FlowMatch Euler Discrete Scheduler for ComfyUI

Custom node that exposes all parameters of the FlowMatchEulerDiscreteScheduler. Outputs SIGMAS for use with SamplerCustom node.

Usage

1. Add FlowMatch Euler Discrete Scheduler (Custom) node to your workflow
2. Connect its SIGMAS output to SamplerCustom node's sigmas input
3. Adjust parameters to control the sampling behavior

Parameters

steps

Type: Integer (1-10000, default: 20)
Description: Number of diffusion steps during sampling.
Example: Use 20-30 for fast previews, 40-50 for quality results.

base_image_seq_len

Type: Integer (default: 256)
Description: Base image sequence length for dynamic shifting calculations.
Example: For 512x512 images, use 256. Matches model's training resolution.

base_shift

Type: Float (default: 1.0986 ≈ log(3))
Description: Stabilizes generation by reducing variation. Higher = more stable/consistent.
Example: Increase to 1.5 for more controlled, predictable outputs.

max_shift

Type: Float (default: 1.0986 ≈ log(3))
Description: Maximum variation allowed. Higher = more exaggerated/stylized results.
Example: Increase to 1.5 for more creative/varied outputs.

shift

Type: Float (default: 1.0)
Description: Global timestep schedule shift value. Affects overall sampling behavior.
Example: Keep at 1.0 unless you understand timestep shifting theory.

shift_terminal

Type: Float (default: 0.0, which means None)
Description: End value for shifted timestep schedule. Set to 0.0 to disable.
Example: Use 0.5 to modify how the schedule ends (advanced).

use_dynamic_shifting

Type: Boolean (default: True)
Description: Automatically adjust timesteps based on image resolution.
Example: Keep True for better results with varying image sizes.

time_shift_type

Type: Choice: "exponential" or "linear" (default: "exponential")
Description: Method for resolution-dependent timestep shifting.
Example: Use "exponential" for most cases, "linear" for experimental control.

use_karras_sigmas

Type: Boolean (default: False)
Description: Use Karras noise schedule (typically gives smoother results).
Example: Enable for potentially higher quality, similar to DPM++ samplers.

use_exponential_sigmas

Type: Boolean (default: False)
Description: Use exponential sigma spacing instead of linear.
Example: Try True for different noise distribution characteristics.

use_beta_sigmas

Type: Boolean (default: False)
Description: Use beta distribution for sigma values.
Example: Enable for alternative noise scheduling (experimental).

invert_sigmas

Type: Boolean (default: False)
Description: Reverse the sigma schedule direction.
Example: Keep False unless you have specific experimental needs.

stochastic_sampling

Type: Boolean (default: False)
Description: Add controlled randomness to each sampling step.
Example: Enable for more varied outputs, similar to ancestral samplers.

num_train_timesteps

Type: Integer (default: 1000)
Description: Number of timesteps the model was trained with.
Example: Match your model's training config (usually 1000).

Tips

• Start simple: Use defaults, only adjust steps initially
• For quality: Try enabling use_karras_sigmas with 30-40 steps
• For variation: Enable stochastic_sampling or increase max_shift
• For stability: Increase base_shift to reduce randomness

Example Workflow

Model -> SamplerCustom
         ^ 
         |
         FlowMatch Euler Scheduler (steps=30, use_karras_sigmas=True)