Configuring the K Sampler in ComfyUI

Step-by-step

1. Connect the Model, Positive Prompt, and Negative Prompt to the K Sampler node
2. Connect an Empty Latent Image node to the K Sampler
3. Set the Latent Image resolution to the native resolution of the model (e.g., 1024x1024 for SD 3.5)
4. Configure the Seed to generate the base noise layer
5. Adjust the Steps to define the number of denoising passes
6. Set the CFG (Classifier Free Guidance) value to control prompt adherence
7. Select a Sampler Name and Scheduler algorithm

Options

• Use a specific Seed number to attempt to recreate or maintain consistency across images
• Select alternative samplers like Euler Ancestral if required
• Add upscaler models after the K Sampler to reach higher resolutions like 1920x1080

Watch out for

• Models have specific latent image requirements based on their training; using non-native resolutions (like 1024x1024 on a 512x512 model) causes the generation to struggle
• Do not set the initial latent image to your final desired high resolution if it exceeds the model's native training size

Tips

• Keep Steps between 20 and 30; exceeding 50 is often a waste of computation and electricity with diminishing returns
• Use a low CFG value to make the model stick strictly to your prompt, or a high value to give the AI more creative leeway
• Stick with the 'Euler' sampler and default scheduler unless you have a specific technical reason to change them
• Hover over settings to view the built-in tooltips for more information
• Use the same seed and settings to achieve approximately 80% more consistency than random generation

Highlights

“the K sampler sort of been described as sort of the heart... of image generation.”