Custom Image Styles¶

Learn how to customize the visual style of your generated images.

Understanding Image Prompts¶

Direktor generates image prompts based on your transcript content. These prompts are stored in image_prompts.json:

[
  {
    "time": 0,
    "prompt": "A stunning view of the James Webb Space Telescope floating in deep space, golden hexagonal mirrors gleaming against a backdrop of distant galaxies"
  },
  {
    "time": 30,
    "prompt": "An artistic rendering of an exoplanet with visible water vapor in its atmosphere, illuminated by a distant star"
  }
]

Editing Prompts Manually¶

After running stage 4, edit the prompts before generating images:

# Run through stage 4
direktor input.txt --stage 4

# Edit prompts
nano temp/*/image_prompts.json

# Continue with image generation
direktor input.txt --stage 5

Style Modifiers¶

Add style modifiers to your prompts for consistent aesthetics:

Photorealistic Style¶

{
  "prompt": "your scene description, photorealistic, 8k resolution, detailed lighting, professional photography"
}

Artistic Style¶

{
  "prompt": "your scene description, digital art, vibrant colors, concept art style, trending on artstation"
}

Minimalist Style¶

{
  "prompt": "your scene description, minimalist design, clean lines, simple shapes, modern aesthetic"
}

Programmatic Customization¶

You can modify prompts programmatically after generation:

import json
from direktor.core.images import generate_images

# Load existing prompts
with open("temp/abc123/image_prompts.json", "r") as f:
    prompts = json.load(f)

# Add style modifier to all prompts
style = ", digital art, vibrant colors, highly detailed"
for prompt in prompts:
    prompt["prompt"] += style

# Save modified prompts
with open("temp/abc123/image_prompts.json", "w") as f:
    json.dump(prompts, f)

# Generate images with new prompts
images = generate_images(prompts, "temp/abc123")

Custom Prompt Generator¶

Override the default prompt generation:

from openai import OpenAI

def custom_prompt_generator(text_segment, timestamp):
    """Generate a custom image prompt for a transcript segment."""
    client = OpenAI()

    response = client.chat.completions.create(
        model="gpt-4-turbo-preview",
        messages=[
            {
                "role": "system",
                "content": """Generate an image prompt in a specific style.
                Style: Cinematic, dramatic lighting, wide angle shot.
                Format: Scene description only, no meta text."""
            },
            {
                "role": "user",
                "content": f"Create an image prompt for: {text_segment}"
            }
        ]
    )

    return {
        "time": timestamp,
        "prompt": response.choices[0].message.content.strip()
    }

Image Generation Parameters¶

The FLUX model accepts several parameters you can customize:

input_data = {
    "prompt": "your prompt",
    "num_outputs": 1,
    "aspect_ratio": "16:9",    # Options: "1:1", "16:9", "9:16", "4:3"
    "output_format": "webp",    # Options: "webp", "png", "jpg"
    "output_quality": 80,       # 1-100
    "seed": 0,                  # For reproducibility
}

Tips for Better Images¶

Be Specific: Include details about lighting, perspective, and mood
Avoid Negatives: Describe what you want, not what you don't want
Keep Consistent: Use similar style modifiers across prompts
Test First: Generate one image to test your style before full batch