gitmyhub

ComfyUI-KSampler-Matrix-Lab

Python ★ 29 updated 2d ago

A ComfyUI add-on that runs every combination of up to nine AI image samplers and nine schedulers and assembles all the results into one labeled grid image so you can compare them side by side.

PythonComfyUIsetup: easycomplexity 2/5

ComfyUI KSampler Matrix Lab is a custom add-on for ComfyUI, a visual tool people use to generate images with AI models. The add-on adds a single node you can drop into your workflow that runs multiple sampler and scheduler combinations back to back and then assembles all the results into one labeled grid image.

The problem it solves is comparison. When generating AI images, two key settings are the sampler (the algorithm that iteratively refines the image) and the scheduler (which controls how the noise is reduced across those steps). Different combinations produce different results, and figuring out which combination you prefer used to mean running each one separately and keeping track of which was which. This node automates that: you pick up to nine samplers and up to nine schedulers, it runs every pairing using the same model and the same settings, and returns a single image where columns are schedulers and rows are samplers.

Each cell in the grid is labeled with the sampler and scheduler it used. An optional header at the top shows the model name, VAE, CLIP, number of steps, CFG scale, and denoise strength. You can also choose between two seed modes: one that uses the same random seed for every cell so the comparison is purely about the algorithm, and one that increments the seed per cell to see variation across results.

Installation is done by cloning the repository into ComfyUI's custom_nodes folder and restarting the application. The node then appears in the node menu and connects to the same inputs as a standard generation workflow: model, conditioning, latent image, and VAE. There is an error-handling option that lets the grid continue generating even if one combination fails, inserting a placeholder cell instead of stopping everything. The project is released under the MIT license.

Where it fits