image-generation

Files (1)

• SKILL.md

SKILL.md (first lines)

# Image Generation Skill

## Setup

```bash
pip install openai replicate pillow requests python-dotenv
# For local Stable Diffusion:
pip install diffusers transformers torch accelerate
```

`.env`:
```
OPENAI_API_KEY=sk-...
REPLICATE_API_TOKEN=r8_...
FAL_KEY=...
```

## DALL-E 3 (OpenAI) — Easiest, Highest Quality

```python
import openai
import os
import requests
from PIL import Image
from io import BytesIO
from dotenv import load_dotenv

load_dotenv()
client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

def generate_dalle3(
    prompt: str,
    size: str = "1024x1024",   # "1024x1024" | "1792x1024" | "1024x1792"
    quality: str = "hd",       # "standard" | "hd"
    style: str = "vivid",      # "vivid" | "natural"
    n: int = 1,
    save_path: str = None
) -> dict:
    """Generate an image with DALL-E 3."""
    response = client.images.generate(
        model="dall-e-3",
        prompt=prompt,
        size=size,
        quality=quality,
        style=style,
        n=n
    )
    
    results = []
    for i, img_data in enumerate(response.data):
        url = img_data.url
        revised_prompt = img_data.revised_prompt
        
        if save_path:
            img_bytes = requests.get(url).content
            path = save_path if n == 1 else save_path.replace(".", f"_{i}.")
            with open(path, "wb") as f:
                f.write(img_bytes)
        
        results.append({"url": url, "revised_prompt": revised_prompt})
    
    return results[0] if n == 1 else results

def edit_image_dalle(
    image_path: str,
    mask_path: str,
    prompt: str,
    size: str = "1024x1024"
) -> dict:
    """Edit an image using DALL-E inpainting (transparent mask area gets replaced)."""
    with open(image_path, "rb") as img, open(mask_path, "rb") as mask:
        response = client.images.edit(
            model="dall-e-2",  # DALL-E 2 for edits
            image=img,
            mask=mask,
            prompt=prompt,
            size=size,
            n=1
        )
    return {"url": response.data[0].url}

def create_variation(image_path: str, n: int = 2) -> list[str]:
    """Create variations of an existing image."""
    with open(image_path, "rb") as img:
        response = client.images.create_variation(model="dall-e-2", image=img, n=n, size="1024x1024")
    return [d.url for d in response.data]
```

## FLUX (via Replicate) — Best Open-Source Quality

```python
import replicate

def generate_flux(
    prompt: str,
    model: str = "black-forest-labs/flux-schnell",  # or "flux-dev", "flux-pro"
    width: int = 1024,
    height: int = 1024,
    num_outputs: int = 1,
    num_inference_steps: int = 4,   # 4 for schnell, 28 for dev/pro
    guidance_scale: float = 3.5,
    save_path: str = None
) -> list[str]:
    """Generate image(s) using FLUX on Replicate."""
    output = replicate.run(
        model,
        input={
            "prompt": prompt,
            "width": width,
            "height": height,
            "num_outputs": num_outputs,
            "num_inference_steps": num_inference_steps,
            "guidance_scale": guidance_scale,
        }
    )
    
    urls = list(output)
    if save_path:
        for i, url in enumerate(urls):

From garri333/Skills, MIT.

More from garri333/Skills