image-generation

name	image-generation
description	Generate and edit images using OpenRouter + Gemini. Use when Daniel asks to create, generate, or edit images. Can also inject generated images into the Draw/whiteboard canvas.
triggers	imagen, image, genera imagen, create image, generate image, foto, picture, edita imagen, edit image, thumbnail, miniatura, banner, ilustración, dibujo, diseña, pon en el draw, inserta en el whiteboard, agrega al canvas, ponlo en la pizarra, imagen en draw, image to whiteboard

Image Generation (OpenRouter → Gemini)

Generate images from text prompts or edit existing images via OpenRouter's Gemini image models. Optionally inject the result into the Draw/whiteboard canvas.

Command

cd /Users/danielcarreon/Developer/software/business-os/claudeclaw
npx tsx scripts/generate-image.ts --prompt "PROMPT" --upload [--image /path/to/input.png] [--output /path/to/output.png] [--size 2K] [--aspect 16:9] [--draw] [--page-id ID] [--draw-name NAME]

Arguments

Arg	Required	Description
--prompt	YES	Text description of what to generate or how to edit
--upload	ALWAYS USE	Uploads to Supabase Storage and prints public URL. MANDATORY for chat display.
--image	NO	Path to input image (for editing/transforming an existing image)
--output	NO	Custom output path. Default: workspace/generated/img-{timestamp}.png
--size	NO	Image size: 1K (default), 2K, 4K
--aspect	NO	Aspect ratio: 1:1, 16:9, 9:16, 4:3, 3:2, etc.
--model	NO	Model ID. Default: google/gemini-3.1-flash-image-preview (Nano Banana 2)
--draw	NO	After generating, inject the image into a Draw/whiteboard canvas
--page-id	NO	Target existing canvas (PATCH injection). If omitted with --draw, creates a new page (POST)
--draw-name	NO	Name for the new draw page. Only used when creating a new page (no --page-id). Default: "Generated Image"

Available Models

Model	Best for
google/gemini-3.1-flash-image-preview	Default. Nano Banana 2, best quality at flash speed
google/gemini-2.5-flash-image	Nano Banana 1, fast, good quality
google/gemini-3-pro-image-preview	Pro quality, supports 4K

Output

The script prints:

• URL:https://... — public URL from Supabase Storage (when --upload is used)
• IMAGE:/path/to/generated/img-1234567890.png — local path to saved image
• TEXT:description — any text the model included
• DRAW:page-id — Draw page UUID (when --draw is used)
• DRAW_URL:https://... — Full Draw page URL (when --draw is used)

Generated images are saved locally to claudeclaw/workspace/generated/ AND uploaded to Supabase Storage.

Examples

# Text to image (with upload for chat display)
npx tsx scripts/generate-image.ts --prompt "A minimalist logo with the letter S in cyan on dark background" --upload

# Edit existing image
npx tsx scripts/generate-image.ts --prompt "Remove the background" --image /path/to/photo.png --upload

# High resolution widescreen
npx tsx scripts/generate-image.ts --prompt "Futuristic city at sunset, cyberpunk" --size 2K --aspect 16:9 --upload

# Use Nano Banana 2 (best quality)
npx tsx scripts/generate-image.ts --model google/gemini-3.1-flash-image-preview --prompt "A watercolor painting of mountains" --upload

# Generate image + inject into current Draw canvas
npx tsx scripts/generate-image.ts --prompt "SaaS Factory funnel diagram" --upload --draw --page-id abc-123

# Generate image + create new Draw page
npx tsx scripts/generate-image.ts --prompt "Logo concepts" --upload --draw --draw-name "Logo Concepts"

# Generate + Draw only (no chat display needed)
npx tsx scripts/generate-image.ts --prompt "Architecture sketch" --draw

When to Use

Use this skill when Daniel:

• Asks to create or generate an image, illustration, design, or visual
• Asks to edit or modify an existing image (pass original with --image)
• Needs a thumbnail, banner, logo, or any visual asset
• Sends an image and asks to change something about it
• Asks to put a generated image on the Draw/whiteboard canvas (add --draw flag)

Draw / Whiteboard Integration (WebP + Storage Pipeline)

REGLA OBLIGATORIA: Cuando uses --draw, el script automaticamente convierte a WebP y sube a Supabase Storage. NUNCA construir payloads de canvas con base64 manualmente. SIEMPRE usar el script generate-image.ts --draw que maneja todo el pipeline.

Generate an image AND place it on the Draw whiteboard in one command. The --draw flag automatically converts to WebP, uploads to Supabase Storage, and injects the public URL into the canvas (no base64 in JSONB).

When to Add --draw

Add the --draw flag when Daniel says any of:

• "pon en el draw", "ponlo en el draw", "mete al draw"
• "inserta en el whiteboard", "agrega al canvas", "ponlo en la pizarra"
• "genera una imagen para el draw", "image on the whiteboard"
• Any request that combines image generation with placing it on a canvas

Context-Aware Page Targeting

When Daniel chats from a Draw page, the system automatically injects context like: [Context: Daniel is viewing draw page (page_id: abc-123)...]

• If context with page_id is present → use --draw --page-id <page_id> to inject into the current canvas
• If NO context is present → use just --draw to create a new Draw page
• If Daniel names a page → use --draw --draw-name "Name Here"

Combining with --upload

--draw and --upload can BOTH be used together:

• --upload gives a public URL for displaying the image in the chat message
• --draw puts the image on the canvas

When both are used, include the markdown image in your response AND confirm the Draw injection.

After Generating with --draw

1. If --upload was also used, include the image in your response as markdown: ![Description](URL)
2. Confirm the Draw injection in plain Spanish (no raw URLs):
   • Injected into existing canvas: "Imagen agregada al canvas."
   • New page created: "Imagen creada en nueva pagina de Draw."

Brand Colors

ALWAYS incorporate Daniel's brand colors in generated images:

• Primary: #8C27F1 (purple/violet)
• Secondary: #f69f02 (amber/orange)

Use these as the dominant color palette unless Daniel explicitly requests different colors.

Default Style

Unless Daniel specifies a different style, ALWAYS prepend this to the prompt:

"Bold thick strokes, simple shapes, didactic and dynamic illustration style, no clutter, high contrast. Primary colors: vibrant purple (#8C27F1) and amber orange (#f69f02) with dark backgrounds."

This produces clear, educational visuals that work great for YouTube, presentations, and teaching. Only override this default if Daniel explicitly asks for a different style (e.g., "photorealistic", "cyberpunk", "watercolor").

Prompt Tips

• Be specific: "A golden retriever wearing sunglasses on a beach at sunset" > "a dog"
• Specify style: "photorealistic", "illustration", "watercolor", "pixel art", "minimalist", "3D render"
• For edits: describe the change clearly: "Make the sky purple" or "Add a hat to the person"
• For logos/text: Gemini can render text in images, specify what text to include

After Generating — CRITICAL

ALWAYS include the image in your response as markdown so it renders in the chat:

![Description of the image](URL_FROM_SCRIPT_OUTPUT)

Extract the URL from the URL:https://... line in the script output and embed it as a markdown image. This way Daniel sees the image directly in Mission Control chat without having to find it on the filesystem.