image-gen

name	image-gen
description	Generate images using FAL (Flux Pro 1.1 Ultra), Gemini (nano-banana 2), or OpenAI. Supports generation, editing, and refinement workflows. Use for any image creation task.

Image Generation Skill

Multi-provider image generation with a preference hierarchy:

1. FAL Flux Pro 1.1 Ultra — Primary generator (best quality, 9:16 native)
2. Gemini nano-banana 2 — Editing/refinement (fix hands, details, consistency)
3. OpenAI DALL-E 3 — Fallback generator

Prerequisites

Provider	Env Var	Status
FAL	FAL_KEY	Required for Flux
Gemini	GEMINI_API_KEY	Required for editing
OpenAI	OPENAI_API_KEY	Optional fallback

Quick Usage

Generate with Flux (9:16 portrait)

node skills/image-gen/scripts/fal-generate.js \
  --prompt "A solarpunk garden city at golden hour" \
  --aspect "9:16" \
  --output ./images/garden.png

Edit with Gemini (nano-banana)

node skills/image-gen/scripts/gemini-edit.js \
  --image ./images/garden.png \
  --instruction "Fix the hands to have exactly 5 fingers" \
  --output ./images/garden-fixed.png

Full Pipeline (generate → refine)

node skills/image-gen/scripts/pipeline.js \
  --prompt "A solarpunk garden city at golden hour" \
  --aspect "9:16" \
  --refine "Ensure all human figures have correct anatomy" \
  --output ./images/final.png

Workflow: Lofn Image Pipeline

1. Concept Phase (skills/image/00-03): Generate aesthetics, concepts, artist/critique
2. Prompt Synthesis (skills/image/04-10): Build final prompt
3. Generation (this skill): FAL Flux Pro 1.1 Ultra @ 9:16
4. Refinement (this skill): Gemini nano-banana for fixes
5. Output: Final image saved to workspace

FAL Flux Pro 1.1 Ultra

Endpoint: fal-ai/flux-pro/v1.1-ultra

Key Parameters:

• prompt: The generation prompt
• aspect_ratio: 21:9, 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16, 9:21
• output_format: jpeg, png
• safety_tolerance: 1 (strict) to 6 (permissive)
• raw: true for less processed, natural look
• seed: for reproducibility

Cost: ~$0.06 per image

Gemini nano-banana 2 (Image Editing)

Uses Gemini's native image generation/editing capabilities.

Models:

• gemini-2.0-flash-exp — Fast editing
• gemini-2.0-pro — Higher quality edits

Use Cases:

• Fix anatomical issues (hands, fingers, faces)
• Adjust clothing/accessories
• Color correction
• Add/remove elements
• Style transfer

Error Handling

• FAL timeout: Retry with exponential backoff (max 3 attempts)
• NSFW block: Adjust safety_tolerance or rephrase prompt
• Gemini edit fail: Fallback to regeneration with refined prompt

Output Structure

All generated images saved with metadata:

images/
  {timestamp}_{seed}.png
  {timestamp}_{seed}.json  # metadata: prompt, params, provider

Integration with Lofn Pipeline

This skill is called by skills/image/08_Generate_Image_Generation.md after prompt synthesis completes. It can also be invoked directly for quick generations.