fal-generate Skill

Overview

This skill enables AI content generation through fal.ai's latest models using a queue-based system. It supports:

• Text-to-Image - Generate images from text prompts
• Image-to-Image - Edit and transform existing images
• Text-to-Video - Create videos from text descriptions
• Image-to-Video - Animate images into videos
• Text-to-Speech - Generate natural speech from text
• Speech-to-Text - Transcribe audio to text
• Text-to-3D - Create 3D models from text
• Image-to-3D - Convert images to 3D models
• LLM / VLM / ALM - Run any LLM, vision, audio or video model via OpenRouter

Scripts

Prerequisites

export FAL_KEY="your-api-key"

Output Format

All generation scripts output JSON to stdout when using --wait. The JSON contains URLs to the generated content:

• Images: {"images": [{"url": "https://fal.media/files/...", "width": 1024, "height": 1024}]}
• Videos: {"video": {"url": "https://fal.media/files/...mp4"}}
• Audio/TTS: {"audio": {"url": "https://fal.media/files/...mp3"}} or {"audio_url": "https://..."}
• 3D Models: {"model_mesh": {"url": "https://fal.media/files/...glb"}}
• Transcription: {"text": "transcribed content..."}
• OpenRouter: {"output": "LLM response text..."}

Without --wait, prints the request ID. With --async, prints only the request ID for later polling.

Examples by Category

Text-to-Image

# Basic image generation
./scripts/generate.sh -m fal-ai/kling-image/v3/text-to-image \
  -p "A majestic mountain at sunrise, cinematic lighting" -w

# With aspect ratio and seed
./scripts/generate.sh -m fal-ai/flux-2/klein/9b \
  -p "Professional headshot, studio lighting" \
  --aspect-ratio "1:1" --seed 42 -w

# Ultra-fast generation
./scripts/generate.sh -m fal-ai/z-image/turbo \
  -p "Quick concept sketch of a robot" -w

# With custom parameters (inference steps, guidance scale)
./scripts/generate.sh -m fal-ai/flux-2/klein/9b \
  -p "Detailed portrait of a scientist" \
  --param num_inference_steps=28 --param guidance_scale=3.5 -w

Image-to-Image (Edit/Transform)

# Upload local image first
IMAGE_URL=$(./scripts/upload.sh ~/photos/portrait.jpg)

# Edit with instructions
./scripts/generate.sh -m fal-ai/qwen-image-max/edit \
  --image-url "$IMAGE_URL" \
  -p "Make the background a sunset beach" -w

# Style transfer
./scripts/generate.sh -m fal-ai/glm-image/image-to-image \
  --image-url "$IMAGE_URL" \
  -p "Convert to oil painting style" -w

Text-to-Video

# Cinematic video with audio (Kling V3 Pro)
./scripts/generate.sh -m fal-ai/kling-video/v3/pro/text-to-video \
  -p "A butterfly emerging from a cocoon in slow motion, macro lens" \
  --duration 5 -w

# Fast video generation
./scripts/generate.sh -m fal-ai/ltx-2-19b/distilled/text-to-video \
  -p "Drone shot flying over a city at golden hour" -w

# Google Veo 3.1 with sound
./scripts/generate.sh -m fal-ai/veo3.1 \
  -p "A cat playing piano, realistic" -w

Image-to-Video (Animate Images)

IMAGE_URL=$(./scripts/upload.sh ~/photos/landscape.jpg)

# Animate a still photo
./scripts/generate.sh -m fal-ai/kling-video/o3/pro/image-to-video \
  --image-url "$IMAGE_URL" \
  -p "Gentle wind moving through the trees, clouds drifting" -w

# Lip-sync avatar from image + audio
AUDIO_URL=$(./scripts/upload.sh ~/audio/speech.mp3)
./scripts/generate.sh -m fal-ai/longcat-multi-avatar/image-audio-to-video \
  --image-url "$IMAGE_URL" --audio-url "$AUDIO_URL" -w

Text-to-Speech

# High-quality TTS (MiniMax)
./scripts/generate.sh -m fal-ai/minimax/speech-2.8-hd \
  -t "Hello! Welcome to the future of AI-generated content." -w

# Fast TTS
./scripts/generate.sh -m fal-ai/minimax/speech-2.8-turbo \
  -t "This is a quick test of fast speech generation." -w

# Custom voice with Qwen-3 TTS
./scripts/generate.sh -m fal-ai/qwen-3-tts/text-to-speech/1.7b \
  -t "Custom voice synthesis with natural intonation." -w

Voice Cloning

# Upload a voice sample (10+ seconds recommended)
VOICE_URL=$(./scripts/upload.sh ~/audio/voice-sample.wav)

# Clone and generate speech
./scripts/generate.sh -m fal-ai/qwen-3-tts/clone-voice/1.7b \
  --audio-url "$VOICE_URL" \
  -t "This sentence will be spoken in the cloned voice." -w

Speech-to-Text (Transcription)

AUDIO_URL=$(./scripts/upload.sh ~/recordings/meeting.mp3)

# Fast transcription
./scripts/generate.sh -m fal-ai/nemotron/asr \
  --audio-url "$AUDIO_URL" -w

# Accurate transcription with timestamps
./scripts/generate.sh -m fal-ai/elevenlabs/speech-to-text/scribe-v2 \
  --audio-url "$AUDIO_URL" -w

Text-to-3D

# Detailed 3D model from text
./scripts/generate.sh -m fal-ai/hunyuan-3d/v3.1/pro/text-to-3d \
  -p "A detailed medieval sword with ornate handle" -w

# Fast 3D generation
./scripts/generate.sh -m fal-ai/hunyuan-3d/v3.1/rapid/text-to-3d \
  -p "Simple wooden chair" -w

Image-to-3D

IMAGE_URL=$(./scripts/upload.sh ~/photos/object.jpg)

# Convert image to 3D model
./scripts/generate.sh -m fal-ai/hunyuan-3d/v3.1/rapid/image-to-3d \
  --image-url "$IMAGE_URL" -w

# High-fidelity geometry
./scripts/generate.sh -m fal-ai/ultrashape \
  --image-url "$IMAGE_URL" -w

OpenRouter — Run Any LLM

# Text chat with any LLM (GPT-5, Claude, Gemini, Llama 4, etc.)
./scripts/generate.sh -m openrouter/router \
  -p "Explain quantum computing in simple terms" \
  --param model=google/gemini-2.5-flash -w

# Vision — analyze an image
IMAGE_URL=$(./scripts/upload.sh ~/photos/chart.png)
./scripts/generate.sh -m openrouter/router/vision \
  --image-url "$IMAGE_URL" \
  -p "Describe what you see in this image" \
  --param model=google/gemini-2.5-flash -w

# Audio — process audio with an ALM
AUDIO_URL=$(./scripts/upload.sh ~/audio/podcast.mp3)
./scripts/generate.sh -m openrouter/router/audio \
  --audio-url "$AUDIO_URL" \
  -p "Summarize the key points discussed" \
  --param model=google/gemini-2.5-flash -w

# Video — analyze a video
VIDEO_URL=$(./scripts/upload.sh ~/videos/demo.mp4)
./scripts/generate.sh -m openrouter/router/video \
  --video-url "$VIDEO_URL" \
  -p "Describe what happens in this video" \
  --param model=google/gemini-2.5-flash -w

Usage Patterns

Queue Mode (Default) — submit and poll

./scripts/generate.sh -m fal-ai/flux-2/klein/9b -p "Portrait" --wait

Async Mode — get ID, poll later

REQUEST_ID=$(./scripts/generate.sh -m fal-ai/kling-video/v3/pro/text-to-video \
  -p "Drone flying over a city" --async)
./scripts/poll.sh fal-ai/kling-video/v3/pro/text-to-video $REQUEST_ID

File Upload — local files to fal CDN

IMAGE_URL=$(./scripts/upload.sh ~/photos/portrait.jpg)
./scripts/generate.sh -m fal-ai/kling-video/o3/pro/image-to-video \
  --image-url "$IMAGE_URL" -p "Gentle wind blowing through hair" -w

Common Parameters

Environment Variables

Tips

1. Always use --wait or --async — Without either, you get the request ID + a manual curl command
2. Use --param for advanced control — Pass any model-specific parameter: --param guidance_scale=7.5
3. Check model schema — ./scripts/models.sh --schema <endpoint> to see all available params
4. Upload files first — Use ./scripts/upload.sh for local images/audio/video before generation
5. Use seeds — Same seed = same output for reproducible results
6. Pro vs Standard — Pro = better quality + longer generation; Standard = cost-effective
7. Flash/Turbo/Distilled — Best for previews and fast iterations

Model Catalog

Image Generation — February 2026

Image Generation — January 2026

Image Generation — December 2025

Video Generation — February 2026

Video Generation — January 2026

Video Generation — December 2025

Audio — February 2026

Audio — January 2026

Audio — December 2025

3D Generation — February 2026

3D Generation — December 2025

OpenRouter Endpoints

Access 100+ LLMs via OpenRouter. Use --param model=<provider/model> to select the model.

Text (LLM)

Vision (VLM)

Audio (ALM)

Video (VLM)

OpenAI-Compatible

Model Selection Guide