ComfyUI Curved Weight Schedule

Advanced ControlNet scheduling and regional prompting nodes for ComfyUI. Control your ControlNet strength across time and space with precision and visual feedback.

🌟 Features

• Curved Timestep Keyframes: Schedule ControlNet strength across generation steps with 14 different curve types
• Visual Feedback: Real-time graph preview showing your strength curve
• Multi-Mask Strength Combiner: Apply different ControlNet strengths to different regions of your image
• Regional Prompting: Use different text prompts for different masked areas
• Multiple Blend Modes: Max, Add, Multiply, and Average for flexible mask combination

📦 Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes/
   

2. Clone this repository:

   git clone https://github.com/diffussy69/comfyui-curved_weight_schedule.git
   

3. Install dependencies (if not already installed):

   pip install matplotlib pillow numpy torch
   

4. Restart ComfyUI

The nodes will appear in:

• conditioning/controlnet → Curved Timestep Keyframes
• mask → Multi-Mask Strength Combiner
• conditioning → Regional Prompting

🎯 Node Overview

1. Curved Timestep Keyframes

Control ControlNet strength across generation steps using mathematical curves.

Key Parameters:

• num_keyframes: Number of control points (2-100)
• start_percent / end_percent: When to start/stop the curve (0.0-1.0)
• start_strength / end_strength: Strength values at start and end
• curve_type: Shape of the strength transition
• curve_param: Controls transition speed/steepness

Available Curve Types:

• strong_to_weak: Start with high control, gradually release
• weak_to_strong: Build up control over time
• linear: Straight line transition
• ease_in / ease_out / ease_in_out: Smooth acceleration curves
• bell_curve: Strong in middle, weak at ends
• reverse_bell: Weak in middle, strong at ends
• exponential_up / exponential_down: Dramatic transitions
• sine_wave: Oscillating control (experimental)
• bounce: Bouncing effect
• custom_bezier: Customizable bezier curve

Outputs:

• TIMESTEP_KF: Connect to Apply Advanced ControlNet's timestep_kf input
• curve_graph: Visual preview (connect to Preview Image)

2. Multi-Mask Strength Combiner

Combine up to 5 separate masks with different ControlNet strengths.

Key Parameters:

• base_strength: Global multiplier for all masks
• mask_X: Individual mask inputs (1-5)
• mask_X_strength: Strength multiplier for each mask
• blend_mode: How overlapping masks combine
• normalize_output: Clamp result to [0,1]

Blend Modes:

• max: Takes highest strength (best for separate regions)
• add: Adds strengths together (for layering)
• multiply: Multiplies strengths (for soft effects)
• average: Averages all strengths (for smooth blending)

Output:

• combined_mask: Connect to Apply Advanced ControlNet's mask_optional input

3. Regional Prompting

Apply different text prompts to different regions of your image.

Key Parameters:

• clip: Your CLIP model
• base_positive: Base prompt applied to entire image
• region_X_mask: Mask for each region (1-5)
• region_X_prompt: Text prompt for that region
• region_X_strength: How strongly the prompt affects the region

Output:

• conditioning: Connect to KSampler's positive input

💡 Usage Examples

Example 1: Composition Lock (Strong Start, Fade Out)

Lock in composition early, then let the model add details freely.

Workflow:

┌─────────────────┐
│ Load ControlNet │
└────────┬────────┘
         │
┌────────▼──────────────────┐
│ Curved Timestep Keyframes │
│ - curve_type: strong_to_weak
│ - start_percent: 0.0
│ - end_percent: 0.4
│ - start_strength: 1.0
│ - end_strength: 0.1
│ - curve_param: 3.0
└────────┬──────────────────┘
         │
┌────────▼──────────────────┐
│ Apply Advanced ControlNet │
│ (timestep_kf input)       │
└───────────────────────────┘

Result: ControlNet strongly guides early steps (structure), then releases control by 40% for creative details.

Example 2: Regional Control with Different Strengths

Apply ControlNet more strongly to some areas than others.

Workflow:

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│ Mountains    │     │ Flowers      │     │ Sky          │
│ Mask         │     │ Mask         │     │ Mask         │
└──────┬───────┘     └──────┬───────┘     └──────┬───────┘
       │                    │                    │
       └────────────────────┼────────────────────┘
                            │
                ┌───────────▼──────────────┐
                │ Multi-Mask Strength      │
                │ Combiner                 │
                │ - mask_1_strength: 1.5   │  (mountains - high)
                │ - mask_2_strength: 0.4   │  (flowers - low)
                │ - mask_3_strength: 0.8   │  (sky - medium)
                │ - blend_mode: max        │
                └───────────┬──────────────┘
                            │
                ┌───────────▼──────────────┐
                │ Apply Advanced ControlNet│
                │ (mask_optional input)    │
                └──────────────────────────┘

Result: Mountains follow reference closely (1.5x), flowers have creative freedom (0.4x), sky is moderately guided (0.8x).

Example 3: Regional Prompting with Different Descriptions

Use different text prompts for different regions.

Workflow:

┌──────────────┐
│ CLIP Model   │
└──────┬───────┘
       │
┌──────▼────────────────────────┐
│ Regional Prompting            │
│ - base_positive:              │
│   "masterpiece, photorealistic"│
│                               │
│ - region_1_mask: mountains    │
│ - region_1_prompt:            │
│   "snowy peaks, dramatic"     │
│ - region_1_strength: 1.2      │
│                               │
│ - region_2_mask: flowers      │
│ - region_2_prompt:            │
│   "wildflowers, soft bokeh"   │
│ - region_2_strength: 1.0      │
└───────────┬───────────────────┘
            │
    ┌───────▼────────┐
    │ KSampler       │
    │ (positive)     │
    └────────────────┘

Result: Mountains get "snowy peaks" style, flowers get "wildflowers" style, with base quality tags applied everywhere.

Example 4: Ultimate Control (All Three Nodes Combined)

Combine timestep curves + regional ControlNet strengths + regional prompts.

Workflow:

                    ┌─────────────┐
                    │ CLIP Model  │
                    └──────┬──────┘
                           │
            ┌──────────────▼─────────────────┐
            │ Regional Prompting             │
            │ - Different prompts per region │
            └──────────┬─────────────────────┘
                       │
                       ├─────────────────────┐
                       │                     │
         ┌─────────────▼──────┐    ┌────────▼─────────────┐
         │ KSampler           │    │ Curved Timestep      │
         │ (positive input)   │    │ Keyframes            │
         └────────────────────┘    │ - Strength over time │
                                   └────────┬─────────────┘
                                            │
                  ┌─────────────────────────┤
                  │                         │
    ┌─────────────▼──────────────┐   ┌─────▼──────────────────┐
    │ Multi-Mask Strength        │   │ Apply Advanced         │
    │ Combiner                   │───│ ControlNet             │
    │ - Different strengths      │   │ - timestep_kf          │
    │   per region               │   │ - mask_optional        │
    └────────────────────────────┘   └────────────────────────┘

Result:

• Different prompts per region (mountains, flowers, sky)
• Different ControlNet strengths per region (strong, medium, weak)
• Strength fades over time (starts strong, ends weak)
• Maximum creative control! 🎨

🎨 Practical Tips

Curve Selection Guide

For Composition Control:

• strong_to_weak with curve_param=2.5-4.0 → Lock composition early, release later
• Set end_percent=0.3-0.5 → Only control first 30-50% of generation

For Detail Refinement:

• weak_to_strong with curve_param=2.0 → Let model generate freely, then guide details
• Set start_percent=0.5 → Only apply ControlNet in second half

For Style Consistency:

• bell_curve with curve_param=2.0-3.0 → Strong guidance during structure formation
• Weak at start/end for creative freedom

For Experimental Effects:

• sine_wave → Oscillating control (wild results!)
• bounce → Rhythmic control variations

Mask Painting Tips

1. Paint each region as a separate mask - don't worry about opacity
2. Use full opacity - the strength settings control the effect
3. Allow small overlaps - use blend_mode=max to handle them
4. Test one region at a time - easier to dial in individual strengths

Regional Prompting Best Practices

Base Prompt:

• Use for universal quality tags: "masterpiece, high quality, detailed"
• Avoid specific objects/subjects

Regional Prompts:

• Be specific about what's in that region
• Include style/lighting/atmosphere for that area
• Can include negative concepts if needed

Strength Values:

• Start at 1.0 for all regions
• Increase if a region isn't responding (1.2-1.5)
• Decrease if a region is overpowering others (0.6-0.8)

🐛 Troubleshooting

Issue: Graph not showing

• Solution: Make sure matplotlib is installed: pip install matplotlib
• Set show_graph=false if you don't need it

Issue: Masks not affecting output

• Solution:
  • Check that masks are actually painted (not empty)
  • Verify mask is connected to correct input
  • Try increasing strength values
  • Enable show_debug=true to see what the node is processing

Issue: Regional prompts bleeding into each other

• Solution:
  • Make sure masks don't overlap
  • Use more specific prompts
  • Adjust region strength values

Issue: ControlNet effect too strong/weak everywhere

• Solution:
  • Adjust base_strength in Multi-Mask Combiner
  • Check strength value on Apply Advanced ControlNet node
  • Verify start_strength and end_strength values in Curved Timestep Keyframes

📋 Requirements

• ComfyUI
• ComfyUI-Advanced-ControlNet
• Python packages: matplotlib, pillow, numpy, torch

🤝 Contributing

Contributions are welcome! Feel free to:

• Report bugs
• Suggest new curve types
• Request features
• Submit pull requests

📄 License

MIT License - feel free to use and modify!

🙏 Credits

Created with assistance from Claude (Anthropic). Special thanks to the ComfyUI and Advanced ControlNet communities.

---

Enjoy creating with precise control! 🎨✨

If you find this useful, consider starring the repo and sharing your creations!