القائمة
Generate and edit images using OpenAI's gpt-image-2 model — the highest-quality image generation model available (Arena
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Process email inboxes, deliver formatted briefs, and manage action items for each team member. Use when any conversation touches briefs, email, action items, inboxes, or whats-happening questions — and ALWAYS when a user states or changes any preference about how they're briefed or how their email is handled (e.g. "stop briefing me about X", "put Y in my morning brief", a reply to a delivered brief). Load this skill BEFORE acting so the preference lands in the right file.
Create educational comic-book-style zines that explain any concept visually using AI image generation. Use this skill whenever someone wants to make a zine, comic, graphic explainer, visual guide, illustrated tutorial, or comic-book-style educational content. Triggers on requests like "make a zine about X", "create a comic explaining Y", "I want to make a visual explainer", "illustrated guide to Z", "graphic novel style walkthrough", or any request to turn a concept into a visual, panel-based, comic-book-style format. Also use when someone references "wizard zines", "Julia Evans style", or wants to create something that teaches through illustrated pages.
Pixel-perfect UI implementation from Figma designs with ruthless visual verification. Use this skill whenever the user wants to implement a UI from a Figma design, mockup, or spec and needs it to be exact — not approximate, not close enough, but pixel-perfect. Triggers include: 'implement this Figma', 'build this design', 'pixel-perfect', 'match the mockup exactly', 'implement this UI spec', 'code this design', 'convert Figma to code', or any request where the user has a visual design and wants it implemented with zero deviation. Also use when the user provides a Figma file, design screenshot, or UI specification and expects faithful reproduction in code.
Help the user write an article or post by interviewing them and polishing their draft. Use this skill when the user wants to write an article, blog post, or share insights about work they've done. They're excited and want to capture that excitement — not have it rewritten by AI. This is NOT ghostwriting. Use when the user says "help me write", "I want to write about", "I have a draft", "help me with my article", or shares a brain dump they want polished.
Start a timed discovery interview with the user. Use this skill when the user wants to be interviewed, wants you to ask them questions to learn about them or what they're working on, says "interview me", "ask me questions", "do a discovery call", or wants a structured Q&A session. Default duration is 5 minutes.
Run a timed brainstorming session with the user. Acts as a great coach — interviews the user to understand their idea, proposes cross-questions, expands on their thinking, and helps them arrive at clarity. Use this skill when the user wants to brainstorm, think through an idea, explore a problem space, or have a structured creative conversation. Default session length is 10 minutes.
| name | openai-imagegen |
| description | Generate and edit images using OpenAI's gpt-image-2 model — the highest-quality image generation model available (Arena |
Generate and edit images using OpenAI's gpt-image-2 API — the highest-quality image model available as of April 2026.
Check if OPENAI_API_KEY is set in the environment.
If OPENAI_API_KEY is NOT available → use ChatGPT Prompt Mode. Instead of running scripts, output a ready-to-paste prompt the user can copy into ChatGPT. Follow these rules:
If OPENAI_API_KEY IS available → use the scripts. Use the venv Python for all script invocations:
SCRIPTS_DIR="${CLAUDE_PLUGIN_ROOT}/skills/openai-imagegen/scripts"
PYTHON="${SCRIPTS_DIR}/.venv/bin/python3"
If the venv doesn't exist yet, create it:
python3.12 -m venv "${SCRIPTS_DIR}/.venv"
"${SCRIPTS_DIR}/.venv/bin/pip" install openai Pillow
Use these scripts for all image generation tasks. Always invoke with $PYTHON:
Use when: User wants to create a new image from a text description
$PYTHON "${SCRIPTS_DIR}/generate_image.py" "prompt" output.png [--quality QUALITY] [--size SIZE] [--format FORMAT] [--preset PRESET] [--n COUNT]
Use when: User wants to edit an image, use reference images, or do inpainting with a mask
# Edit with instruction
$PYTHON "${SCRIPTS_DIR}/edit_image.py" "instruction" output.png --image input.png [--quality QUALITY] [--size SIZE]
# Edit with mask (inpainting)
$PYTHON "${SCRIPTS_DIR}/edit_image.py" "instruction" output.png --image input.png --mask mask.png
# Compose from multiple reference images
$PYTHON "${SCRIPTS_DIR}/edit_image.py" "instruction" output.png --image ref1.png --image ref2.png
| Option | Values | Default |
|---|---|---|
--quality | low (drafts, fast), medium (default, good balance), high (print-ready — only use when explicitly requested), auto | medium |
--size | Any WxH where both are multiples of 16, max edge 3840px, aspect ≤3:1. Or auto | auto |
--format | png, jpeg, webp | png |
--compression | 0-100 (for jpeg/webp only) | None |
--n | 1-8 (number of images) | 1 |
--preset | draft, social, print, wide, portrait | None |
| Preset | Quality | Size | Format | Best for |
|---|---|---|---|---|
draft | low | 1024x1024 | jpeg | Quick iterations, brainstorming |
social | medium | 1080x1080 | png | Instagram, social posts |
print | medium | 2048x2048 | png | Print materials, high-res assets |
wide | medium | 1920x1080 | png | Presentations, headers, banners |
portrait | medium | 1080x1920 | png | Stories, mobile, posters |
Images are saved to the specified output path. For Slack sharing, save to ~/work/ or /tmp/ and use the attach: prefix in your response:
Here's your image attach:/tmp/generated-image.png
When generating multiple images (--n > 1), files are saved as output_1.png, output_2.png, etc.
gpt-image-2 is a generational leap. It has native reasoning — it thinks about your prompt before drawing. Throw away every instinct from DALL-E, Midjourney, or Stable Diffusion about what AI image generation "can't do." This model:
Be as specific and ambitious as you want. Describe exactly what you see in your head. The more precise and detailed your vision, the better the output. Don't simplify for the model — it can handle it.
Order your prompt for clarity:
Just write the exact text you want. The model renders it accurately. For best results:
Default is medium quality — good enough for most uses at ~$0.05/image. Only bump to high when the user explicitly asks for print-ready or highest quality output. Use low for rapid iteration and brainstorming.
| Scenario | Use |
|---|---|
| Highest quality, client-facing assets | OpenAI (this skill) |
| Accurate text in images | OpenAI |
| Multi-panel with character consistency | OpenAI |
| Quick/cheap generation, exploration | Gemini |
| Speed-critical (1-3s generation) | Gemini |
| 4K output | Gemini (pro model) |
| Budget-sensitive bulk generation | Gemini |
At 1024x1024:
Default quality is medium (~$0.05/image). Only use high when the user explicitly requests it.