# fal.ai through toolfare

> Run a specific image or video generation model hosted on fal — by its model id, with that model’s own inputs — and read back runs that are still in progress, answered inside ChatGPT, Claude, Codex or Cursor. No fal account or key of your own: toolfare runs each generation and tells your AI the price before it starts.

Page: https://toolfare.ai/fal

## Use it from your agent

Paste this into Claude Code, Codex, Cursor or any agent with a shell:

> Set up https://toolfare.ai/skill.md, then use toolfare to answer: Generate a product shot with this image model on fal.

Apps without a terminal connect toolfare as an MCP server: https://toolfare.ai/docs#connect

## Use it from a script

```sh
mkdir -p ~/.local/bin
curl -fsSL 'https://toolfare.ai/toolfare/toolfare.mjs' -o ~/.local/bin/toolfare
chmod +x ~/.local/bin/toolfare
export PATH="$HOME/.local/bin:$PATH"
export TOOLFARE_MCP_URL='https://api.toolfare.ai/mcp'
toolfare login
```

Sign in once per machine; the CLI renews the sign-in itself. On a CI runner or another server, create an API key in the toolfare console and set it as `TOOLFARE_API_KEY` instead of signing in. Paid calls on one sign-in or key run one at a time. `--max-cost` refuses a call quoted above that many credits (1,000 credits = $3).

## Tools

### Run a named generation model

Tool: `fal_run_model`

Invokes one model on fal by its id, with that model’s own input schema, and returns what it produced — or a queue position when it is still running.

Ask:
- Generate a product shot with this image model on fal.
- Make a five-second video from this image with a fal video model.

### Collect a run that was still going

Tool: `fal_get_queue_result`

Fetches the result of a run that returned a queue position instead of an image or video.

Ask:
- Is that video generation finished yet?
- Get the result of the run I started earlier.

### Also available

- `fal_get_queue_status`: Check whether a queued fal run has finished

## Price

Quoted to your AI before the run starts, then charged on the result fal returns for that exact generation. Metered on the finished output, wherever it lands — immediately or on a later read-back — and charged once. A video result settles as a fixed fifteen-second clip whatever the clip’s own length, which makes video among the most expensive calls in the catalogue. No fixed price table on the page, on purpose: every model prices differently, and the rates live in the provider’s billing module. A run that fails is never charged.

Balances and top-ups are in US dollars. Quotes and `--max-cost` count credits: 1,000 credits = $3 ($0.003 each).

## Use it when

- You need an image or a video from a specific named model.
- You are testing creative variants across models and want to choose the model yourself.

## Look elsewhere when

- You just want an image and do not mind which model makes it.
- You do not know the model id. The address is the model, so there is nothing sensible to run without one.
- You want a finished ad video from a product page, with a script and presenter.

## Limits

- You name the model and supply its inputs; a wrong id or input is rejected by fal.
- Video generation is slow and costly — a run may come back as a queue position to collect later, and a finished video is charged as a fifteen-second clip even when it is shorter.
- Pay as you go from your toolfare balance, in US dollars; there is no subscription. A call runs when your balance covers its quote.