fal

fal - Claude Code Plugin

Generate images, videos, audio, and more using fal.ai AI models directly from Claude Code.

Installation

/plugin marketplace add Cdlane24399/fal-ai-skill
/plugin install fal@fal-ai-skill

Setup

1. Get your API key at fal.ai/dashboard/keys
2. The setup script is interactive - it prompts for your key directly. Run it yourself in the Claude Code prompt:

! bash skills/fal/scripts/setup.sh --add-fal-key

This saves your key to a .env file in your current working directory. The plugin reads it automatically from there.

If you already have a key, you can also create the .env file manually:

echo 'FAL_KEY="your-key-here"' > .env

Capabilities

Category	Examples
Image Generation	Text-to-image with Flux, Gemini (Nano Banana Pro), Ideogram, Recraft
Video Generation	Text-to-video with Veo 3.1, Seedance 2.0, Sora 2; image-to-video with Kling, Veo 3.1
Audio	Text-to-speech (MiniMax, ElevenLabs), speech-to-text (Whisper)
Image Editing	Style transfer, object removal (Finegrain Eraser), background replacement, inpainting
Upscaling	Image and video upscaling (Aura SR, Clarity, Topaz)
Workflows	Chain multiple models into pipelines
Platform	Pricing, usage tracking, cost estimation

Quick Start

Just ask Claude naturally:

• "Generate an image of a mountain landscape"
• "Create a video of ocean waves"
• "Convert this text to speech"
• "Remove the background from this image"
• "What fal models are available for video generation?"

Model Discovery

The plugin includes the fal MCP server for searching documentation and discovering models. It also provides scripts for fetching exact OpenAPI schemas:

/fal search models for text-to-video
/fal get schema for fal-ai/flux/dev

Troubleshooting

FAL_KEY not set: Run ! bash skills/fal/scripts/setup.sh --add-fal-key in the Claude Code prompt to interactively set your key. It saves to .env in your working directory.

Video generation timeout: Video models can take several minutes. The plugin uses queue mode by default with a 10-minute timeout. For very long tasks, use async mode: /fal generate --mode async --prompt "..." and check status later.

Unknown model parameters: Use /fal get schema for MODEL_ID to see the exact parameters a model accepts. Not all models support all options (e.g., image_size, num_images).

License

MIT