cliffnodes/krea2-multi-character-lora-node-w-bounding-box

This project is a custom node for ComfyUI, a visual tool for building AI image generation workflows. It lets you place multiple characters in a single generated image, where each character comes from a different LoRA, a small add on file that teaches an image model a specific look or identity. Normally, mixing several LoRAs in one image causes them to bleed into each other, blending faces or styles together. This node fixes that by letting you draw a box for each character on the canvas and locking each LoRA strictly to its own box, so one character's LoRA never affects another region of the image. Beyond characters, the same boxes can control other parts of a scene too, such as objects, props, or backgrounds, when paired with a compatible prompt style, letting someone lay out an entire image by drawing and describing regions. The newest version adds a second feature called Reference Lock. Even with boxes preventing bleed within a single image, a LoRA alone does not guarantee a character looks exactly the same across many separate generations, since faces can drift slightly from one image to the next. Reference Lock lets you attach a reference image to any box, and during the image generation process, that box is nudged step by step toward matching the reference, in addition to the LoRA masking. This works even for boxes with no LoRA at all, useful for props or backgrounds you have a reference photo for but no trained LoRA for. Settings control how strongly the reference pulls the image and during which part of the generation process it applies. The README also documents a recent fix for a type of LoRA file called LoKr, which previously failed silently and appeared to do nothing when loaded. It is now correctly recognized and applied. Older versions of the node remain available and unaffected by these changes, so existing workflows continue to work.