
Two ways to build multi-step AI media pipelines on fal.ai. Mode A generates declarative JSON workflow files that run in fal's cloud runtime, good when you need a portable, shareable asset someone else can execute. Mode B orchestrates the genmedia CLI locally with scripts, better for exploratory work or pipelines with conditional logic. The skill includes model catalogs, output reference syntax (like $node.images.0.url), and patterns for fan-out generation and frame bridging. The JSON mode is picky about structure: only "run" and "display" node types exist, no string interpolation allowed, and dependencies must explicitly match every reference. If you're chaining Flux to video models or building dataset generators, this covers both the declarative and imperative approaches.
npx -y skills add fal-ai-community/skills --skill fal-workflow --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.