Image Generation with fal.ai

Use this skill when asked to generate images, create visual assets, produce illustrations, or work with AI image generation.

MCP Tools Available

Two MCP servers are configured in .mcp.json:

fal-docs (Documentation)

Remote server at https://docs.fal.ai/mcp. Use to look up model parameters, API details, or troubleshoot.

fal-ai (Generation)

Local server via uvx fal-ai-mcp-server. Available tools:

Generating Images

Use the fal-ai MCP generate tool:

generate(
  model="fal-ai/gpt-image-1.5",
  prompt="A flat-style rocket icon, minimalist, clean lines",
  background="transparent",
  quality="high",
  output_format="png"
)

Images are saved to ./generated/ directory.

Preferred Models

When to Use Each

• •Icons/UI assets: GPT Image 1.5 (supports transparent background)
• •Quick iterations: Nano Banana Pro (fastest)
• •Editing existing images: Flux 2 (best I2I via /edit endpoint)
• •High resolution: Nano Banana Pro (supports 4K)

Common Parameters

GPT Image 1.5

{
  "prompt": "...",
  "image_size": "1024x1024",  // or 1536x1024, 1024x1536
  "background": "transparent", // or opaque, auto
  "quality": "high",          // or low, medium
  "num_images": 1,            // 1-4
  "output_format": "png"      // or jpeg, webp
}

Nano Banana Pro

{
  "prompt": "...",
  "aspect_ratio": "1:1",      // or 16:9, 9:16, 4:3, etc.
  "resolution": "1K",         // or 2K, 4K
  "num_images": 1,
  "output_format": "png"
}

Flux 2

{
  "prompt": "...",
  "guidance_scale": 2.5,      // 0-20
  "num_inference_steps": 28,  // 4-50
  "acceleration": "regular",  // or none, high
  "num_images": 1,
  "output_format": "png"
}

Prompt Engineering Tips

For Icons

"A flat-style [subject] icon, minimalist design, clean lines, solid colors"
"A 3D [subject] icon with soft shadows, rounded edges, gradient fill"
"A line-art [subject] icon, single stroke weight, geometric"

For Illustrations

"A modern illustration of [subject], flat design, limited color palette"
"An isometric view of [subject], clean vectors, soft shadows"

Style Modifiers

• •Flat: "flat design", "flat-style", "2D"
• •3D: "3D rendered", "soft shadows", "depth"
• •Minimal: "minimalist", "simple", "clean"
• •Detailed: "intricate", "detailed", "complex"

Background Control

• •Use background: "transparent" for icons/assets
• •Add "on a white background" or "on a [color] background" to prompt
• •Use background: "auto" to let model decide

Image-to-Image (Editing)

Use the /edit endpoint variants:

generate(
  model="fal-ai/gpt-image-1.5/edit",
  prompt="Change the color to blue",
  image_urls=["https://..."],
  input_fidelity="high"
)

Or with Flux 2:

generate(
  model="fal-ai/flux-2/edit",
  prompt="Add a shadow beneath",
  image_urls=["https://..."]
)

Output Location & Naming

Generated images are saved to ./generated/ (configured in .mcp.json).

Naming convention:

{subject}-{style}-{size}.png
# Examples:
rocket-flat-1024.png
calendar-3d-512.png
star-line-256.png

For production assets, move to:

./public/assets/

Fallback: CLI Usage

If MCP is unavailable, use the fal CLI:

# Install fal CLI
npm install -g @fal-ai/cli

# Generate (requires FAL_KEY env var)
fal run fal-ai/gpt-image-1.5 \
  --prompt "A flat-style rocket icon" \
  --background transparent \
  --quality high \
  --output_format png

Project Integration

This project uses fal.ai through:

• •MCP servers: Configured in .mcp.json
• •Web proxy: /api/fal/proxy (Next.js route)
• •Client: @fal-ai/client in apps/web/lib/fal.ts
• •Models: Defined in apps/web/lib/models/

When generating images for this project's icon sets, prefer:

1. •GPT Image 1.5 for transparent icons
2. •1024x1024 size for consistency
3. •PNG format for transparency support