Zum Inhalt springen
Node packÜberprüftExtern
Quelle öffnen

Comfyui-Latent-Transform-Sampler

Advanced latent space transformations during diffusion sampling

Betreuer@pizurnyBBewertung

Paketprofil

Status
Überprüft
Verleger
@pizurny
Quelle
GitHub
Sterne
3

README

ComfyUI Latent Transform Sampler

Advanced latent space transformations during diffusion sampling with precise control over timing and distribution.

Demo

Example use with 128 pixel shift at steps 3, 6 and 9

Features

  • Transform Types: shift, mirror (H/V), rotate (90°/180°), random, sequences, spiral
  • Distribution Strategies: even, golden ratio, fibonacci, front/back loaded, edges, center, exponential, logarithmic, random, manual
  • Advanced Controls: transform strength, accumulating shifts, reversible transforms, minimum spacing

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/pizurny/Comfyui-Latent-Transform-Sampler

Restart ComfyUI. Find the node under sampling/transform > Latent Transform Sampler

Requirements

  • ComfyUI (latest)
  • PyTorch >= 2.0.0
  • NumPy >= 1.19.0

Usage

Connect like a standard KSampler:

  • Model, positive/negative conditioning, latent_image inputs
  • Output to VAE Decode

Basic Example

transform_count: 3
transform_type: shift
distribution: golden_ratio
shift_pixels_x: 16
shift_pixels_y: 0

Parameters

Core Settings

  • transform_count (0-20): Number of transforms to apply
  • transform_type: Type of transformation
  • distribution: How to space transforms across steps
  • steps (1-10000): Total sampling steps

Transform Settings

  • shift_pixels_x/y (-512 to 512): Shift amount in pixels
  • transform_strength (0.0-1.0): Blend strength
  • accumulate_shifts: Whether shifts accumulate or reset

Advanced Options

  • min_spacing (1-10): Minimum steps between transforms
  • reverse_on_second_half: Undo transforms in second half
  • transform_sequence: Custom sequence (comma-separated)
  • manual_steps: Exact steps (comma-separated)
  • debug_mode: Console output for debugging

Transform Shorthand

For sequences:

  • s or shift - Shift transform
  • h or mh - Mirror horizontal
  • v or mv - Mirror vertical
  • cw - Rotate 90° clockwise
  • ccw - Rotate 90° counter-clockwise
  • 180 - Rotate 180°
  • r - Random

Example: "s,h,cw,s,v,ccw"

License

MIT License - see LICENSE file for details.

Knoten in diesem Paket

2 Knoten (Plural)
  • LatentTransformSampler
  • NTransformSampler

Quellen

3 Quellen
GitHubQuelleVersion bca5ebdbd5032298a6250d3fa4c84073ce5a2a07
github.com/pizurny/Comfyui-Latent-Transform-Sampler

Verifiziert 3. Okt.

Quellenauszüge

1 Auszug

Source context: Repo pizurny/Comfyui-Latent-Transform-Sampler