Flow Builder

A visual canvas for connecting image, video, and audio generation into reusable workflows.

Creative toolsAI
Flow Builder canvas after a run: a FLUX image feeds a Runway Gen-4.5 video, and the video and an ElevenLabs line meet in a composition node
The whole graph after one real run: FLUX dev on fal.ai for the image, Runway Gen-4.5 animating it for 4 seconds, ElevenLabs for the line, then the composition.
Status
Working prototype
Tools
Next.js, React Flow, Zustand
Links
Open prototypeSource

Flow Builder makes a media pipeline visible as a graph. Each node owns one step, with its inputs, settings, output, and execution state together. The canvas lets a person revise part of a workflow while retaining the surrounding structure.

Run all, in dependency order: the image first, then the video from it and the speech, then the composition. The outputs are the real results of one run of this graph; the wait for the providers is cut from the recording.

Decisions

Make connections meaningful

Text, image, video, and audio handles have distinct types. A wire only connects compatible handles, so an invalid media combination is caught during editing. Running the graph follows its dependency order.

Keep failures at the step that failed

Provider settings and task errors live on their nodes. A stub mode allows the entire interaction to be explored without credentials, while configured routes can call generation providers. The two modes need to be distinguishable in any demonstration.