---

name: fal-ai-image description: Generate images using fal.ai nano-banana-pro model

fal-ai-image

Generate images via fal.ai nano-banana-pro (Gemini 3 Pro Image). Best for: infographics, text rendering, complex compositions.

Config

Requires FAL_KEY in config/.env or environment. Get key: https://fal.ai/dashboard/keys

Two Modes

1. Generate (text-to-image)

Create images from text prompt only. Script: scripts/generate.sh

2. Edit (image-to-image)

Create images using reference images (up to 14). Script: scripts/edit.sh

Workflow

IMPORTANT: Run generation via Task tool with Haiku subagent to avoid blocking main context.

For Generate mode:

1. Clarify params (if not specified):

   • Aspect ratio: 1:1, 16:9, 9:16, 4:3, etc.
   • Resolution: 1K (default), 2K, 4K

2. Propose save path based on project structure:

   • Check for ./images/, ./assets/, ./static/
   • Suggest: "Save to ./images/infographic_coffee.png?"

3. Show price & confirm:

   CopyCost: $0.15/image (4K: $0.30)
   Confirm? (yes/no)
   

4. Launch subagent:

   CopyTask tool:
   - subagent_type: "Bash"
   - model: "haiku"
   - run_in_background: true
   - prompt: Run generate.sh with params
   

5. Report result: Parse JSON output, show image URL, read saved file.

For Edit mode:

1. Get reference images:

   • If URLs provided → use directly
   • If local files → run upload.sh first to get URLs

2. Clarify prompt: What to do with references?

3. Show price & confirm: Same as generate

4. Launch subagent with edit.sh

5. Report result

Scripts

generate.sh

Copybash scripts/generate.sh \
  --prompt "infographic about coffee brewing" \
  --aspect-ratio "9:16" \
  --resolution "1K" \
  --output-dir "./images" \
  --filename "coffee_infographic"

edit.sh

Copybash scripts/edit.sh \
  --prompt "combine these into a collage" \
  --image-urls "https://example.com/img1.png,https://example.com/img2.png" \
  --aspect-ratio "16:9" \
  --output-dir "./images" \
  --filename "collage"

upload.sh (for local files)

Copy# Get URL for local file
URL=$(bash scripts/upload.sh --file /path/to/image.png)

# Or get base64 data URI (for small files)
URI=$(bash scripts/upload.sh --file /path/to/image.png --base64)

Parallel Generation

For multiple images — launch several subagents in parallel:

CopyTask 1: generate "cat in space" → cat_space.png
Task 2: generate "dog on moon" → dog_moon.png
Task 3: generate "bird in ocean" → bird_ocean.png

Each runs independently via Haiku, results collected when done.

Pricing

• Generate: $0.15/image
• Edit: $0.15/edit
• 4K resolution: $0.30 (2x)
• Web search: +$0.015

Formula: price = num_images * (resolution == "4K" ? 0.30 : 0.15) + (web_search ? 0.015 : 0)

Notes

• URLs expire in ~1 hour — save locally if needed
• Uploaded files stored 7 days on fal.ai, then auto-deleted
• Model excels at text rendering and infographics