SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillscreative-image-generator

creative-image-generator

Design & creation

View repository

Updated 10-15 21:08

About this skill

Creative Image Generator

Generate advertising images via Gemini 2.0 Flash Experimental.

Workflow

Step 1: Identify the account

Ask the user which account the creative is for. Check the list of accounts in .claude/ads-agent/config/ad_accounts.md

Step 2: Read the brief

CRITICAL: Read .claude/ads-agent/config/briefs/{account}.md

Step 3: Gather the texts

Get from the user or call /creative-copywriter:

  • offer — headline (6-12 words)
  • bullets — 3 bullets via \n
  • profits — benefit/bonus
  • cta — call to action

Step 4: Ask for the prompt

Find out from the user:

  • What style/mood?
  • What colors?
  • What objects in the background?
  • Is there a reference? (path to image)

Step 5: Generate

Call MCP tool generate_creative_image() with parameters

Step 6: Show the result

  • Read the created file via Read tool → show a preview to the user
  • Output the file path for opening in Finder/Preview

Step 7: Stories format (optional)

If 9:16 Stories is needed → call expand_to_stories()

MCP Tools

generate_creative_image

Generate a 4:5 creative (1080×1350).

Parameters:

  • prompt — style, colors, objects, mood
  • offer — headline
  • bullets — 3 bullets via \n
  • profits — benefit
  • cta — call to action (default: "Learn more")
  • reference_image_path — (optional) path to reference
  • output_dir — (optional) directory for saving

Result: JSON с file_path to PNG file

expand_to_stories

Extend 4:5 to 9:16 via outpainting.

Parameters:

  • image_path — path to the 4:5 image
  • output_path — (optional) path for the result

Result: JSON с file_path to 9:16 PNG file

References

The user can attach a reference image for the style:

  • Pass the path to reference_image_path
  • Gemini will take colors, composition, mood from the reference
  • Texts will be added on top in the style of the reference

Viewing the result

After generation:

  1. Read tool — show a preview in the chat (Claude Code multimodal)
  2. File path — the user will open it in Finder/Preview
  3. The file is saved to the working directory

Requirements

  • GEMINI_API_KEY — environment variable with the Gemini API key
  • google-generativeai — Python package (pip install google-generativeai)

Workflow example

Request: "Generate a creative for bas_dent about implantation"

Actions:

  1. I read the brief .claude/ads-agent/config/briefs/bas_dent.md
  2. I ask for the texts or call /creative-copywriter
  3. I ask for the style:
    • "What style? (minimalism, UGC, bright...)"
    • "What colors? (blue, white...)"
    • "Is there a reference?"
  4. I call generate_creative_image()
  5. I read the file via Read → show a preview
  6. I ask: "Do you need the Stories format 9:16?"
  7. If yes → expand_to_stories()