ComfyUI advanced nodes: sampling and inpainting tutorial

This ComfyUI tutorial explains node workflows for image generation and video processing. The speaker demonstrates Detail Demon with a detail amount of 0.8 in an eight-step Z Image Turbo workflow. It requires a sampler that accepts a sampler input, and its start and end controls determine when it affects generation.

RIFE interpolates between video frames to smooth motion. Reference Latent adds an encoded image to prompt conditioning, but the speaker notes that only some models accept this input.

The KSampler Advanced example splits a 20-step generation between two samplers. The speaker explains how leftover noise lets the second sampler continue the first sampler's work, potentially with a different model. Inpaint Crop and Inpaint Stitch isolate a masked region for editing and place the result back into the original image. The example uses a 1024 by 1024 crop with a 1.2 context expansion factor.

String Multiline and Concatenate assemble text with a chosen delimiter. The core concatenate node accepts two strings; chaining nodes allows more inputs. The all-in-one AUX preprocessor provides access to depth, line art and other preprocessors, but omits some controls, including Canny thresholds.

The final example uses Conditioning Zero Out for workflows that do not use negative prompts. The speaker describes a small resource benefit, especially on the first run, rather than a measured performance gain.