AgentSkillsCN

Generate Image

生成图像

SKILL.md

Generate Image Skill

Generate and iterate on images using Gemini 3 Pro Image.

Usage

code
/generate-image [prompt]

Or: "Generate an image of..." / "Create a visual for..."

What It Does

  1. •Takes your prompt and generates an image via Gemini 3 Pro Image
  2. •Saves output to tools/image_gen/outputs/ directory
  3. •Maintains session for multi-turn refinement
  4. •Supports reference images for style guidance

The Script

Uses .scripts/image_gen.py which provides:

python
import sys; sys.path.insert(0, '.scripts')
from image_gen import generate, new_session, revert

# Generate new image
generate("a PM dashboard with dark mode", output_name="dashboard")

# Iterate on it (same session)
generate("make the charts more colorful", output_name="dashboard")

# Start fresh
new_session("dashboard")

# Undo last change
revert(1, "dashboard")

Parameters

ParameterDefaultOptions
output_nameautoCustom filename (also determines session)
aspect_ratio"1:1""1:1", "3:4", "4:3", "16:9", "9:16"
resolution"1K""1K", "2K", "4K"
reference_imagesNoneList of image paths for style reference

Multi-Turn Refinement

Each output_name gets its own session. To iterate:

python
# Turn 1: Initial generation
generate("minimalist logo for a PM tool called Moltbot", output_name="logo")

# Turn 2: Refine (same session, remembers context)
generate("make it more playful, add a subtle robot element", output_name="logo")

# Turn 3: Color adjustment
generate("try it in purple and teal", output_name="logo")

Instructions for Claude

When the user asks you to generate an image:

  1. •

    Run the generation:

    bash
    cd "/Users/carl/Documents/Carl's Life OS/📦 Projects/peter-yang-podcast-prep/demo-folder/tools/image_gen" && python3 -c "
import sys; sys.path.insert(0, '.scripts')
from image_gen import generate
generate('USER_PROMPT_HERE', output_name='descriptive_name')
"
  2. •

    Show the output path so the user can view it

  3. •

    For refinements, use the same output_name to continue the session

  4. •

    If they want to start over, call new_session('name')

Example Conversation

code
User: "Generate a hero image for the Moltbot landing page"

Claude: [Runs generate with prompt, output_name="moltbot-hero"]
Generated: tools/image_gen/outputs/moltbot-hero.png

User: "Make it more dynamic, add some motion blur"

Claude: [Runs generate with refinement prompt, same output_name]
Refined: tools/image_gen/outputs/moltbot-hero.png (updated)

User: "Actually go back to the previous version"

Claude: [Runs revert(1, "moltbot-hero")]
Reverted to previous version.

Requirements

  • •GEMINI_API_KEY in .env
  • •pip install google-genai pillow python-dotenv