A drop-in replacement for ComfyUI's stock ReferenceLatent node with per-image strength, per-image timestep gating, per-image MediaPipe auto-masks, a megapixel cap on attention cost, and 1–4 image inputs in a single node. Works with Flux, Flux2/Klein, Lumina2, Z-Image, Wan, Hunyuan, and Qwen-Image.
https://github.com/shootthesound/comfyui-ReferenceLatentPlus
git clone https://github.com/shootthesound/comfyui-ReferenceLatentPlus
comfy node install comfyui-reference-latent-plus