fal.ai

Run generative media models on fal.ai directly from your workflows. Agents generate and transform images, create video from text or images, upload files, and manage queued jobs across hosted models with sync and async execution.

11 actions API key available

fal.ai website ↗ Platform Documentation ↗ Capability data verified 2026-07-27

The content calendar calls for the launch week's product clip set

Upload File stages the product stills to fal storage as model inputs

Animate Image renders the first clip from the hero still with the brief's motion prompt

The creative lead reviews the first render and locks the prompt, model, and duration before the batch spends anything more

Run Model (Async) submits the remaining clips to the queue with the approved settings

Get Queue Status tracks each render as it moves from IN_QUEUE through IN_PROGRESS

Get Queue Result collects each finished video into FlowRunner file storage as it completes

Durable URLs post to the campaign channel, ready for scheduling

What This Integration Enables

fal.ai's real product is velocity: the newest image and video models, hosted behind one consistent API, usually within days of release. This connector passes that velocity through to your workflows. FlowRunner agents run any hosted model by ID with schema-matched JSON input, synchronously for fast models or through the queue for long renders, and use the friendly wrappers, Generate Image, Transform Image, Generate Video, Animate Image, when the standard cases are enough. Upload File stages input assets on fal storage, and because fal's output URLs expire, every result is downloaded into FlowRunner file storage by default and returned as a durable URL. The generic Run Model action is the strategic piece. When a better video model ships next quarter, your flow changes a model ID string, not its architecture. That is what it means to treat generative media as infrastructure rather than as a per-vendor integration project, and it pairs with a [human-in-the-loop](/concepts/human-in-the-loop) discipline for the renders that spend real money.

Without FlowRunner

Model access means engineering time Every new image or video model needs someone to hand-roll HTTP calls and auth

Renders babysat in browser tabs Long video jobs get watched manually because nothing else is tracking them

Output links that rot Generated media URLs expire, and last month's approved asset is a dead link

With FlowRunner

Any hosted model is a flow step One connector runs FLUX, Kling, Veo, and whatever ships next, by model ID

Queues managed, not watched Async jobs submit, poll, resume, and cancel under agent control

Every output made durable Results download into FlowRunner file storage automatically, with URLs that keep working

Use Case Scenarios

Product stills become launch videos

New product photography lands in [Dropbox](/integrations/dropbox-service). The agent stages each hero shot with Upload File, and Animate Image renders short clips with the campaign's motion language: slow push-in for the hero, orbit for the detail shots. After the creative lead approves the first render, the batch fans out through the queue and finished clips post with durable URLs. The launch gets motion assets in an afternoon, from photography that already existed.

The creative pipeline with interchangeable engines

A weekly content flow generates blog and social imagery with Generate Image on FLUX, then applies brand-consistent edits with Transform Image using Kontext: same wardrobe of prompts, new subjects each week. Prompts, model IDs, seeds, and durable URLs log to [Google Sheets](/integrations/google-sheets), so the team can reproduce any asset exactly and compare model output over time. When a new model tops the quality bar, the flow adopts it by changing one field.

Long renders under real management

A brand film sequence needs minutes of generated video across many shots. The agent submits everything with Run Model (Async), records request IDs, and polls Get Queue Status on a schedule, pulling each completed shot with Get Queue Result and posting progress to [Slack](/integrations/slack). When the director cuts a scene mid-render, Cancel Request kills the affected jobs while they are still in queue, and the budget goes to shots that will actually ship. Run Model and Wait covers the short jobs where blocking is simpler than bookkeeping.

Human-in-Loop Highlight

Video generation is the spend that compounds quietly: each Generate Video or Animate Image render takes minutes of paid GPU inference, and a batch of forty clips commits the budget forty times over to whatever the prompt actually says, not what the brief meant. This connector's flows therefore run dailies, not gambles. The first render goes to the creative lead as a review step, prompt, model, and duration attached, and the approval is what releases the batch. While the queue runs, Cancel Request is the recall lever: it can stop any job still IN_QUEUE or IN_PROGRESS, and an agent watching Get Queue Status can present the option the moment the brief changes, but the decision to kill paid renders mid-flight, trading spent inference against a wrong deliverable, stays with the person who owns the budget.

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

11 actions

Model Execution

Run Model Runs any fal.ai model synchronously by ID with schema-matched JSON input, saving expiring output URLs to durable file storage by default. The universal adapter for fast models.
Run Model (Async) Submits a run to the queue and returns immediately with a request ID and polling URLs, with optional webhook notification. The entry point for long renders.
Run Model and Wait Submits to the queue, polls, and returns the finished output in one blocking call, up to roughly 9.5 minutes. The convenience path for medium-length jobs.

Queue Management

Get Queue Status Retrieves a queued request's state: IN_QUEUE, IN_PROGRESS, or COMPLETED, with errors and optional runner logs.
Get Queue Result Retrieves a completed request's output, downloading expiring media into FlowRunner file storage by default.
Cancel Request Cancels a request still in queue or in progress. The recall lever this page's human gate controls.

Assets

Upload File Uploads a file from a public URL to fal storage and returns a fal URL for model input. The staging step for image-to-image and image-to-video work.

Media Generation

Generate Image Text-to-image across FLUX, Stable Diffusion 3.5, Recraft V3, and Ideogram V2 with friendly parameters: size presets, steps, guidance, negative prompt, seed. Durable URLs in savedFiles.
Transform Image Image-to-image editing with FLUX Kontext and related models: source URL plus a prompt describing the change, with strength control.
Generate Video Text-to-video across Kling, MiniMax Hailuo, Veo 3, Luma Dream Machine, and Wan, polling the queue until the render completes and saving durable output.
Animate Image Image-to-video with an optional motion prompt across Kling, Hailuo, Luma, and Stable Video Diffusion. The still-to-clip step behind launch videos.

Frequently Asked Questions

What can FlowRunner do with fal.ai?

FlowRunner agents can run Run Model, Run Model (Async), and Get Queue Status in fal.ai, plus 8 more actions.

Does connecting fal.ai to FlowRunner require OAuth?

No. fal.ai connects to FlowRunner with an API key, no OAuth flow required.

Can fal.ai trigger a FlowRunner workflow automatically?

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

Start building with fal.ai

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

Start Building Free Talk to us