Fal Model Guide

Quick Reference

When to Use This Skill

Use for model selection decisions:

• Comparing FLUX vs SDXL for image generation
• Choosing video models by quality tier
• Optimizing costs with appropriate model selection
• Understanding model-specific parameters
• Building production vs development pipelines

Related skills:

• For text-to-image: see fal-text-to-image
• For text-to-video: see fal-text-to-video
• For audio models: see fal-audio

fal.ai Model Selection Guide

Comprehensive guide to selecting the right fal.ai model for your use case.

Image Generation Models

FLUX Family

FLUX.2 Pro - Latest and best quality

const result = await fal.subscribe("fal-ai/flux-2-pro", {
  input: {
    prompt: "Professional product photo of a watch",
    image_size: "square_hd",
    num_inference_steps: 28,
    guidance_scale: 3.5
  }
});

FLUX.1 Dev - Best open-source option

const result = await fal.subscribe("fal-ai/flux/dev", {
  input: {
    prompt: "A serene mountain landscape",
    image_size: "landscape_16_9",
    num_inference_steps: 28
  }
});

FLUX Schnell - Fast 4-step generation

const result = await fal.subscribe("fal-ai/flux/schnell", {
  input: {
    prompt: "Quick concept sketch",
    num_inference_steps: 4  // Optimized for 4 steps
  }
});

Stable Diffusion Models

// Fast SDXL - Good balance of speed and quality
const result = await fal.subscribe("fal-ai/fast-sdxl", {
  input: {
    prompt: "A colorful abstract painting",
    image_size: "square_hd",
    num_inference_steps: 25
  }
});

Specialized Image Models

Image-to-Image Models

// Image-to-Image transformation
const result = await fal.subscribe("fal-ai/flux/dev/image-to-image", {
  input: {
    image_url: "https://example.com/photo.jpg",
    prompt: "Transform into watercolor painting style",
    strength: 0.75  // 0-1, how much to change
  }
});

// Inpainting (edit specific regions)
const result = await fal.subscribe("fal-ai/flux/dev/inpainting", {
  input: {
    image_url: "https://example.com/photo.jpg",
    mask_url: "https://example.com/mask.png",
    prompt: "A golden retriever"
  }
});

// ControlNet (structural guidance)
const result = await fal.subscribe("fal-ai/flux/dev/controlnet", {
  input: {
    prompt: "Modern house design",
    control_image_url: "https://example.com/edges.png",
    controlnet_conditioning_scale: 0.8
  }
});

Video Generation Models

Text-to-Video

Kling 2.6 Pro - Best overall quality

const result = await fal.subscribe("fal-ai/kling-video/v2.6/pro", {
  input: {
    prompt: "A majestic eagle soaring over mountains at golden hour",
    duration: 5,
    aspect_ratio: "16:9",
    negative_prompt: "blurry, distorted",
    cfg_scale: 0.5
  }
});

LTX-2 Pro - Fast with audio

const result = await fal.subscribe("fal-ai/ltx-video-2-pro", {
  input: {
    prompt: "Ocean waves crashing on rocks",
    resolution: "720p",
    enable_audio: true
  }
});

Image-to-Video

// Animate a still image
const result = await fal.subscribe("fal-ai/minimax/video-01", {
  input: {
    image_url: "https://example.com/portrait.jpg",
    prompt: "Person slowly turns head and smiles",
    prompt_optimizer: true
  }
});

Video-to-Video (Editing)

// Edit/transform existing video
const result = await fal.subscribe("fal-ai/kling-video/o1", {
  input: {
    video_url: "https://example.com/video.mp4",
    prompt: "Change to anime style"
  }
});

Audio Models

Speech-to-Text

const result = await fal.subscribe("fal-ai/whisper", {
  input: {
    audio_url: "https://example.com/speech.mp3",
    task: "transcribe",  // or "translate"
    language: "en",
    chunk_level: "segment"
  }
});

console.log(result.text);
console.log(result.chunks);  // With timestamps

Text-to-Speech

// Text-to-speech with Elevenlabs
const result = await fal.subscribe("fal-ai/elevenlabs", {
  input: {
    text: "Hello, welcome to our service.",
    voice_id: "voice_id_here"
  }
});

3D Generation Models

// Generate 3D mesh from image
const result = await fal.subscribe("fal-ai/triposr", {
  input: {
    image_url: "https://example.com/object.jpg"
  }
});

console.log(result.model_mesh.url);  // GLB/OBJ file

Model Selection Decision Tree

What do you want to create?
├── Image
│   ├── From text only?
│   │   ├── Need best quality? → FLUX.2 Pro
│   │   ├── Need fast iteration? → FLUX Schnell
│   │   ├── Need open-source? → FLUX.1 Dev
│   │   └── Budget conscious? → Fast SDXL
│   ├── Transform existing image?
│   │   ├── Style transfer → FLUX i2i
│   │   ├── Edit specific region → FLUX Inpainting
│   │   └── Follow structure → FLUX ControlNet
│   └── Need text in image? → Ideogram
│
├── Video
│   ├── From text only?
│   │   ├── Need best quality? → Kling 2.6 Pro
│   │   ├── Need fast preview? → Runway Gen-3 Turbo
│   │   └── Need audio? → LTX-2 Pro or Kling
│   ├── Animate image? → MiniMax Hailuo
│   └── Edit video? → Kling O1
│
├── Audio
│   ├── Speech to text?
│   │   ├── Accurate → Whisper
│   │   └── Fast → Whisper Turbo
│   └── Text to speech? → Elevenlabs / F5-TTS
│
└── 3D
    └── Image to 3D? → TripoSR

Performance Comparison

Image Generation Speed

Video Generation Speed

Cost Optimization Strategies

1. Development Phase

   • Use FLUX Schnell for prompt iteration
   • Use Fast SDXL for quick tests
   • Use Runway Gen-3 Turbo for video previews

2. Production Phase

   • FLUX.2 Pro for final images
   • Kling 2.6 Pro for final videos

3. Batch Processing

   • Process during off-peak hours
   • Use appropriate image sizes (don't upscale unnecessarily)
   • Cache results by seed for reproducibility

4. Resource Efficiency

   • Generate at target resolution (don't generate larger and downscale)
   • Use webhooks for high-volume (avoid polling overhead)
   • Implement client-side caching

Model-Specific Tips

FLUX Tips

• Guidance scale 3-4 works best
• 28 steps is optimal default
• Schnell works best with 4 steps only
• Detailed prompts produce better results

Video Tips

• Include camera movement in prompts
• Describe action/motion explicitly
• Use negative prompts to avoid artifacts
• Start with shorter duration for testing

Whisper Tips

• Provide language hint for better accuracy
• Use "translate" task for non-English to English
• chunk_level="segment" gives timestamps

Explore More Models

Visit https://fal.ai/models for the complete catalog of 600+ models including:

• Face restoration/enhancement
• Background removal
• Upscaling
• Style transfer
• OCR
• Object detection
• And many more specialized models