7.3 KiB
7.3 KiB
Scheduler Select Helper
Overview
The Scheduler Select Helper node provides intelligent scheduler selection with sampler-aware recommendations and model-specific optimizations. Adapted from comfyui-essentials-nodes (now in maintenance mode), this tool ensures optimal scheduler selection for different sampling algorithms and models.
Attribution
This node is based on work from comfyui-essentials-nodes by cubiq. The original project is in maintenance-only mode, and we've adopted and enhanced these tools to ensure continued support and compatibility with modern ComfyUI workflows.
Features
- Sampler-Aware Selection: Recommends best schedulers for each sampler
- Model Optimization: Specific scheduler tuning for different models
- Noise Schedule Profiles: Pre-configured curves for various use cases
- Visual Feedback: Preview noise schedules
- Batch Testing: Compare multiple schedulers
Node Properties
- Category:
ComfyAssets/🧰 xyz-helpers - Node Name:
SchedulerSelectHelper - Function:
select_scheduler
Inputs
Required
| Parameter | Type | Default | Description |
|---|---|---|---|
sampler_name |
STRING | - | Current sampler being used |
model_type |
DROPDOWN | auto | [auto, sdxl, sd15, flux] |
schedule_type |
DROPDOWN | smooth | [smooth, sharp, linear, custom] |
Optional
| Parameter | Type | Default | Description |
|---|---|---|---|
override |
DROPDOWN | none | Force specific scheduler |
beta_schedule |
STRING | - | Custom beta schedule values |
visualize |
BOOLEAN | False | Show schedule curve |
Outputs
| Name | Type | Description |
|---|---|---|
scheduler |
STRING | Selected scheduler name |
schedule_curve |
IMAGE | Visualization of noise schedule |
beta_values |
FLOAT_ARRAY | Beta schedule values |
Scheduler Types Explained
Normal
- Curve: Linear noise reduction
- Best For: General purpose
- Samplers: euler, dpm_fast
Karras
- Curve: Improved noise schedule
- Best For: High quality
- Samplers: dpmpp_2m, dpmpp_2m_sde
Exponential
- Curve: Exponential decay
- Best For: Fine details
- Samplers: dpmpp_3m_sde
Simple
- Curve: Basic linear
- Best For: Fast generation
- Samplers: euler, lcm
SGM Uniform
- Curve: Uniform distribution
- Best For: FLUX models
- Samplers: euler, dpmpp_2m
Schedule Types
Smooth (Default)
# Gradual noise reduction
# Good for most content
→ karras or exponential
Sharp
# Aggressive early reduction
# Good for high contrast
→ normal or simple
Linear
# Constant reduction rate
# Predictable results
→ normal
Custom
# User-defined curve
# Advanced control
→ based on beta_schedule
Usage Examples
Automatic Selection
KSampler Settings → SchedulerSelectHelper → KSampler
sampler_name: "dpmpp_2m_sde"
model_type: auto
→ scheduler: "karras"
Visual Comparison
SchedulerSelectHelper → Display
visualize: True
→ Shows noise schedule curve
Batch Testing
For each scheduler:
SchedulerSelectHelper → KSampler → Save
→ Compare results
Sampler-Scheduler Compatibility
Optimal Pairings
| Sampler | Best Scheduler | Good Alternatives |
|---|---|---|
| euler | normal | karras |
| euler_a | karras | normal |
| heun | normal | - |
| dpm_fast | normal | simple |
| dpm_adaptive | normal | - |
| dpmpp_2m | karras | exponential |
| dpmpp_2m_sde | karras | exponential |
| dpmpp_3m_sde | exponential | karras |
| dpmpp_2s_a | karras | normal |
| lcm | simple | normal |
Model-Specific Recommendations
SDXL
preferred_schedulers = ["karras", "exponential"]
# Better convergence for high-res
SD 1.5
preferred_schedulers = ["karras", "normal"]
# Classic combinations
FLUX
preferred_schedulers = ["simple", "sgm_uniform"]
# Optimized for FLUX architecture
Best Practices
Selection Strategy
- Let auto-detection handle defaults
- Override for specific artistic goals
- Test multiple schedulers for hero images
- Use visualization to understand curves
Performance Tips
- Simple/normal for quick previews
- Karras/exponential for quality
- SGM uniform specifically for FLUX
- Match scheduler to sampler type
Testing Workflow
schedulers = ["normal", "karras", "exponential"]
for scheduler in schedulers:
generate_image(scheduler)
save_with_metadata(scheduler)
compare_results()
Advanced Features
Beta Schedule Customization
# Custom exponential curve
beta_schedule = "0.00085, 0.0012, 0.0018, ..."
# Sharp early reduction
beta_schedule = "0.001, 0.002, 0.004, 0.006, ..."
Schedule Visualization
- Plots noise reduction curve
- Shows sigma values
- Compares with standard schedules
- Exports schedule data
Adaptive Selection
- Learns from user preferences
- Adapts to hardware capabilities
- Optimizes for generation speed
Integration Examples
Complete Pipeline
Sampler Combo → SchedulerSelectHelper → KSampler
↓ ↓
sampler_name → Optimal scheduler selection
A/B Testing
Same prompt → Different schedulers → Grid comparison
normal vs karras vs exponential
Noise Schedule Analysis
SchedulerSelectHelper → Plot Parameters
visualize: True
→ Analyze noise curves
Tips and Tricks
Quality Optimization
# For maximum quality
if sampler in ["dpmpp_3m_sde"]:
use scheduler="exponential"
elif sampler in ["dpmpp_2m_sde"]:
use scheduler="karras"
Speed Optimization
# For fast generation
use scheduler="simple" or "normal"
reduce step count by 20%
Artistic Effects
- Sharp details: normal scheduler
- Smooth gradients: karras scheduler
- Fine textures: exponential scheduler
Troubleshooting
Artifacts or Noise
- Try different scheduler
- Check sampler compatibility
- Adjust step count
Slow Convergence
- Switch from simple to karras
- Increase step count
- Check model compatibility
Inconsistent Results
- Use same scheduler for batch
- Avoid random scheduler selection
- Fix seed for testing
Visual Guide
Noise Schedule Curves
Normal: ████████████████
Linear reduction
Karras: ███████████▓▓▓░░
Smooth curve
Exponential: ██████▓▓▓░░░░░
Fast early reduction
Common Workflows
Scheduler Comparison
Test same seed with different schedulers to find optimal setting.
Model Migration
When switching models, automatically adjust scheduler for best results.
Quality Ladder
Progress through schedulers from fast to quality for different use cases.
Version History
- 1.0.0: Initial adaptation from comfyui-essentials-nodes
- 1.0.1: Added visualization features
- 1.0.2: Enhanced model detection
- 1.0.3: Improved compatibility matrix
Credits
Original implementation by cubiq in comfyui-essentials-nodes. Adapted and maintained by the ComfyAssets team.