pollinations-ai

SKILL.md

Pollinations.ai Image Generation

Free, open-source AI image generation through simple URL parameters. No API key or signup required.

When to use this skill

Quick prototyping: Generate placeholder images instantly

Marketing assets: Create hero images, banners, social media content

Creative exploration: Test multiple styles and compositions rapidly

No-budget projects: Free alternative to paid image generation services

Automated workflows: Script-friendly URL-based API

Instructions

Step 1: Understand the API Structure
Pollinations.ai uses a simple URL-based API:

https://image.pollinations.ai/prompt/{YOUR_PROMPT}?{PARAMETERS}

No authentication required - just construct the URL and fetch the image.

Available Parameters:

width / height: Resolution (default: 1024x1024)

model: AI model (flux, turbo, stable-diffusion)

seed: Number for reproducible results

nologo: true to remove watermark (if supported)

enhance: true for automatic prompt enhancement

Step 2: Craft Your Prompt

Use descriptive prompts with specific details:

Good prompt structure:

[Subject], [Style], [Lighting], [Mood], [Composition], [Quality modifiers]

Example:

A father welcoming a beautiful holiday, warm golden hour lighting, 
cozy interior background with festive decorations, 8k resolution, 
highly detailed, cinematic depth of field

Prompt styles:

Photorealistic: "photorealistic shot, 8k resolution, highly detailed, cinematic"

Illustrative: "digital illustration, soft pastel colors, disney style animation"

Minimalist: "minimalist vector art, flat design, simple geometric shapes"

Step 3: Generate via URL (Browser Method)

Simply open the URL in a browser or use curl:

# Basic generation
curl "https://image.pollinations.ai/prompt/A_serene_mountain_landscape" -o mountain.jpg

# With parameters
curl "https://image.pollinations.ai/prompt/A_serene_mountain_landscape?width=1920&height=1080&model=flux&seed=42" -o mountain-hd.jpg

Step 4: Generate and Save (Python Method)

For automation and file management:

import requests
from urllib.parse import quote

def generate_image(prompt, output_file, width=1920, height=1080, model="flux", seed=None):
    """
    Generate image using Pollinations.ai and save to file
    
    Args:
        prompt: Description of the image to generate
        output_file: Path to save the image
        width: Image width in pixels
        height: Image height in pixels
        model: AI model ('flux', 'turbo', 'stable-diffusion')
        seed: Optional seed for reproducibility
    """
    # Encode prompt for URL
    encoded_prompt = quote(prompt)
    url = f"https://image.pollinations.ai/prompt/{encoded_prompt}"
    
    # Build parameters
    params = {
        "width": width,
        "height": height,
        "model": model,
        "nologo": "true"
    }
    if seed:
        params["seed"] = seed
    
    # Generate and save
    print(f"Generating: {prompt[:50]}...")
    response = requests.get(url, params=params)
    
    if response.status_code == 200:
        with open(output_file, "wb") as f:
            f.write(response.content)
        print(f"✓ Saved to {output_file}")
        return True
    else:
        print(f"✗ Error: {response.status_code}")
        return False

# Example usage
generate_image(
    prompt="A father welcoming a beautiful holiday, warm lighting, festive decorations",
    output_file="holiday_father.jpg",
    width=1920,
    height=1080,
    model="flux",
    seed=12345
)

Step 5: Batch Generation

Generate multiple variations:

prompts = [
    "photorealistic shot of a father at front door, warm lighting, festive decorations",
    "digital illustration of a father in snow, magical winter wonderland, disney style",
    "minimalist silhouette of father and child, holiday fireworks, flat design"
]

for i, prompt in enumerate(prompts):
    generate_image(
        prompt=prompt,
        output_file=f"variant_{i+1}.jpg",
        width=1920,
        height=1080,
        model="flux"
    )

Step 6: Document Your Generations

Save metadata for reproducibility:

import json
from datetime import datetime

metadata = {
    "prompt": prompt,
    "model": "flux",
    "width": 1920,
    "height": 1080,
    "seed": 12345,
    "output_file": "holiday_father.jpg",
    "timestamp": datetime.now().isoformat()
}

with open("generation_metadata.json", "w") as f:
    json.dump(metadata, f, indent=2)

Examples

Example 1: Hero Image for Website

generate_image(
    prompt="serene mountain landscape at sunset, wide 16:9, minimal style, soft gradients in blue tones, clean lines, modern aesthetic",
    output_file="hero-image.jpg",
    width=1920,
    height=1080,
    model="flux"
)

Expected output: 16:9 landscape image, minimal style, blue color palette

Example 2: Product Thumbnail

generate_image(
    prompt="futuristic dashboard UI, 1:1 square, clean interface, soft lighting, professional feel, dark theme, subtle glow effects",
    output_file="product-thumb.jpg",
    width=1024,
    height=1024,
    model="flux"
)

Expected output: Square thumbnail, dark theme, app store ready

Example 3: Social Media Banner

generate_image(
    prompt="LinkedIn banner for SaaS startup, modern gradient background, abstract geometric shapes, colors from purple to blue, space for text on left side",
    output_file="linkedin-banner.jpg",
    width=1584,
    height=396,
    model="flux"
)

Expected output: LinkedIn-optimized dimensions (1584x396), text-safe zone

Example 4: Batch Variations with Seeds

# Generate 4 variations of the same prompt with different seeds
base_prompt = "A father welcoming a beautiful holiday, cinematic lighting"

for seed in [100, 200, 300, 400]:
    generate_image(
        prompt=base_prompt,
        output_file=f"variation_seed_{seed}.jpg",
        width=1920,
        height=1080,
        model="flux",
        seed=seed
    )

Expected output: 4 similar images with subtle variations

Best practices

Use specific prompts: Include style, lighting, mood, and quality modifiers

Specify dimensions early: Prevents unintended cropping

Use seeds for consistency: Same seed + prompt = same image

Model selection:

flux: Highest quality, slower

turbo: Fast iterations

stable-diffusion: Balanced

Save metadata: Track prompts, seeds, and parameters for reproducibility

Batch similar requests: Generate style sets with consistent parameters

URL encode prompts: Use urllib.parse.quote() for special characters

Common pitfalls

Vague prompts: Add specific details about style, lighting, and composition

Ignoring aspect ratios: Check target platform requirements (Instagram 1:1, LinkedIn 1584x396, etc.)

Overly complex scenes: Simplify for clarity and better results

Not saving metadata: Difficult to reproduce or iterate on successful images

Forgetting URL encoding: Special characters break URLs

Troubleshooting

Issue: Inconsistent outputs

Cause: No seed specified
Solution: Use a fixed seed for reproducible results

generate_image(prompt="...", seed=12345, ...)  # Same output every time

Issue: Wrong aspect ratio

Cause: Incorrect width/height parameters
Solution: Use platform-specific dimensions

# Instagram: 1:1
generate_image(prompt="...", width=1080, height=1080)

# LinkedIn banner: ~4:1
generate_image(prompt="...", width=1584, height=396)

# YouTube thumbnail: 16:9
generate_image(prompt="...", width=1280, height=720)

Issue: Image doesn't match brand colors

Cause: No color specification in prompt
Solution: Include HEX codes or color names

prompt = "landscape with brand colors deep blue #2563EB and purple #8B5CF6"

Issue: Request fails (HTTP error)

Cause: Network issue or service downtime
Solution: Add retry logic

import time

def generate_with_retry(prompt, output_file, max_retries=3):
    for attempt in range(max_retries):
        if generate_image(prompt, output_file):
            return True
        print(f"Retry {attempt + 1}/{max_retries}...")
        time.sleep(2)
    return False

Output format

## Image Generation Report

### Request
- **Prompt**: [full prompt text]
- **Model**: flux
- **Dimensions**: 1920x1080
- **Seed**: 12345

### Output Files
1. `hero-image-v1.jpg` - Primary variant
2. `hero-image-v2.jpg` - Alternative style
3. `hero-image-v3.jpg` - Different lighting

### Metadata
- Generated: 2026-02-13T14:30:00Z
- Iterations: 3
- Selected: hero-image-v1.jpg

### Usage Notes
- Best for: Website hero section
- Format: JPEG, 1920x1080
- Reproducible: Yes (seed: 12345)

Multi-Agent Workflow

Validation & Quality Check

Round 1 (Orchestrator - Claude):

Validate prompt completeness

Check dimension requirements

Verify seed consistency

Round 2 (Executor - Codex):

Execute generation script

Save files with proper naming

Generate metadata JSON

Round 3 (Analyst - Gemini):

Review style consistency

Check brand alignment

Suggest prompt improvements

Agent Roles

Agent
Role
Tools

Claude
Prompt engineering, quality validation
Write, Read

Codex
Script execution, batch processing
Bash, Write

Gemini
Style analysis, brand consistency check
Read, ask-gemini

Example Multi-Agent Workflow

# 1. Claude: Generate prompts and script
# 2. Codex: Execute generation
bash -c "python generate_images.py"

# 3. Gemini: Review outputs
ask-gemini "@outputs/ Analyze brand consistency of generated images"

Metadata

Version

Current Version: 1.0.0

Last Updated: 2026-02-13

Compatible Platforms: Claude, ChatGPT, Gemini, Codex

Related Skills

image-generation - MCP-based image generation

design-system - Design system implementation

presentation-builder - Presentation creation

API Documentation

Official Site: https://pollinations.ai

API Endpoint: https://image.pollinations.ai/prompt/{prompt}

Models: flux, turbo, stable-diffusion

Tags

#pollinations #image-generation #free #api #url-based #no-signup #creative
1d:[