ComfyUI-ReferenceLatentPlus Enables Per-Image Strength Dialing

ComfyUI-ReferenceLatentPlus is a custom node that completely replaces ComfyUI’s original ReferenceLatent tool. It gives creators per-image control over how reference images influence the final output during image and video generation. Instead of a simple on/off switch, this node lets you dial in strength, isolate specific subjects like faces or clothes, and manage up to four reference images in a single place.

Developer Shootthesound who recently released ComfyUI-Clippy-Reloaded built this drop-in replacement after hitting walls with the stock node. The standard tool only accepts one reference and provides no way to adjust its impact or timestep range. By adding MediaPipe auto-masks, per-image strength, and a megapixel cap, Shootthesound solved the most common frustrations encountered when trying to guide models like Flux, Klein, and Wan.

Per-image control and auto-masking

This node helps ComfyUI users who need fine-grained direction when guiding image models. You can attach reference images and automatically extract only the relevant subject, like a face while ignoring the background. The collapsible per-image controls let you set different rules for each input, which removes the need to chain multiple nodes together by hand.

Developer notes and known limits

The auto-mask feature relies on MediaPipe being installed; without it, ticking a region checkbox does nothing silently. The two advanced mask fill modes, latent_zero and latent_noise, work outside the model’s training distribution, so extreme mask coverage may produce unexpected results. Negative strength values are experimental and can somewhat push the output away from the reference image, though results remain hard to predict.

“The masking-of-ref-latents idea is novel. No other ComfyUI node does it (I checked).” — Source: GitHub