Runware Run is the foundation Openclaw Skills skill that teaches agents how to execute Runware tasks safely, schema-valid, and cost-aware across every modality.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install runware-run
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install runware-run using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Runware Run?
Runware Run is the execution contract for Openclaw Skills that turns a model choice into a real, reliable Runware call.
It tells agents how to select the right AIR and taskType, resolve the model’s JSON Schema, upload or encode inputs correctly, choose synchronous or asynchronous delivery, and read the modality-specific response without guessing. The result is fewer invalid requests, cleaner automation, and predictable cost control whether you call Runware through the TypeScript SDK, Python SDK, MCP tools, CLI, or the OpenAI-compatible endpoint at api.runware.ai/v1.
Runware Run Use Cases
Openclaw Skills teams use this skill when:
- An agent is about to make a real Runware generation call and needs the correct execution flow.
- You must choose between imageInference, videoInference, audioInference, or modelUpload.
- A newly launched model is not yet in the SDK registry and needs an explicit taskType.
- You need to validate a payload against the model schema before sending it live.
- Local files must be uploaded first and referenced as URLs or base64 inputs.
- A long-running job needs async polling with taskUUID and getResponse.
- You want cost visibility through X-Runware-Dry-Run: 1 or includeCost: true.
- The same Runware contract must work across SDK, MCP, CLI, or API surfaces.
How Runware Run Works
- Pick the model and execution type. Retrieve the model from runware-models, choose the AIR, and set taskType based on modality; pass taskType explicitly for newly launched or unregistered models.
- Resolve the schema. Inspect the model’s JSON Schema and populate only documented field names; never invent parameter names.
- Prepare inputs. Send images, video, audio, masks, and other assets as URLs or base64; upload local files first when using SDK or MCP helpers.
- Execute with the right delivery mode. Use synchronous calls for images and other fast tasks; use asynchronous execution for video, audio, 3D, and training, then poll getResponse using the returned taskUUID until terminal.
- Read the terminal output. Consume modality-specific results such as image URLs, videoURL, or outputs.files[].url for 3D, and capture cost with includeCost when needed.
- Validate and optimize before production. Use X-Runware-Dry-Run: 1 to verify the request and preview cost without executing or being charged.
Runware Run Setup
- Install the surface you plan to use.
npm install @runware/sdk
pip install runware
- Prefer the SDK or MCP when possible, because they handle schema resolution, file upload, and async polling for you.
- Before production, validate uncertain payloads with
X-Runware-Dry-Run: 1and enableincludeCost: trueon live runs. - For long-running jobs, store the returned taskUUID and poll getResponse until the task reaches a terminal state.
Runware Run Data Schema & Taxonomy
Runware Run follows a schema-first request/response model:
| Entity | Purpose | Rules |
|---|---|---|
taskType |
Selects the modality pipeline | Use imageInference, videoInference, audioInference, or modelUpload |
model / AIR |
Identifies the model to run | Use AIRs such as bfl:7@1; slugs also resolve |
| Parameters | Model-specific options | Must match the model’s JSON Schema exactly |
| Inputs | Media and reference assets | Pass URLs or base64; upload local files first |
| Delivery mode | Controls runtime behavior | Sync for images and fast tasks, async for video, audio, 3D, and training |
taskUUID |
Async job handle | Returned for long-running tasks; poll getResponse |
| Outputs | Terminal artifacts | Images return URLs, video returns videoURL, 3D returns outputs.files[].url |
| Cost metadata | Pricing visibility | X-Runware-Dry-Run: 1 for estimate-only validation, includeCost: true for live runs |
Metadata taxonomy
- Identity: AIR or slug, plus
taskTypewhen required. - Input payload: prompt, reference media, masks, audio, training assets, and any other schema-approved fields.
- Execution metadata: sync vs async, returned
taskUUID, and polling state. - Billing metadata: dry-run validation and live cost reporting.
- Output artifacts: modality-specific URLs or file lists, never a single assumed response shape.
Runware Run Advanced Features
- Schema resolution before execution eliminates guesswork and invalid parameter names.
- Dry-run mode validates request shape and returns cost without executing or charging.
includeCost: truesurfaces real spend on live runs for budget-aware automation.- Async orchestration with
taskUUIDandgetResponsekeeps long jobs out of sync timeouts. - Cross-surface parity across the TypeScript SDK, Python SDK, MCP, CLI, and api.runware.ai/v1.
- AIR-based addressing plus explicit taskType supports newly launched or test models before registry updates.
- Built-in upload flow for local files, reference images, source video, audio, and masks.
SKILL.md
Loading
Related Openclaw Skills
Text in Image Generation
A specialized workflow for generating high-fidelity images with accurate, legible, and styled text layers.
Voiceover and Narration
Openclaw Skills turns scripts into expressive one-speaker voiceover audio with bracket-based direction, voice selection, and model-aware normalization.
Open Source Issue-First Workflow
An Openclaw Skills workflow that forces you to search upstream issues and releases before writing custom workarounds for open-source bugs.
AI Music Composer
A professional end-to-end AI music generator that clones your voice and composes complete, personalized songs based on your lyrics and style preferences.
Vision and Media Understanding
Openclaw Skills turns media inputs into grounded text outputs like captions, OCR, transcripts, summaries, tags, and structured JSON.
Restore and Upscale Images
Openclaw Skills restores blurry, noisy, or low-resolution images with the right model for sharper, more faithful results.
Featured*
