Advanced Styling Reference#

Complete reference for all advanced styling and configuration options in visualkeras.

This page documents the powerful but less-commonly-used customization features available across all visualization modes.

Overview#

The options system in visualkeras supports:

Grouped highlights: Visually group related layers with backgrounds and labels
Logos and brand elements: Add logos and create legend entries
Per-layer styling: Customize individual layers with fine-grained control
Image embedding: Embed custom images or textures in visualizations
Text callables: Control what text appears in visualizations
Dimension configuration: Fine-tune how tensor shapes are displayed

All features are accessed through options objects and the styles parameter.

Grouped Highlights (layered_groups)#

Group related layers with colored backgrounds and labels.

Purpose#

Use grouped highlights to:

Visually separate architectural components (e.g., “Feature Extraction”, “Classification”)
Highlight specific stages of processing
Create visual hierarchy in complex models
Make diagrams easier to understand at a glance

Basic Syntax#

Groups are defined via the layered_groups parameter in LayeredOptions:

from visualkeras.options import LayeredOptions
import visualkeras

options = LayeredOptions(
    layered_groups=[
        {
            'name': 'Feature Extraction',
            'type': 'block',
            'indices': [0, 1, 2, 3],  # Layer indices to group
            'fill': '#e8f4f8',         # Background color
            'outline': '#4a90e2',      # Border color
        },
        {
            'name': 'Classification',
            'type': 'block',
            'indices': [4, 5, 6],
            'fill': '#f8e8e8',         # Different color
            'outline': '#e24a4a',
        }
    ]
)

image = visualkeras.show(model, mode='layered', options=options)
image.show()

Parameter Details#

Each group definition is a dictionary with:

Parameter	Type	Description
`name`	str	Label for the group (displayed above the block)
`type`	str	Must be `'block'` (groups consecutive layers into a visual block)
`indices`	list/tuple	Layer indices to include in the group (0-indexed from model.layers)
`fill`	color	Background fill color (hex string like `'#e8f4f8'` or CSS color name)
`outline`	color	Border color around the group

Practical Example: ResNet-style Model#

model = keras.Sequential([
    # Input/preprocessing layers
    keras.layers.Input(shape=(224, 224, 3)),
    keras.layers.Conv2D(64, (7, 7), padding='same'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.MaxPooling2D((3, 3)),

    # Feature extraction block
    keras.layers.Conv2D(128, (3, 3), padding='same'),
    keras.layers.BatchNormalization(),
    keras.layers.ReLU(),
    keras.layers.Conv2D(128, (3, 3), padding='same'),

    # Classification block
    keras.layers.GlobalAveragePooling2D(),
    keras.layers.Dense(256),
    keras.layers.ReLU(),
    keras.layers.Dense(10, activation='softmax'),
])

options = LayeredOptions(
    layered_groups=[
        {
            'name': 'Input & Preprocessing',
            'type': 'block',
            'indices': [0, 1, 2, 3, 4],
            'fill': '#e3f2fd',      # Blue
            'outline': '#1976d2',
        },
        {
            'name': 'Feature Extraction',
            'type': 'block',
            'indices': [5, 6, 7, 8],
            'fill': '#f3e5f5',      # Purple
            'outline': '#7b1fa2',
        },
        {
            'name': 'Classification Head',
            'type': 'block',
            'indices': [9, 10, 11, 12],
            'fill': '#e8f5e9',      # Green
            'outline': '#388e3c',
        }
    ]
)

image = visualkeras.show(model, mode='layered', options=options, preset='presentation')
image.show()

Logos and Branding (logo_groups, logos_legend)#

Add logos and create visual legends for your diagrams.

Purpose#

Use logos to:

Add company or institutional branding
Create visual legends explaining color coding
Add informational graphics
Improve publication figures

Basic Logo Syntax#

Define logos via the logo_groups and logos_legend parameters:

from visualkeras.options import LayeredOptions

options = LayeredOptions(
    logo_groups=[
        {
            'indices': [0, 1, 2],      # Layers to label
            'fill': '#3498db',         # Color
            'outline': '#2980b9',      # Border
            'text': 'Conv Block 1',    # Logo/label text
        }
    ],
    logos_legend=True  # Show legend explaining logos
)

image = visualkeras.show(model, mode='layered', options=options)
image.show()

Logo Parameter Details#

Parameter	Type	Description
`text`	str	Text label for the logo
`indices`	list/tuple	Layer indices to apply this logo to
`fill`	color	Background color for logo
`outline`	color	Border color for logo
`positions`	str	When to show: `'top'`, `'bottom'`, `'above'`, `'below'` (depends on context)

logos_legend can be:

True/False to enable/disable legend
A dict with legend customization options

Per-Layer Styling (styles Parameter)#

Fine-grained control over individual layers.

Purpose#

Use per-layer styling to:

Customize specific layers differently from others
Embed images in particular layers
Apply layer-specific colors or effects
Create highly customized publication figures

Basic Syntax#

Styles are passed via the styles parameter in LayeredOptions:

from visualkeras.options import LayeredOptions
import visualkeras

styles = {
    0: {                              # Layer 0 (by index)
        'fill': '#1f77b4',
        'outline': '#0d47a1',
    },
    'conv_1': {                       # Layer by name
        'fill': '#ff7f0e',
        'outline': '#d84315',
    },
    keras.layers.Dense: {             # By layer type (applies to all Dense layers)
        'fill': '#2ca02c',
        'outline': '#1b5e20',
    }
}

options = LayeredOptions(styles=styles)
image = visualkeras.show(model, mode='layered', options=options)
image.show()

Style Keys Reference#

Available style parameters per layer:

Key	Type	Description
`fill`	color	Layer background/fill color
`outline`	color	Layer border color
`line_width`	int	Border thickness in pixels
`text`	str	Custom text to display in or for the layer

Image Embedding (face_image)#

Embed custom images or textures in layer visualizations (LeNet mode).

Purpose#

Use image embedding to:

Show kernel visualizations
Display activation maps
Add custom graphics to layers
Create distinctive, publication-quality diagrams

Basic Syntax (LeNet Mode)#

from visualkeras.options import LenetOptions

styles = {
    0: {           # Layer 0
        'face_image': '/path/to/kernel_viz.png',
        'face_image_fit': 'cover',   # How to fit image
        'face_image_alpha': 200,     # Transparency (0-255)
        'face_image_inset': 2,       # Border inset in pixels
    }
}

options = LenetOptions(styles=styles)
image = visualkeras.show(model, mode='lenet', options=options)
image.show()

Image Options Reference#

Key	Type	Description
`face_image`	path/str	Path to an image file or a `PIL.Image` object
`face_image_fit`	str	Fit mode: `'cover'`, `'contain'`, `'fill'`, or `'match_aspect'`
`face_image_alpha`	int	Transparency from 0 (transparent) to 255 (opaque)
`face_image_inset`	int	Border inset in pixels

Fit Mode Behavior#

'cover': Fill entire space, crop image if necessary
'contain': Fit entire image, may have empty space (letterboxing)
'fill': Stretch to fill (may distort)
'stretch': Same as 'fill'
'scale-down': Use largest fitting size without upscaling

Image-Fitting Options (image_axis, image_fit)#

Control how images are displayed in layered visualizations.

Purpose#

These options control global image behavior in layered mode:

from visualkeras.options import LayeredOptions

options = LayeredOptions(
    image_axis='z',      # Which axis represents image depth (batch processing)
    image_fit='fill',    # How to fit images in layer blocks
)

ImageAxis Options#

Axis	Description
`'x'`	Image axis along width
`'y'`	Image axis along height
`'z'`	Image axis along depth/channels (default)

ImageFit Options#

Same options as face_image_fit: 'cover', 'contain', 'fill', 'stretch', 'scale-down'

Text Callables and Labels#

Customize what text appears in visualizations.

Using Built-in Callables#

from visualkeras.options import LAYERED_TEXT_CALLABLES, LayeredOptions
import visualkeras

# Use a built-in text callable
options = LayeredOptions(
    text_callable=LAYERED_TEXT_CALLABLES['name_shape']
)

image = visualkeras.show(model, mode='layered', options=options)
image.show()

Available built-in callables:

'name' - Layer name only
'type' - Layer type/class name
'shape' - Output tensor shape
'name_shape' - Both name and shape (recommended for presentation)

Custom Text Callables#

Define your own text function:

from visualkeras.options import LayeredOptions

def custom_label(layer_index, layer):
    """Return (text_string, is_above_layer)"""
    name = layer.__class__.__name__
    output_shape = layer.output_shape if hasattr(layer, 'output_shape') else 'unknown'
    text = f"{name}\n{output_shape}"
    above = False  # Show below the layer
    return (text, above)

options = LayeredOptions(
    text_callable=custom_label
)

image = visualkeras.show(model, mode='layered', options=options)
image.show()

LeNet Label Callables#

LeNet mode has special label callables:

from visualkeras.options import LenetOptions

def top_info(layer, render_shape):
    """Text above each layer"""
    return f"{render_shape.dH}×{render_shape.dW}×{render_shape.dZ}"  # h x w x channels

def bottom_info(layer, render_shape):
    """Text below each layer"""
    return layer.__class__.__name__

options = LenetOptions(
    top_label_callable=top_info,
    bottom_label_callable=bottom_info,
)

image = visualkeras.show(model, mode='lenet', options=options)
image.show()

RenderShape Properties (for LeNet)#

When using label callables in LeNet mode, the RenderShape object provides:

Attr	Description
`dH`	Height of the feature-map visualization in pixels
`dW`	Width of the feature-map visualization in pixels
`dZ`	Number of channels (depth)

Dimension Display Options#

Control how tensor dimensions are shown.

ShowDimension#

Enable dimension display on layers:

from visualkeras.options import LayeredOptions

options = LayeredOptions(
    show_dimension=True,      # Show tensor shapes on layers
    dimension_caps=None,      # Custom dimension display
)

image = visualkeras.show(model, mode='layered', options=options)
image.show()

DimensionCaps#

Limit dimension display formatting:

options = LayeredOptions(
    show_dimension=True,
    dimension_caps={
        'max_width': 100,     # Max characters for width display
        'max_height': 100,    # Max characters for height display
        'decimal_places': 2,  # Decimal precision
    }
)

SizingMode Strategies#

Control how layer sizes are calculated.

Available Modes#

Mode	Description
`'accurate'`	Use actual tensor dimensions (default and most precise)
`'balanced'`	Balance readability with accuracy
`'logarithmic'`	Use logarithmic scaling to handle a large range of sizes
`'relative'`	Scale relative to a base size controlled by `relative_base_size`

Example#

from visualkeras.options import LayeredOptions

# Logarithmic scaling for models with extreme size variations
options = LayeredOptions(
    sizing_mode='logarithmic',
    relative_base_size=20,  # Reference size for relative mode
)

image = visualkeras.show(model, mode='layered', options=options)
image.show()

Complete Advanced Example#

Combining multiple advanced features:

from visualkeras.options import LayeredOptions, LAYERED_TEXT_CALLABLES
import visualkeras

# Define advanced options
options = LayeredOptions(
    # Layout
    spacing=20,
    padding=30,

    # Styling
    background_fill='white',
    font_color='black',

    # Groups
    layered_groups=[
        {
            'name': 'Feature Extraction',
            'type': 'block',
            'indices': [0, 1, 2, 3],
            'fill': '#e3f2fd',
            'outline': '#1976d2',
        },
        {
            'name': 'Dense Classifier',
            'type': 'block',
            'indices': [4, 5, 6],
            'fill': '#e8f5e9',
            'outline': '#388e3c',
        }
    ],

    # Per-layer styling
    styles={
        keras.layers.Conv2D: {'fill': '#1976d2'},
        keras.layers.Dense: {'fill': '#388e3c'},
    },

    # Text
    text_callable=LAYERED_TEXT_CALLABLES['name_shape'],

    # Dimensions
    show_dimension=True,
    sizing_mode='balanced',

    # Legends
    legend=True,
    logos_legend=True,
)

image = visualkeras.show(
    model,
    mode='layered',
    options=options,
    preset='presentation'
)
image.save('advanced_diagram.png')

Tips and Best Practices#

Start with presets: Begin with preset='presentation' and customize from there
Use grouped highlights: Improve readability of complex models with visual grouping
Test fit modes: Try different image_fit values to find what looks best
Combine overlays: Use colors + groups + text for clearest communication
Save high-DPI: For publication, save PNG with sufficient resolution
Limit complexity: Too many style customizations can make diagrams harder to parse

Advanced Styling Reference#

Overview#

Grouped Highlights (layered_groups)#

Purpose#

Basic Syntax#

Parameter Details#

Practical Example: ResNet-style Model#

Logos and Branding (logo_groups, logos_legend)#

Purpose#

Basic Logo Syntax#

Logo Parameter Details#

Per-Layer Styling (styles Parameter)#

Purpose#

Basic Syntax#

Style Keys Reference#

Image Embedding (face_image)#

Purpose#

Basic Syntax (LeNet Mode)#

Image Options Reference#

Fit Mode Behavior#

Image-Fitting Options (image_axis, image_fit)#

Purpose#

ImageAxis Options#

ImageFit Options#

Text Callables and Labels#

Using Built-in Callables#

Custom Text Callables#

LeNet Label Callables#

RenderShape Properties (for LeNet)#

Dimension Display Options#

ShowDimension#

DimensionCaps#

SizingMode Strategies#

Available Modes#

Example#

Complete Advanced Example#

Tips and Best Practices#

See Also#