Runway Agent Builds Entire AI Video Pipelines From Plain Language
Runway Agent now builds and edits node-based video pipelines from plain text descriptions, removing the need to wire nodes manually.
Takeaways
- Runway Agent now builds, edits, and runs node-based Workflows from plain-language descriptions via the new
/Workflowskill. - The feature removes the need to manually place and wire nodes in the Workflow Editor, replacing it with a single conversational prompt.
- Workflows chain Input nodes, Media model nodes (Gen-4.5, Veo, Kling, etc.), LLM nodes (Claude, Gemini), and utility nodes into automated pipelines.
- LLM nodes inside Workflows consume 3-4 credits per execution, so high-frequency pipelines can drain credit budgets quickly.
- Workflows are available on all paid plans starting at $12/month; a 30% discount applies with code
AGENTat app.runwayml.com/agent. - Early user sentiment is positive on time savings, but some flag that debugging agent-generated graphs can be harder than fixing manually built ones.
Runway Agent can now build, edit, and run entire node-based Workflows from a plain-language description. Type what you want, and the Agent assembles the pipeline: nodes placed, outputs wired to inputs, settings configured, ready to run.
What Workflows do
Workflows let you chain multiple AI models into a single automated pipeline. Instead of manually exporting from one tool and uploading to the next, every step runs in sequence inside one graph. The standard Runway interface handles quick generations well; Workflows handle the repeatable, multi-step processes that would otherwise require constant manual intervention.
Until now, building those pipelines still required manual work: knowing which nodes to place, how to connect them, and how to configure each one. This update removes that requirement entirely.
How to use it
The capability lives inside Agent as a Workflow skill. Type /Workflow in the chat, describe your pipeline in natural language, and the Agent creates the nodes, connects them, and runs the result. A pipeline like "generate a prompt with Claude, feed it into Gen-4.5, extract the last frame, then upscale it" goes from a sentence to a running graph without touching the canvas.
The node system
Every Workflow is built from four node types:
- Input nodes accept text or uploaded media
- Media model nodes handle image and video generation
- LLM nodes generate or modify prompts mid-pipeline, with a system prompt input for custom instructions
- Media utility nodes perform operations like stitching clips, extracting frames, or adding audio
Nodes are color-coded for quick navigation:
- Orange: Text
- Blue: Image
- Yellow: Audio
- Green: Video
LLM nodes support three backends: Gemini 2.5 Flash, Claude Sonnet 4.5, and Claude Opus 4.5. They're useful for automating prompt engineering and keeping outputs consistent across multi-step generations.
What you can build
The main advantage is repeatability. Save a Workflow as a template and your team can run the same pipeline on demand without rebuilding it each time. Pipelines people are already running include:
- Text-to-video chains that route prompts through an LLM node for refinement before hitting a Gen-4.5 video node
- Product shot pipelines that take a raw image, restyle it, upscale it, and extract the final frame in one run
- Character generation and storyboarding sequences that maintain visual consistency across scenes
- Short film, motion graphics, and social content production for studios and marketing teams
Where it breaks down
When the Agent misreads your intent and builds the wrong graph, debugging a structure you didn't construct yourself takes longer than fixing a node you placed manually. That's the core tradeoff with any abstraction layer: less setup friction, more investigation friction when something goes wrong.
Credit costs are also worth tracking carefully. LLM nodes charge per execution, and those costs compound fast in automated pipelines:
- Claude Opus 4.5: 4 credits per execution
- Claude Sonnet 4.5: 3 credits per execution
A Workflow triggered 100 times a month with a single Claude node consumes 300 to 400 credits before any media generation runs.
Pricing and availability
The Workflow skill is available now through the Agent interface. Every paid plan at Standard tier and above includes Workflows. Runway's current plan pricing (billed annually):
| Plan | Monthly price | Credits/month |
|---|---|---|
| Standard | $12 | 625 |
| Pro | $28 | 2,250 |
| Unlimited | $76 | 2,250 + Explore Mode |
| Enterprise | Custom | Custom |
Runway is offering 30% off any plan with the code AGENT to coincide with the Agent 2.0 launch. Try it at app.runwayml.com/agent.
The broader bet
Runway Agent already covers the full production pipeline: concept development, story beats, visual direction, multi-scene generation, voiceover, music, and final assembly. Natural-language Workflow construction extends that further, letting the Agent design pipelines rather than just execute them.
Most AI video tools have assumed that power users need to understand the underlying graph to get consistent, production-grade results. Runway is betting a capable enough agent can abstract that layer away. For teams currently spending more time wiring nodes than creating, it's a reasonable bet to test.