ComfyUI tutorial: switch and compare image refiners

This ComfyUI tutorial builds a reusable image refiner section with the Lenient Switch custom nodes. The Simple Selector provides a dropdown with custom labels and up to five inputs. The presenter connects three selectors to one combo box so a single choice selects the model and its matching CLIP and VAE.

Lenient Switch handles conditional routing separately. In the example, a text replacement turns the Qwen Image label into an empty string. With empty strings treated as false, the switch selects the original Qwen conditioning instead of the other source. The presenter also uses a vision model to describe the generated image, replacing an editing instruction with a prompt suited to a non-editing refiner.

The comparisons cover fantasy artwork, a photographic portrait and a masked cyborg edit. The presenter tests Ernie, Microsoft Lens and Z Image, with an optional Qwen finishing pass. In these examples, that final pass can repair lettering or recover some facial likeness, but it can also smooth away detail. Higher denoise settings produce larger changes, including changes to faces and hands, so the results illustrate tradeoffs rather than a universal ranking.

The final example copies the refiner section into a HiDream reference-image workflow. This open source workflow tutorial focuses on routing and visual comparisons rather than installation.