# Run without installing
uvx --from git+https://github.com/wuyoscar/gpt_image_2_skill gpt-image -p "a cat astronaut"

# Install to PATH permanently
uv tool install git+https://github.com/wuyoscar/gpt_image_2_skill
gpt-image -p "a cat astronaut"

/plugin marketplace add wuyoscar/gpt_image_2_skill
/plugin install gpt-image@wuyoscar-skills

$skill-installer install https://github.com/wuyoscar/gpt_image_2_skill/tree/main/skills/gpt-image

git clone https://github.com/wuyoscar/gpt_image_2_skill.git
cd gpt_image_2_skill

export AGENT_SKILLS_DIR="/path/to/your/agent/skills"
mkdir -p "$AGENT_SKILLS_DIR"
ln -s "$PWD/skills/gpt-image" "$AGENT_SKILLS_DIR/gpt-image"

export OPENAI_API_KEY="sk-..."

# Basic generation
gpt-image -p "a photorealistic convenience store at 10pm"

# With size, quality, and explicit output file
gpt-image -p "a neon-lit Tokyo alley at midnight" \
  --size portrait --quality high -f tokyo-alley.png

# Square, low quality (cheap draft)
gpt-image -p "watercolor mountains at sunrise" \
  --size 1k --quality low -f draft.png

# Batch: generates 4 variants, saved as out_0.png … out_3.png
gpt-image -p "product shot of a ceramic mug on white" \
  --size square --quality medium -n 4 -f out.png

# Single reference restyle
gpt-image -p "Make it a winter evening with heavy snowfall" \
  -i chess.png --quality high -f chess-winter.png

# Multi-reference composite: dog from image 2, scene from image 1
gpt-image -p "Place the dog from image 2 next to the woman in image 1. \
Match the same lighting, composition, and background." \
  -i woman.png -i dog.png --size portrait --quality medium -f woman-with-dog.png

# opaque pixels = keep, transparent pixels = regenerate
gpt-image -p "replace sky with aurora borealis" \
  -i photo.jpg -m sky_mask.png -f aurora.png

from openai import OpenAI

client = OpenAI()  # reads OPENAI_API_KEY from environment

result = client.images.generate(
    model="gpt-image-2",
    prompt="A photorealistic ceramic mug on a white studio background, "
           "soft directional light, light shadow beneath",
    size="1024x1024",   # square
    quality="high",
)

# Save result
import base64
from pathlib import Path

image_bytes = base64.b64decode(result.data[0].b64_json)
Path("mug.png").write_bytes(image_bytes)
print("Saved mug.png")

result = client.images.generate(
    model="gpt-image-2",
    prompt="Minimalist event poster: 'Boston Spring Jazz Festival · April 2026' "
           "in bold serif, pastel cherry-blossom watercolor background, centered layout",
    size="1024x1536",   # portrait (3:4)
    quality="high",
)

result = client.images.edit(
    model="gpt-image-2",
    image=open("chess.png", "rb"),
    prompt="Make it a winter evening with heavy snowfall, keep the chess pieces identical",
    size="1024x1024",
    quality="high",
)

result = client.images.edit(
    model="gpt-image-2",
    image=[open("woman.png", "rb"), open("dog.png", "rb")],
    prompt="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.",
    size="1024x1536",
    quality="medium",
)

result = client.images.edit(
    model="gpt-image-2",
    image=open("photo.jpg", "rb"),
    mask=open("sky_mask.png", "rb"),   # transparent = regenerate
    prompt="Replace the sky with dramatic aurora borealis, keep everything below the horizon identical",
    size="1024x1024",
    quality="high",
)

import base64
from pathlib import Path
from openai import OpenAI

def generate_batch(prompt: str, n: int = 4, size: str = "1024x1024",
                   quality: str = "medium", out_prefix: str = "variant") -> list[Path]:
    client = OpenAI()
    result = client.images.generate(
        model="gpt-image-2",
        prompt=prompt,
        size=size,
        quality=quality,
        n=n,
    )
    paths = []
    for i, item in enumerate(result.data):
        path = Path(f"{out_prefix}_{i}.png")
        path.write_bytes(base64.b64decode(item.b64_json))
        paths.append(path)
        print(f"Saved {path}")
    return paths

# Usage
variants = generate_batch(
    prompt="product shot of a blue glass water bottle, white background, studio lighting",
    n=4,
    quality="low",   # cheap sweep; rerun winner at high
)

[background/scene] → [subject] → [key details] → [constraints/intended use]

gpt-image -p "Clean scientific diagram: transformer architecture overview. \
White background, labeled encoder/decoder blocks with arrows, \
color-coded attention heads in teal and orange, \
sans-serif labels, publication-ready, 4K resolution" \
--size landscape --quality high -f transformer-diagram.png

gpt-image -p "Mobile app UI mockup, iOS style, dark mode. \
Fitness tracking dashboard: circular progress ring in neon green, \
daily steps '8,432', heart rate '74 bpm', \
bottom nav with 4 icons, pixel-perfect, no lorem ipsum" \
--size portrait --quality high -f fitness-app.png

gpt-image -p "Event poster. Text: 'SUMMER SONIC 2026' in bold condensed sans-serif. \
Subtext: 'Tokyo · August 9–10'. Vivid sunset gradient background (magenta to amber). \
Geometric grid overlay, high contrast, print-ready" \
--size portrait --quality high -f poster.png

gpt-image -p "Photorealistic product photo: matte black insulated coffee thermos, \
condensation droplets, placed on dark slate surface, \
single soft key light from upper-left, shallow depth of field, \
shot on Canon 5D, 85mm lens, commercial quality" \
--size square --quality high -f thermos.png

# Any text that must appear verbatim in the image — put in straight quotes in the prompt
prompt = '''Storefront sign reading "OPEN 24/7" in red neon. 
Below it: "Est. 1987" in smaller white block letters.
Realistic neon glow, night scene, rain-slicked pavement.'''

# Step 1: cheap sweep
gpt-image -p "minimalist logo for a coffee brand" --quality low -n 4 -f logo.png

# Step 2: pick winner (e.g. logo_2.png), rerun at high
gpt-image -p "minimalist logo for a coffee brand" --quality high -f logo-final.png

# image 1 = person, image 2 = garment
result = client.images.edit(
    model="gpt-image-2",
    image=[open("person.png", "rb"), open("shirt.png", "rb")],
    prompt="Dress the person in image 1 wearing the shirt from image 2. "
           "Keep the person's face, pose, and background identical. "
           "Natural fabric draping and lighting.",
    size="1024x1536",
    quality="high",
)

result = client.images.edit(
    model="gpt-image-2",
    image=open("billboard_photo.jpg", "rb"),
    mask=open("billboard_mask.png", "rb"),
    prompt='Replace the billboard face with: "SALE ENDS SUNDAY" '
           'in bold white text on solid red background. '
           'Match perspective and lighting of surrounding scene.',
    size="1536x1024",
    quality="high",
)

gpt-image -p "Anime key visual style (Studio Ghibli-inspired): \
young woman standing on a hillside overlooking a coastal town at golden hour, \
painterly backgrounds, soft cel shading, \
detailed environmental storytelling, cinematic composition" \
--size landscape --quality high -f anime-scene.png

# Replace text in an existing image in a different language
result = client.images.edit(
    model="gpt-image-2",
    image=open("menu_english.png", "rb"),
    prompt='Replace all English text with Japanese translations. '
           'Keep the exact same layout, fonts, colors, and imagery. '
           'Translate "Grilled Salmon" → "グリルサーモン", '
           '"Caesar Salad" → "シーザーサラダ".',
    size="1024x1024",
    quality="high",
)

export OPENAI_API_KEY="sk-..."
# or add to ~/.env — the CLI reads it automatically