wuyoscar/gpt_image_2_skill: Curated GPT Image 2 Prompts, Agent Skills & CLI
wuyoscar/gpt_image_2_skill: Curated GPT Image 2 Prompts, Agent Skills & CLI
Developers working with OpenAI's latest image generation models face a familiar friction: crafting effective prompts, managing multi-modal workflows across agent runtimes, and avoiding the black-box limitations of built-in tools. wuyoscar/gpt_image_2_skill addresses this directly — a curated prompt gallery, agentic skill implementation, and standalone CLI for GPT Image 2 that integrates with Claude Code, Codex, and other skill-capable agents.
What is wuyoscar/gpt_image_2_skill?
wuyoscar/gpt_image_2_skill is an open-source Python↗ Bright Coding Blog project (MIT License, 3,749 GitHub stars, 315 forks as of July 2026) that bundles three complementary surfaces: a prompt gallery with copy-paste examples across 25+ visual categories, an agentic skill installable into Claude Code, Codex, OpenClaw, and Hermes Agent runtimes, and a Python CLI (gpt-image) that wraps OpenAI's /v1/images/generations and /v1/images/edits endpoints.
The repository is maintained by wuyoscar and last committed on 2026-07-01. It requires Python ≥ 3.11 and reads OPENAI_API_KEY from process environment, then .env, then ~/.env without overriding existing variables. The project explicitly positions itself as "small but mighty — curated for signal, not volume" rather than attempting to collect every prompt on the internet.
What distinguishes this project from generic prompt collections is its runtime integration: the skill format allows capable agents to discover, route, and execute image generation tasks through natural language requests like "generate the Boston Spring poster from the skill gallery." The CLI provides the same capabilities for terminal-driven workflows, with full parameter coverage for size, quality, format, and reference-image editing.
Key Features
Curated Prompt Gallery with 25+ Categories
The gallery spans anime/manga, gaming, retro/cyberpunk, cinematic animation, character design, typography/posters, illustration, watercolor, ink/Chinese art, pixel art, isometric, product/food, brand systems, photography, screen photography, infographics, research paper figures, official OpenAI cookbook examples, edit endpoint showcases, UI/UX mockups, data visualization, technical illustration, architecture/interior, scientific/educational content, fashion editorial, fine art painting, and tattoo design. Each entry includes the prompt text, recommended size, quality setting, and source attribution (Curated, X, Reddit, WeChat, Xiaohongshu, or GitHub archive).
Agentic Skill for Multiple Runtimes
The skills/gpt-image folder implements a portable skill format installable via:
- Claude Code:
/plugin marketplace add wuyoscar/gpt_image_2_skill - Codex:
$skill-installerwith GitHub URL, or manual copy to~/.codex/skills/ - Cross-agent
npx skillsinstaller:npx --yes skills@latest add wuyoscar/gpt_image_2_skill --skill gpt-image --agent codex --copy - Manual symlink to any runtime's skills directory
Full-Featured CLI
The gpt-image command supports text-to-image, single/multi-reference editing, and mask-based inpainting with complete parameter coverage: --size (1k/2k/4k/portrait/landscape/square/wide/tall or literal dimensions), --quality (auto/low/medium/high), --background (auto/opaque), --moderation (auto/low), --format (png/jpeg/webp), and --compression for lossy formats. Batch generation via -n produces suffixed filenames.
Reference-First Documentation Architecture
The skill loads category-specific reference files on demand rather than dumping all prompts into context: gallery.md serves as a routing index, gallery-*.md files contain per-category prompt atlases, craft.md provides a 19-section prompt-craft checklist, and openai-cookbook.md archives the official OpenAI guide verbatim.
Safety-Conscious Design
The README explicitly cautions against dropping generated images directly into academic papers, includes API-key safety tips (unset OPENAI_API_KEY before agent invocation if desired), and guards against accidental overwrites during skill installation with existence checks.
Use Cases
Agent-Assisted Creative Workflows Developers using Claude Code or Codex for iterative prototyping can request image generation without leaving their agent context. The skill surfaces gallery examples through natural language, reducing context-switching between chat interfaces and image generation UIs.
Research Visualization Prototyping
The dedicated "Research Paper Figures" category and explicit caution note make this suitable for generating reference visuals, workflow sketches, and reproducible style targets — not final publication assets. The quality=high recommendation for diagrams, small labels, and multi-panel layouts aligns with academic needs.
Game Development Pre-Production Gaming-focused prompts include AAA screenshot generation, HUD mockups, pixel-art RPG scenes, and worldbuilding boards. The multi-reference edit endpoint supports compositing character/environment elements iteratively.
Brand and Marketing Asset Generation
Typography/posters, product/food photography, brand systems, and event experience categories provide starting points for campaign visuals. The CLI's --quality budget knob (low for drafts, high for shipping assets) maps directly to production workflows.
UI/UX Mockup Rapid Prototyping The UI/UX mockups and data visualization categories target product designers needing quick, disposable screens for stakeholder alignment before committing to Figma/Sketch polish.
Installation & Setup
Before installing, verify existing installations to avoid conflicts:
command -v gpt-image || true
command -v uv >/dev/null && uv tool list | grep -E '^gpt-image-cli([[:space:]]|$)' || true
test -n "${OPENAI_API_KEY:-}" && echo "OPENAI_API_KEY is already set (value hidden)"
Claude Code (plugin marketplace):
/plugin marketplace add wuyoscar/gpt_image_2_skill
/plugin install gpt-image@wuyoscar-skills
Codex (built-in installer):
$skill-installer
Install this skill from GitHub:
https://github.com/wuyoscar/gpt_image_2_skill/tree/main/skills/gpt-image
Restart Codex after installation. Manual fallback:
git clone https://github.com/wuyoscar/gpt_image_2_skill.git
cd gpt_image_2_skill
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
test -e "${CODEX_HOME:-$HOME/.codex}/skills/gpt-image" && echo "gpt-image skill already exists; stop before overwriting" && exit 1
cp -R skills/gpt-image "${CODEX_HOME:-$HOME/.codex}/skills/"
Cross-agent installer (avoids --global by default):
npx --yes skills@latest add wuyoscar/gpt_image_2_skill \
--skill gpt-image --agent codex --copy
CLI installation via uv:
uvx --from git+https://github.com/wuyoscar/gpt_image_2_skill gpt-image -p "a cat astronaut"
# Or install to PATH
command -v gpt-image >/dev/null || uv tool install git+https://github.com/wuyoscar/gpt_image_2_skill
Update:
cd gpt_image_2_skill && git pull # manual clone
uv tool upgrade gpt-image-cli # CLI
Real Code Examples
Basic text-to-image generation:
# CLI: generate a photorealistic scene with explicit size and quality
gpt-image -p "a photorealistic convenience store at 10pm" --size 1k --quality high -f store.png
This maps to POST /v1/images/generations with model=gpt-image-2. The --size 1k shortcut resolves to 1024x1024; literal dimensions must be 16-pixel multiples with maximum edge 3840 and 3:1 aspect ratio cap.
Single-reference image editing:
# Restyle an existing image with text guidance
gpt-image -p "Make it a winter evening with heavy snowfall" \
-i chess.png --quality high -f chess-winter.png
The -i flag routes through POST /v1/images/edits as multipart form data. This is the official OpenAI cookbook endpoint, supporting image, prompt, size, quality, background, output_format, and n.
Multi-reference compositing:
# Combine elements from two source images with explicit constraints
gpt-image -p "Place the dog from image 2 next to the woman in image 1. Match the same lighting, composition, and background. Do not change anything else." \
-i woman.png -i dog.png --size portrait --quality medium -f woman-with-dog.png
Multiple -i inputs are supported for multi-reference edits. The prompt's explicit constraint language ("Do not change anything else") demonstrates the precision required for controlled compositing.
Mask-based inpainting:
# Opaque regions preserved, transparent regions regenerated
gpt-image -p "replace sky with aurora" \
-i photo.jpg -m sky_mask.png -f aurora.png
The mask must be PNG with alpha channel. This workflow enables surgical edits without regenerating the entire image.
OpenAI SDK equivalent (Python):
from openai import OpenAI
client = OpenAI()
result = client.images.generate(
model="gpt-image-2",
prompt="<PROMPT FROM GALLERY ENTRY>",
size="1024x1536", # CLI "portrait" maps to this
quality="high",
)
For edits, use client.images.edit(...) with image=[open(p, "rb") for p in refs].
Advanced Usage & Best Practices
Quality as Budget Control
The README explicitly notes there is no separate budget flag — use --quality as the dial. Start with low for cheap drafts and large sweeps, medium for normal exploration, and high only for finalists requiring text-heavy or shipping-facing polish. This mirrors practical MLOps and creative-ops workflows where cost scales with iteration count.
Prompt Structure
The embedded prompting fundamentals recommend: background/scene → subject → key details → constraints, with stated intended use (ad, UI mock, infographic) so the model selects appropriate polish level. Required text must be in straight quotes. Aspect ratio should be decided before prompt writing and reinforced in text, not just via --size.
Context-Efficient Skill Usage
The reference files are intentionally split: gallery.md for routing, category files loaded on demand. When working with agents, request specific categories rather than dumping the full atlas — this preserves context window for iterative refinement.
API Key Hygiene
The README's safety note bears repeating: if you do not want an agent runtime to accidentally invoke image generation with your OpenAI API key, run unset OPENAI_API_KEY before agent invocation. The CLI's env-var resolution (process → .env → ~/.env, no override) is predictable but requires awareness in shared environments.
Comparison with Alternatives
| Tool | Surface | Prompt Curation | Agent Integration | Open Source |
|---|---|---|---|---|
| wuyoscar/gpt_image_2_skill | Gallery + Skill + CLI | 162+ curated, 25+ categories | Native: Claude Code, Codex, OpenClaw, Hermes | MIT |
| OpenAI's built-in Codex image skill | Black-box skill | None exposed | Codex only | No |
| Midjourney / DALL-E web UIs | Web app | Community prompts | None | No |
| ComfyUI / Stable Diffusion workflows | Node/graph UI | Community models | Manual | Partial |
The trade-off is clear: wuyoscar/gpt_image_2_skill sacrifices the visual node-based control of ComfyUI for agent-native integration and curated prompt quality, while offering more transparency than OpenAI's built-in Codex skill. For developers already in Claude Code or Codex workflows, the skill eliminates context switching. For terminal-preferring users, the CLI provides equivalent capability without IDE lock-in. Web-native designers will still prefer Midjourney's interface for exploratory play.
FAQ
Q: Does this require a paid OpenAI API key?
Yes — GPT Image 2 is a paid API. The tool reads OPENAI_API_KEY but does not provide one.
Q: Can I use this with GPT-4o or DALL-E 3?
The CLI hardcodes model=gpt-image-2. For other models, modify the source or use the OpenAI SDK directly.
Q: Is the gallery content free to use commercially? The MIT license covers code and skill infrastructure. Generated images are subject to OpenAI's terms; curated prompts are provided as-is — verify attribution requirements for community-sourced entries.
Q: Why does the skill use symlinks for manual install?
Symlinks allow git pull in the clone directory to update the skill without re-copying. The installer guards against overwriting existing skills.
Q: What happens if I omit --quality?
Defaults to high. The README recommends overriding to low or medium for cost-sensitive batch work.
Q: Are the generated images suitable for academic publication? Explicitly not recommended as-is. The README's caution note advises treating outputs as references or workflow sketches only.
Q: How do I uninstall?
Remove the skill directory from your runtime's skills folder, or run uv tool uninstall gpt-image-cli for the CLI.
Conclusion
wuyoscar/gpt_image_2_skill fills a specific gap: developers and technical users who want curated GPT Image 2 prompts, agent-native integration, and terminal-accessible generation without surrendering to black-box tools or web-only workflows. The 3,749-star traction suggests this resonates with Claude Code and Codex users in particular.
It's best suited for: AI-assisted creative technologists, researchers needing rapid visualization prototypes, game developers in pre-production, and designers who prefer version-controlled, copy-paste prompt libraries over ephemeral web interfaces. It's not a replacement for professional design tools or a bypass for OpenAI's API costs — it's a productivity layer that reduces friction between intent and image.
Explore the gallery, install the skill for your agent runtime, or grab the CLI at https://github.com/wuyoscar/gpt_image_2_skill.
Outils recommandés
Explore on the BrightCoding network
Hand-picked resources from our other sites.
Stop Flying Blind in Claude Code: This Powerline Exposes Everything
Transform Claude Code from a black box into a glass cockpit with claude-powerline. Real-time usage tracking, deep git integration, context window intelligence,...
topoteretes/awesome-ai-memory: Curated AI Memory Stack Guide
topoteretes/awesome-ai-memory is a curated, actively maintained list of 40+ AI memory tools, frameworks, optimizers, and storage layers. It uses a three-dimensi...
GiovanniPasq/agentic-rag-for-dummies: Build Agentic RAG with LangGraph
Modular Agentic RAG system built with LangGraph featuring hierarchical indexing, conversation memory, multi-agent map-reduce, and human-in-the-loop query clarif...
Continuez votre lecture
The Ultimate Guide to Self-Hosted Workflow Automation Executors: Take Control of Your Automation Empire
AI Research Assistant: How Real-Time Web Scraping is Revolutionizing Knowledge Work in 2025
🎮 The Ultimate Guide to Open Source JavaScript Games: 100+ Free Games & Dev Tools You Can Use Today
Stop Coding Alone: OPC-Skills Gives Your AI Agent Superpowers
Commentaires 0
Aucun commentaire pour l'instant. Soyez le premier à réagir !