TL;DR
- ISC-licensed Claude Code skill that wraps the fal.ai API. Adds 50+ generative models (Flux, Veo, Kling, Minimax, HiDream, ElevenLabs, and more) to your terminal workflow.
- Install by pasting the repo URL into Claude Code and asking it to install. Claude handles clone, build, API key setup, and verification end to end.
- No local GPU needed. Every job runs on fal.ai hosted GPUs and returns the result URL to your Claude Code chat.
System Requirements
| RAM | 4GB |
| GPU | None required |
| VRAM | N/A |
✓ Apple Silicon
If you already use Claude Code, adding AI image, video, audio, and music generation to your terminal workflow is now a single prompt away. The claude-skill-falai skill wires Claude Code directly into the fal.ai generative platform, giving your agent access to 50+ hosted models including Flux, Veo, Kling, Minimax, HiDream, ElevenLabs, and more. This guide covers the easy install path (just ask Claude Code to do it for you), the manual path if you prefer to see every step, the API key setup flow, and a few first prompts to confirm everything works.
What the Skill Actually Does
claude-skill-falai is an ISC-licensed Claude Code skill written by karamble. It wraps a small Go CLI binary that talks to the fal.ai async queue API. Once installed, Claude can generate and edit media on your behalf just by calling the skill from inside a normal conversation. You never leave the terminal. You never paste JSON payloads by hand. You describe what you want, Claude builds the request, fal.ai runs the model on a hosted GPU, and the result URL lands in your chat.
The skill exposes the full fal.ai catalog through one unified interface:
- Text to image: Flux Schnell, Flux Pro v1.1, Flux Pro v1.1 Ultra, Flux 2, Flux 2 Pro, HiDream, SDXL, Stable Diffusion 3.5 Large
- Image to image: Ghibli style, cartoon style, SVG vectorization, AI image editing via Flux 2 Pro Edit
- Text and image to video: Veo 2, Kling, Minimax Director, Hunyuan, Luma, Seedance
- Text to speech: ElevenLabs, Minimax TTS, Chatterbox
- Text to music: Minimax Music, Stable Audio
- Audio transcription: ElevenLabs Scribe
- Voice changing: ElevenLabs Voice Changer
- Video to audio: MMAudio soundscape generation
Every job is async: Claude submits it to fal's queue, polls until it completes, and returns the output URL (and an optional local download path) in the same conversation turn. All fifty-plus models are reachable through a single skill, so you do not need one tool per provider.
What You Will Need
- Claude Code. The CLI (or one of the IDE extensions for VS Code or JetBrains) installed and working. If you have never used Claude Code, start with our Claude Code in 10 minutes tutorial.
- Go 1.24 or newer. The skill ships a compiled Go CLI binary that Claude invokes under the hood. The install script needs
go on your PATH to build it.
- Git. To clone the repo if you take the manual path. The easy path does not require you to clone anything by hand, Claude handles it.
- A fal.ai API key. Grab one at fal.ai/dashboard/keys. The first job credits are free so you can verify the install works before adding any billing.
That is the entire shopping list. No Python environment, no Docker, no GPU on your machine. All the heavy lifting runs on fal.ai's side.
The Easy Install: Ask Claude Code To Do It
This is the install path the skill was designed for, and it takes one prompt. Open Claude Code in any project folder and paste the following:
Please install this Claude Code skill from https://github.com/karamble/claude-skill-falai and set it up for me.
Claude will do the full sequence itself:
- Fetch the repo metadata and read the README so it understands what the skill is.
- Clone
https://github.com/karamble/claude-skill-falai.git to a working directory.
- Run the bundled
./install.sh script that compiles the Go binary and copies the skill files into ~/.claude/skills/falai/.
- Verify that
~/.claude/skills/falai/SKILL.md exists and the binary is executable.
- Ask you for your fal.ai API key (you paste it, Claude stores it, nobody logs it).
- Write the config to
~/.config/falai/config.yaml with restrictive permissions (file mode 0600, directory mode 0700) so other users on the machine cannot read your key.
- Run a
ping action to confirm the key is valid and fal.ai is reachable.
- Report ready.
The whole flow takes about two minutes. You never have to touch a shell prompt, a YAML editor, or a Go toolchain. Claude handles every step and narrates what it is doing as it goes. If you hit a missing prerequisite (no Go installed, for example), Claude will flag it and tell you exactly what to fix.
The Manual Install: Three Commands
If you would rather see every step yourself, the manual path is three shell commands. This is also the path to take if you want to inspect the install script first or fork the repo before installing.
git clone https://github.com/karamble/claude-skill-falai.git
cd claude-skill-falai
./install.sh
The install.sh script runs go build to compile the CLI binary, copies the skill into ~/.claude/skills/falai/, and sets the right permissions. Rerun it any time you pull new changes from the repo.
Once installed, the first time you ask Claude Code to generate anything, Claude will detect that the config file is missing and walk you through the same API key setup flow described above.
Set Up Your fal.ai API Key
If you took the easy path, this step already happened. If you took the manual path, the first generation prompt triggers it. The key lives in a small YAML file:
api_key: "your-fal-ai-api-key"
timeout: 300
debug: false
The file path is ~/.config/falai/config.yaml. Only api_key is required. timeout (in seconds, default 300) caps how long the CLI waits for a queued job before giving up. debug: true makes the CLI log each HTTP request to stderr, useful for troubleshooting but noisy otherwise.
If you prefer to keep the key out of a file, set the environment variable FALAI_API_KEY in your shell profile. The CLI checks the environment first and falls back to the config file.
Your First Generation: Ten Prompts That Prove It Works
Once the skill is installed and configured, you drive it by describing what you want in plain English. Claude picks the right fal.ai model for the job and calls the skill. Here are ten prompts to try on a fresh install:
- Quick image: "Generate a hyperrealistic image of a red vintage car on a mountain road at sunset."
- High quality hero: "Use Flux Pro Ultra to generate a dark cyberpunk server room at 16:9 aspect ratio."
- Image editing: "Take this photo URL and replace the sky with a starry night using Flux 2 Pro Edit."
- Ghibli style: "Convert this photo into a Studio Ghibli style illustration."
- Text to video: "Create a 5-second video of a golden retriever running on a beach at sunset."
- Image to video: "Animate this still product photo with a slow zoom and gentle camera drift."
- Text to speech: "Read this paragraph aloud with a calm narrator voice using ElevenLabs."
- Music generation: "Generate a 20-second ambient soundtrack suitable for a documentary opening."
- Transcription: "Transcribe this audio file URL and return the text with timestamps."
- Voice changing: "Take this voice recording and change it to a different speaker using ElevenLabs Voice Changer."
For every prompt, Claude builds the right JSON payload, calls the skill, waits for the fal.ai queue to finish, and returns the result URL. If you pass download: true in the options, the file also lands locally in your working directory.
How the Skill Fits a Real Workflow
The real payoff is in how naturally the skill slots into work you already do in Claude Code. A few concrete examples:
Hero images for articles. Ask Claude to write a blog post, then ask it to generate a 16:9 hero image that matches the opening paragraph. Claude picks the model (Flux Pro Ultra is the current default for cinematic hero shots), generates the image, downloads it, and drops the local path into your draft. Our own site uses this workflow on every new article.
Product mockups from a logo. Point Claude at a logo PNG, ask for a set of product photos with the logo prominently featured. Claude uses Flux 2 Pro Edit with the logo as the image input and generates branded shots for landing pages, social posts, or press kits.
Video ads for marketing tests. Generate a short clip for an A/B test without opening any creative tool. "Create a 5-second ad showing a coffee mug steaming on a wooden table, warm morning light, shallow depth of field." Claude runs Veo 2 or Kling through the skill and returns a ready-to-upload MP4.
Voice over for a tutorial. Write a script, ask Claude to narrate it with ElevenLabs, and pipe the resulting audio into your video editor. No separate TTS tool, no extra account.
Transcripts of meeting recordings. Drop an audio URL into the chat and ask for a transcript. The skill calls ElevenLabs Scribe and returns the full text, ready for summarization by Claude in the same conversation.
Troubleshooting and Tips
Go 1.24 is a hard requirement. The install script calls go build. Older Go versions will fail with an obscure module error. Check your version with go version before running the install script.
First job can take extra time. The fal.ai queue has cold-start costs on some models. Image jobs usually complete in under 10 seconds, but the first video generation can take 1 to 3 minutes. The skill polls automatically, so Claude just waits.
Keep the API key out of git. The config file lives in ~/.config, not inside any project. Never paste your key into a README, a committed .env file, or a notebook you plan to publish. The skill ships a .gitignore that covers its own config, but external configs are on you.
Costs are per second of GPU time. Flux Schnell is pennies per image. Flux Pro Ultra is a few cents. Veo 2 video is the expensive one. Check the fal.ai pricing page before running 100 videos in a loop.
Debug mode is your friend. If a job fails and Claude cannot figure out why, set debug: true in the config and rerun. The CLI will print every HTTP request and response, and the failure root cause becomes obvious almost every time.
What to Read Next
For the full tour of the fal.ai platform itself, including the model catalog, pricing mechanics, SDKs, and a hands-on comparison with self-hosting, read our fal.ai tool spotlight. That article is the natural pair with this install tutorial: one covers the service, this one covers the skill that brings it into your Claude Code workflow.
If you want to see what else Claude Code can do once you start attaching skills to it, start with the Claude Code getting-started tutorial and the Automate MODX with Claude Code deep dive. Both articles show the same skill-installation pattern applied to very different domains.
Sources and Further Reading
You now have Claude Code talking to fal.ai through a single installed skill. Pick one of the ten prompts above, run it in your next Claude Code session, and watch a finished media file land in your working directory. From there, every prompt you write can pull from the full fal.ai catalog without opening another browser tab. The install is the hard part, and on this skill the install is a single sentence.
