fal.ai Model API Skill

Run 1000+ generative AI models on fal.ai.

Arguments

• Command: $0 (search | schema | run | status | result | upload)
• Arg 1: $1 (model_id, search query, or file path)
• Arg 2+: $2, $3, etc. (additional parameters)
• All args: $ARGUMENTS

Session Output

Save generated files to session folder:

mkdir -p ~/.fal/sessions/${CLAUDE_SESSION_ID}

Downloaded images/videos go to: ~/.fal/sessions/${CLAUDE_SESSION_ID}/

Authentication

Requires FAL_KEY environment variable. If requests fail with 401, tell user:

Get an API key from https://fal.ai/dashboard/keys
Then: export FAL_KEY="your-key-here"

Command: $0

If $0 = "search"

Search for models matching $1:

curl -s "https://api.fal.ai/v1/models?q=$1&limit=15" \
  -H "Authorization: Key $FAL_KEY" | jq -r '.models[] | "• \(.endpoint_id) — \(.metadata.display_name) [\(.metadata.category)]"'

For category search, use:

curl -s "https://api.fal.ai/v1/models?category=$1&limit=15" \
  -H "Authorization: Key $FAL_KEY" | jq -r '.models[] | "• \(.endpoint_id) — \(.metadata.display_name)"'

Categories: text-to-image, image-to-video, text-to-video, image-to-3d, training, speech-to-text, text-to-speech

If $0 = "schema"

Get input schema for model $1:

curl -s "https://api.fal.ai/v1/models?endpoint_id=$1&expand=openapi-3.0" \
  -H "Authorization: Key $FAL_KEY" | jq '.models[0].openapi.components.schemas.Input.properties'

Show required vs optional fields to help user understand what inputs are needed.

If $0 = "run"

Run model $1 with parameters from remaining arguments.

Step 1: Parse parameters

Extract --key value pairs from $ARGUMENTS after the model_id to build JSON payload.

Example: /fal run fal-ai/flux-2 --prompt "a cat" --image_size landscape_16_9

→ Model: fal-ai/flux-2

→ Payload: {"prompt": "a cat", "image_size": "landscape_16_9"}

Step 2: Submit to queue

curl -s -X POST "https://queue.fal.run/$1" \
  -H "Authorization: Key $FAL_KEY" \
  -H "Content-Type: application/json" \
  -d '<JSON_PAYLOAD>'

Step 3: Poll until complete

# Get request_id from response, then poll:
while true; do
  STATUS=$(curl -s "https://queue.fal.run/$1/requests/$REQUEST_ID/status" \
    -H "Authorization: Key $FAL_KEY" | jq -r '.status')
  echo "Status: $STATUS"
  if [ "$STATUS" = "COMPLETED" ]; then break; fi
  if [ "$STATUS" = "FAILED" ]; then echo "Job failed"; break; fi
  sleep 3
done

Step 4: Get result and save

# Fetch result
RESULT=$(curl -s "https://queue.fal.run/$1/requests/$REQUEST_ID" \
  -H "Authorization: Key $FAL_KEY")

# Create session output folder
mkdir -p ~/.fal/sessions/${CLAUDE_SESSION_ID}

# Download images/videos
# For images: jq -r '.images[0].url' and curl to download
# Save as: ~/.fal/sessions/${CLAUDE_SESSION_ID}/<timestamp>_<model>.png

If $0 = "status"

Check status of request $2 for model $1:

curl -s "https://queue.fal.run/$1/requests/$2/status?logs=1" \
  -H "Authorization: Key $FAL_KEY" | jq '{status: .status, queue_position: .queue_position, logs: .logs}'

If $0 = "result"

Get result of completed request $2 for model $1:

curl -s "https://queue.fal.run/$1/requests/$2" \
  -H "Authorization: Key $FAL_KEY" | jq '.'

If $0 = "upload"

Upload file $1 to fal CDN:

curl -s -X POST "https://fal.run/fal-ai/storage/upload" \
  -H "Authorization: Key $FAL_KEY" \
  -F "file=@$1"

Returns URL to use in model requests.

Quick Reference

Popular models:

• fal-ai/flux-2 — Fast text-to-image
• fal-ai/flux-2-pro — High quality text-to-image
• fal-ai/kling-video/v2/image-to-video — Image to video
• fal-ai/minimax/video-01/image-to-video — Image to video
• fal-ai/whisper — Speech to text

Common parameters for text-to-image:

• --prompt "description" — What to generate
• --image_size landscape_16_9 — Aspect ratio (square, portrait_4_3, landscape_16_9)
• --num_images 1 — Number of images

Example invocations:

• /fal search video — Find video models
• /fal schema fal-ai/flux-2 — See input options
• /fal run fal-ai/flux-2 --prompt "a sunset over mountains"
• /fal status fal-ai/flux-2 abc-123
• /fal upload ./photo.png