CliffNodes GitHub avatar

Krea2-Multi-Character-Lora-Node-w-bounding-box-By-Fedor

CliffNodes

A ComfyUI custom node for placing multiple character LoRAs in a single image with hard spatial masking to prevent bleeding, plus per-region reference image guidance for consistent identity across generations.

Stars

56

7-day growth

No data

Forks

6

Open issues

7

License

MIT

Last updated

2026-07-07

AI repository intelligence
FR-AI / ANALYSIS

Why it is worth attention

It solves two orthogonal problems—spatial bleeding between LoRAs and identity drift across seeds—using activation-delta masking (not attention bias) and a novel latent-mold guidance that works at the sampler level, all without modifying model weights.

Who it is for

  • ComfyUI users working with the Krea 2 model family
  • Character artists generating multi-character scenes with trained LoRAs
  • AI image generation enthusiasts needing consistent multi-character outputs across different seeds
  • Users who want bounding-box controlled layout (like Ideogram 4) combined with regional identity locking

Use cases

  • Generate an image with two or more characters each driven by its own LoRA, with no cross-contamination of features
  • Create a full scene layout using bounding boxes for objects, backgrounds, and characters, then lock identity only where needed
  • Produce a series of shots where a character's appearance stays consistent across different compositions and lighting
  • Use reference images to anchor a character’s identity without a LoRA, useful for props or untrained subjects

Strengths

  • Hard activation-delta masking makes it structurally impossible for a LoRA to affect any pixel outside its assigned box
  • Reference Lock (v3) provides per-region identity anchoring via latent-mold guidance, solving cross-seed identity drift
  • Supports unlimited regions with auto-syncing rows from a bounding-box builder, and works with fp8/quantized Krea 2 checkpoints
  • Full layout control: the same bounding boxes drive both positional prompts (for any object) and LoRA identity masking, all in one workflow

Considerations

  • Requires LoRAs trained specifically on Krea 2 (FLUX or SDXL LoRAs load but produce poor likeness)
  • Reference Lock ties both composition and identity together—cropping or early window end is needed to avoid full-pose copy
  • Needs higher LoRA strength (~1.5 or above) compared to full-image LoRAs, and canvas dimensions must match exactly between the prompt builder and the node

README quick start

Installation

Related repositories

Similar projects matched by category, topics, and programming language.

lopopolo
Featured
lopopolo GitHub avatar

harness-engineering

Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.

AI & Machine LearningAI Agents
2,390
slvDev
Featured
slvDev GitHub avatar

esp32-ai

A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.

AI & Machine LearningLarge Language Models
1,960
littledivy
Featured
littledivy GitHub avatar

mimic

mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.

AI & Machine Learning
1,482