aspect ratio instead of regenerating from scratch.

Overnight Batch With a Morning Review

For a launch, the agent queues a dozen variations with Wait for Completion disabled, letting Luma work overnight. In the morning it sweeps List Generations, collects finished work with Wait for Generation, persists everything with Save Generation to Files, and posts the contact sheet to the review channel in [Slack](/integrations/slack). Failed generations are listed with their prompts so a person decides what gets re-queued rather than the agent silently doubling the spend.

Human-in-Loop Highlight

Rendering is spend. Every video and image generation draws down a finite credit balance, and the expensive options, 4k resolution and long durations, draw it down fastest. So the batch agent treats escalation to high resolution as a decision, not a default. Before upscaling, it posts to the review channel: "Six clips approved at 1080p. Upscale Video to 4k is requested for the hero placement. Current balance from Get Credits is attached; upscaling all six draws it down substantially. Upscale all, just the hero, or none?" The [human-in-the-loop](/concepts/human-in-the-loop/) owns the spend. The same review guards Delete Generation, which permanently removes a generation and cancels it mid-render if still in progress.

Agent processes routinely
Detects exception requiring judgment
Clear match Continues automatically
Ambiguous Routes to human via preferred channel
Human decides
Agent resumes with decision

Agent Capabilities

15 actions

Video Generation

5
  • Generate Video Renders video from a text prompt with Ray 2, Ray Flash 2, or Ray 1.6. Start and end keyframes accept image URLs or prior generation references, enabling extend, interpolate, and loop-back work. Controls aspect ratio, resolution up to 4k, duration, looping, and camera motion. Polls to completion and persists the asset by default.
  • Extend Video Continues an existing generation from its final frame, or bridges into another generation, from a prompt describing what happens next. A convenience wrapper that wires the keyframes for you.
  • Upscale Video Upscales a completed generation to a higher resolution, up to 4k. A credit-heavy call that batch flows route through human approval.
  • Reframe Video Outpaints a completed video to a new aspect ratio, generating surrounding content to fill the expanded frame. One render becomes every placement format.
  • Add Audio to Video Generates a synchronized audio track for a completed video, guided by a prompt and optional negative prompt.

Image Generation

2
  • Generate Image Creates images from a prompt with Photon 1 or Photon Flash 1. Image References guide content, Style References transfer art direction, a Character Reference locks a consistent identity, and a Modify Image Reference edits an existing image toward the prompt.
  • Reframe Image Outpaints an existing image to a new aspect ratio, with optional grid positioning of the original within the new canvas.

Generation Lifecycle

5
  • Get Generation Retrieves a generation's status and details as it moves queued, dreaming, completed, or failed. The polling primitive for fire-and-collect flows.
  • Wait for Generation Blocks until a generation reaches a terminal state, optionally downloading and persisting the finished assets. Bounded by a timeout so flows never hang.
  • Save Generation to Files Downloads a completed generation's video, image, or audio assets into FlowRunner file storage and returns durable URLs, before Luma's ephemeral links expire.
  • List Generations Lists generations newest first with state filtering and pagination. How morning-after flows reconcile overnight batches.
  • Delete Generation Permanently deletes a generation, cancelling it if still in progress. Cannot be undone, so cleanup flows route it for review.

Account and Reference

3
  • Get Credits Retrieves the account's available credit balance. Runs before expensive generations so spend is a checked decision, not a surprise.
  • Ping Verifies connectivity and API key validity. The setup smoke test.
  • List Camera Motions Lists the supported camera-motion concept keys, such as dolly_zoom and orbit_right, for use as Camera Concepts on Generate Video.

Frequently Asked Questions

What can FlowRunner do with Luma AI?

FlowRunner agents can run Generate Video, Extend Video, and Generate Image in Luma AI, plus 12 more actions.

Does connecting Luma AI to FlowRunner require OAuth?

No. Luma AI connects to FlowRunner with an API key, no OAuth flow required.

Can Luma AI trigger a FlowRunner workflow automatically?

Luma AI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Luma AI

$100 in credits. No card required. Connect in minutes.