Runway Agent is a chat-based collaborative agent that analyzes what you give it — a product, image, an ad campaign, or idea — and uses what it learns to create what comes next. Instead of generating a single asset in isolation, Agent can plan, produce, and scale entire creative projects while picking the best model for the job at each step.

A built-in timeline editor lets you cut, reorder, and upload additional assets when generating or otherwise working with video content.

This article walks you through the basics of creating with Agent, with examples to show what's possible.

What can Agent do?

Agent has a wide range of functionality across planning, generating, analyzing, and editing, and new capabilities are added regularly.

Generate

• Images — from text, reference images, or both. Good for character references, product shots, style exploration, CTAs, and end cards.
• Video — text-to-video or image-to-video, single shots or multi-shot sequences, with or without native audio depending on the model.
• Audio — sound effects, ambience, foley, dialogue scenes, and audio stories from text or reference audio/images.
• Music — standalone instrumental tracks from a descriptive prompt.
• Voiceover — scripted narration using a catalog of cast voices.

Edit

• Video editing — restyle, change environments/seasons/lighting, swap or alter elements in an existing clip, all while preserving the original motion and framing.
• Image editing — modify or extend existing images using text instructions or reference images.
• Timeline editing — arrange clips, voiceover, and music into a cut in the built-in editor.

Create & organize

• Build and render timelines (multi-clip cuts)
• Generate reference sheets, storyboards, mood boards, and character sheets
• Write scripts and narrative structure
• Diagnose and fix prompts that aren't working

What to keep in mind when working with Agent

Agent can help plan your project, provide suggestions, and will communicate its intent for each generation step. However, generating media is inherently variable — Agent's plan describes its intent, not a guaranteed outcome. Like all AI models, Agent can make mistakes at times, and results may require iteration to get right.

Step 1 – Accessing Agent

1. Sign in to your Runway account
2. Click Agent in the left sidebar, or go directly to app.runwayml.com/agent

From here, you have two ways to start:

• New chat — start a fresh Agent conversation to plan, generate, or produce a new project from scratch.
• New timeline ("Edit existing footage") — jump straight into the timeline editor to arrange and trim footage you already have.

If you've used Agent before, you'll land on your most recent session. To start a new one, double-click Agent in the left sidebar, or click the session name in the top left and select Start New Chat.

Agent settings

Before or during an Agent chat, you can configure whether Agent waits for your approval before generating and what it optimizes for when selecting models. These settings are available in the bottom right corner of the prompt panel. 

Once you've configure your preferences, click Set as default for new sessions to apply your choice to all new sessions going forward.

Agent generations use credits, and costs vary by model and output type. For a full cost breakdown, see What are my credits worth?

Generation mode

Choose when Agent generates media:

• Ask before generating media: Agent shows you its plan for each step — including the model, prompt, and estimated credit cost — and waits for your go-ahead before generating.
• Automatically generate (default): Agent generates immediately after planning each step, without waiting for confirmation.

Model selection preference

Under Optimize generations for…, choose how Agent picks models:

• Speed: Prioritize faster generations
• Cost: Prioritize lower credit usage
• Quality: Prioritize the highest-quality output
• Custom — write your own natural language instructions, including preferred models, resolutions and aspect ratios. See Setting custom generation preferences in Runway Agent

Agent weighs this preference when selecting a model, but will always choose a model capable of fulfilling your request — if only one model can do the job, Agent uses it regardless of your preference setting.

This defaults to Quality on desktop and Speed on mobile. 

Step 2 – Prompting Agent

The prompt sets the creative foundation of what Agent generates. 

1. Click into the message box at the top of the panel
2. Type a description of what you want to start creating in natural language

Think of your prompt as a simple description of where to start. A single, simple sentence works well in most cases, as Agent will ask for more details as needed to ensure alignment before generating.

After entering your prompt and optional assets, send the first message to have Agent start analyzing the request.

You can also type / in the message box to trigger an Agent Skill — a ready-made workflow like Ad Campaign or Mood Board that guides Agent through a common creative task. See Using agent skills to learn more.

Example prompts

Some examples of initial prompts include, but are not limited to:

Step 3 – Reviewing and Refining with Agent

After submitting your initial message, Agent analyzes your request to build a plan and may ask questions to ensure it generates what you want. Review this plan and let Agent know whether you wish to:

• Proceed with generating or
• Refine and workshop the concept before generating

If you're ready to proceed, simply let Agent know that you like the plan. 

If changes are needed before starting the generation, you can provide feedback directly through the chat for automated updates.

Agent may also request more information to ensure it generates what you want. 

Everything Agent generates during a session is collected in the Generations tab, where you can view it all as a grid.

Step 4 – Touching up videos

If Agent creates a multi-shot video, it will automatically combine your clips and audio in the Final Cut tab — where Agent assembles your videos, and where you can also manually trim, split, and arrange generated or uploaded media yourself.

For minor adjustments, use the tools directly in Final Cut. Alternatively, you can ask Agent to make edits for you through chat.

Split

Split separates media files into two. To use Split to separate a single media clip:

1. Click the media you want to split in the timeline editor
2. Move the scrubber to the location to split
3. Click Split

Selecting a media clip (which adds the white outline) will split only that file. To split all tracks at a point, make sure no single media clip is selected.

Examples

Agent works across a wide range of creative use cases. Studios, marketing teams, and individual creators can use Agent for short films, motion graphics, social content, and more.

Use case	What you bring	What you ask	What Agent creates
Campaign creation		Tell me the strongest angles for this product and build a campaign around it.
Ad testing		Build three stronger variations of this ad to test.
Film & narrative		Create a multi-shot scene with this character across three different locations.
Performance analysis		Analyze what's working in this campaign and create new creative that builds on it.
Social content		Make 6 days of social content in the style of my recent posts. Use a 1:1 aspect ratio.

Next Steps

From here, you can take your Agent output further with these tools:

• Upscale Video — upscale your finished video to 4K quality
• Ad Localization — take an image ad and generate localized versions for different markets in seconds
• Start a new Agent session — return to Agent to analyze, create, or scale your next project

FAQs