Wildcard Divide

ComfyUI custom node that specifies wildcard prompts for multiple regions

The above workflow is docs/example.json.

Wildcard Divide Node

This node incorporates the syntax of Impact Pack Wildcards while introducing additional syntactical features.

Weighted Child Selection

You can assign selection weights to options by prefixing them with a numerical value. This number determines the likelihood of that particular option being chosen.

hair:
  - 4, blonde
  - 5, black
  - 1, red

In this example, invoking __hair__ will result in "blonde" being selected with a probability of 4/(4+5+1) = 4/10 = 0.4. When a numerical prefix is omitted, a default weight of 1 is assumed.

This weighted selection mechanism is functionally equivalent to the following syntax in Impact Pack Wildcards:

hair:
  - {4::blonde|5::black|1::red}

Pattern-Based Selection

Entries beginning with / are evaluated against the preceding prompt context. The system selects candidates based on pattern matches. Here's an example:

outfit:
  - blouse, skirt, __legs__
  - shirt, pants, __legs__
  - swimsuit, __legs__
legs:
  - /skirt/ stockings
  - /pants/ socks
  - bare feet

Selection Logic

When __outfit__ expands (with equal 1/3 probability for each option):

• If it resolves to blouse, skirt, then __legs__ will select from either stockings or bare feet, as the /skirt/ pattern matches the context.
• If it resolves to shirt, pants, then __legs__ will select from either socks or bare feet, as the /pants/ pattern matches the context.
• Entries without patterns (like bare feet) are always included in the candidate pool.

This allows for contextually appropriate selections based on previously expanded wildcards.

Pattern Alternatives

When a line includes !, the text after ! will be selected when the pattern does not match the prompt. For example:

outfit:
  - blouse, skirt
  - dress
  - swimsuit
legs:
  - /swimsuit/ bare feet ! stockings

In this example, stockings will be selected when the outfit doesn't contain swimsuit (i.e., when blouse, skirt or dress is selected). Conversely, if swimsuit is selected, bare feet will be chosen.

Exclusive Pattern Matching

When a pattern ends with =, it becomes an exclusive pattern that will remove all non-matching options from consideration. For example:

outfit:
  - blouse, skirt
  - dress
  - swimsuit
legs:
  - /skirt/= stockings
  - bare feet

In this example:

• When __outfit__ selects blouse, skirt (1/3 probability):

  • The /skirt/= pattern matches
  • Due to the = suffix, all non-matching options (in this case, bare feet) are excluded
  • Therefore, stockings will be selected with 100% probability

• When __outfit__ selects either dress or swimsuit:

  • The /skirt/= pattern doesn't match
  • Only the non-pattern option bare feet remains available
  • Therefore, bare feet will be selected with 100% probability

Pattern Matching with Conditional Exclusion

The =~ suffix creates a sophisticated pattern matching rule that combines conditional exclusion with fallback behavior. When a pattern ends with =~, it implements the following logic:

outfit:
  - blouse, skirt
  - dress
  - swimsuit
legs:
  - /skirt/=~ stockings
  - bare feet
  - socks

This operates in two distinct modes:

1. When Pattern Matches: If __outfit__ contains skirt (probability: 1/3):

   • The /skirt/=~ pattern activates
   • All non-matching options (bare feet, socks) are excluded
   • stockings is selected with 100% probability

2. When Pattern Fails: If "skirt" is not present in __outfit__:

   • The pattern-matched option (stockings) remains in the candidate pool
   • All options become eligible for selection
   • Random selection occurs between stockings, bare feet, and socks

This mechanism provides a elegant way to enforce specific combinations while maintaining flexibility when conditions aren't met.

Split region

You can use [SEP] to divide an image into different regions. Each [SEP] divides the image into n equal parts.

scene: 2girls [SEP] blonde hair [SEP] black hair

For example, if written as above, 2girls would be applied to the entire image, blonde hair to the left half of the image, and black hair to the right half.

Split Direction

You can specify the orientation of the split using the opt:horizontal and opt:vertical options.

scene:
  - opt:horizontal 2girls [SEP] blonde hair [SEP] black hair
  - opt:vertical sky [SEP] blue sky [SEP] red sky

This syntax allows for precise control over image segmentation:

1. Horizontal Split (Left to Right): If the first option is selected, the image is divided horizontally. In this case:

   • 2girls applies to the entire image
   • blonde hair is applied to the left half
   • black hair is applied to the right half

2. Vertical Split (Top to Bottom): If the second option is chosen, the image is segmented vertically:

   • sky is applied across the entire image
   • blue sky affects the top half
   • red sky influences the bottom half

Image Size Specification

You can define the dimensions of the output image using the opt:widthxheight syntax. This feature allows for dynamic image size adjustment based on the selected option.

scene:
  - opt:1216x832 2girls [SEP] blonde hair [SEP] black hair
  - opt:832x1216 sky [SEP] blue sky [SEP] red sky

In this example, selecting the second option would result in an image with dimensions of 832x1216 pixels.

To implement this functionality, ensure that you connect the width and height outputs to the empty latent image node in your workflow. This connection enables the dynamic resizing of the output based on the specified dimensions.

Wild Prompt Generator Node

This node helps create prompts for the Wildcard Divide node.

Usage

Adding a New Slot

Click the Add Slot button to open a dialog where you can add a new slot to the wildcards file.

After clicking Save, the wildcards will be stored in custom_nodes/WildDivide/wildcards/m.yaml. For image generation, be sure to add a template slot. Use the format __m/slot_name__ to reference other slots within the template.

Configuring Slot Values

Select slot values from the dropdown menu:

• disabled: Replaces the slot with an empty string.
• random: Randomly selects a value from the available options.
• specific value: Replaces __m/slot__ with your selected value.

Retrieving Previous Random Values

The Get last random values button retrieves the most recently generated random selections, making it easier to fine-tune your prompt.

Creating Groups

Click the Add group button to create a new group of related slots.

Use the format __m/group/slot__ to reference slots within the group. Make sure to update the template slot to include these group references.

Reordering Slots

Rearrange slots easily using drag-and-drop functionality.

Viewing Random Selections

Check the box to display tooltips showing the last generated random values for each slot.

Adding Conditions to Slot Values

You can apply conditions to slot values, which are evaluated against the prompt generated up to that point. For example:

In this example, brown eyes is added to the candidate pool only if black hair is part of the prompt.

Special characters define conditions:

• ~: Negates a pattern match.
• &: Logical AND.
• |: Logical OR.

For instance, ~(black hair|brown hair) means the condition is true if neither black hair nor brown hair matches.

There are two candidate pools: normal and exclusive. You can control which pool a value is added to by ending the pattern with ? or =:

When selecting a value, the system prioritizes the exclusive pool if it’s not empty. Otherwise, it selects from the normal pool.

Examples:

In this case, if pants matches the prompt, the value is selected from the normal pool (stockings, thigh highs, and socks). If not, socks is excluded, leaving only stockings and thigh highs as options.

If the pattern is pants= and it matches, the value is chosen exclusively from the pool (socks). If it doesn’t match, socks is excluded entirely, leaving only stockings and thigh highs.

If the pattern is pants? and it matches, the value is chosen from the exclusive pool (socks). If it doesn’t match, the value is selected from the normal pool (stockings, thigh highs, and socks).