picasso（picasso 搜索）

Picasso

Generate production-ready visual assets via fal.ai API.

Design Philosophy

Commit to a BOLD aesthetic direction—no generic AI slop:

• Choose distinctive, memorable visual concepts over safe defaults
• Use intentional, purposeful compositions that feel genuinely designed
• Avoid overused patterns: gradient text, neon-on-dark, glassmorphism clichés

Prerequisites

command -v bun ffmpeg convert 2>/dev/null | xargs -I{} basename {} || echo "Missing deps"

Required: bun, FAL_API_KEY in .env. Optional: ffmpeg (video/gif), convert (ImageMagick).

Quick Start

bun scripts/fal-generate.ts "minimalist letter A logo, flat design" --size square --out favicon.png
bun scripts/fal-generate.ts "mobile app icon, weather app" --num 4
bun scripts/fal-generate.ts "pixel art sword, transparent background" --format png

Script Usage

Usage: bun fal-generate.ts <prompt> [options]

Options:
  --model <id>        fal.ai model endpoint (see Model Selection below)
  --size <size>       Size: square, landscape_4_3, portrait_4_3, etc.
  --num <n>           Number of images 1-4
  --format <fmt>      png, jpeg, webp
  --out <path>        Save first image to file
  --edit <url>        Image URL for editing (can be repeated)
  --resolution <res>  Edit output resolution: 1K, 2K, 4K

Model Selection & Cost Efficiency

→ *Consult models reference for discovery commands and selection criteria.*

Strategy: Cheap iterations, quality finals

1. Explore with fast/cheap models — most generations get discarded
2. Upgrade for final output only — when quality is the deliverable
3. Post-process instead of regenerate — ImageMagick is free

# Discover available models
curl -s "https://api.fal.ai/v1/models?category=text-to-image&status=active" \
  -H "Authorization: Key $FAL_API_KEY" | jq '.models[] | {id: .endpoint_id, name: .metadata.display_name}'

# Search by capability
curl -s "https://api.fal.ai/v1/models?q=fast" -H "Authorization: Key $FAL_API_KEY"

When to use what

Browse models: https://fal.ai/models

Prompting Fundamentals

→ *Consult image generation reference for complete prompt engineering guide.*

Prompt structure: Subject + Style/Medium + Lighting + Composition + Color/Mood

# Well-structured prompt
bun scripts/fal-generate.ts "portrait of a barista, film photo, soft rim light, 50mm close-up, warm mood, teal-orange palette"

DO: Be specific about subject, lighting, camera, mood DON'T: Use vague terms like "cool" or "nice"

Image Editing

→ *Consult image editing reference for inpainting, outpainting, and compositing.*

Edit existing images with --edit. The script auto-selects appropriate edit endpoint.

# Inpainting (edit within image)
bun scripts/fal-generate.ts "replace the red car with a blue motorcycle" --edit photo.png

# Outpainting (extend image)
bun scripts/fal-generate.ts "extend to landscape, continue the forest background" --edit portrait.png

# Compositing (combine images)
bun scripts/fal-generate.ts "put the person from first image into the scene from second" \
  --edit person.png --edit background.png

Edit tips:

• Be specific about what to change AND what to preserve
• Describe spatial relationships: "add a cat sitting on the left"
• Reference image order matters: first = subject, second = context/style

Mockups

→ *Consult mockups reference for device, product, and presentation mockups.*

Generate realistic product presentations:

# Device mockup
bun scripts/fal-generate.ts "iPhone on wooden desk, app interface on screen, coffee shop setting, shallow depth of field"

# Product mockup
bun scripts/fal-generate.ts "t-shirt on hanger, front view, neutral background, soft studio lighting"

# Two-step for exact screen content
bun scripts/fal-generate.ts "phone mockup, blank screen, lifestyle setting" --out base.png
bun scripts/fal-generate.ts "place this screenshot on the phone screen" --edit base.png --edit my_screenshot.png

Asset Type Quick Reference

Detailed Guides

Core:

• Image Generation - Prompt structure, lighting, composition, style
• Image Editing - Inpainting, outpainting, compositing
• Models - Model discovery, selection criteria, cost/quality tradeoffs

Asset-Specific:

• Game Assets - Spritesheets, pixel art, tilesets, character rotations
• Icons & Favicons - Size requirements, platform guidelines
• Brand Assets - Logos, marketing imagery, style consistency
• Mockups - Device, product, and presentation mockups

Post-Processing (ImageMagick)

# Resize for favicon
convert logo.png -resize 32x32 favicon.ico
convert logo.png -resize 180x180 apple-touch-icon.png

# Remove background
convert logo.png -fuzz 10% -transparent white transparent.png

# Create spritesheet from frames
montage frame*.png -tile 8x1 -geometry +0+0 -background none spritesheet.png

# Nearest-neighbor resize (pixel art)
convert sprite.png -filter point -resize 64x64 sprite_2x.png

# Batch uniform sizing
for f in frame*.png; do convert "$f" -gravity center -extent 64x64 "fixed_$f"; done

Post-Processing (ffmpeg)

# PNG sequence to GIF
ffmpeg -framerate 10 -i frame_%03d.png -loop 0 animation.gif

# Video to GIF
ffmpeg -i input.mp4 -vf "fps=10,scale=320:-1" output.gif

Example Workflow

# 1. Discover suitable models
curl -s "https://api.fal.ai/v1/models?q=logo&category=text-to-image" \
  -H "Authorization: Key $FAL_API_KEY" | jq '.models[0:3] | .[].endpoint_id'

# 2. Generate logo concepts (use fast model for exploration)
bun scripts/fal-generate.ts "modern tech startup logo, geometric, blue" --num 4

# 3. Refine chosen concept
bun scripts/fal-generate.ts "minimalist geometric logo, blue" --size square --out logo.png

# 4. Create favicon set
convert logo.png -resize 32x32 favicon.ico
convert logo.png -resize 180x180 apple-touch-icon.png