Extensions/ComfyUI-Krea2-Projector-Tuner
ComfyUI Extension

ComfyUI-Krea2-Projector-Tuner

ComfyUI-Krea2-Projector-Tuner is a specialized custom node package for ComfyUI that allows you to patch and tune the layers of Krea2 models in-memory.

By BlackSnowSkill·Created 2 months ago·Updated about a month ago· 17
BlackSnowSkill/ComfyUI-Krea2-Projector-Tuner
Nodes1
On cloudLocal install
CategoryBSS/Krea2
Stars17
Updatedabout a month ago
Readme

ComfyUI-Krea2-Projector-Tuner (BSS)

🇺🇸 English | 🇷🇺 Читать на русском языке

ComfyUI-Krea2-Projector-Tuner (BSS) is a specialized custom node package for ComfyUI that allows you to patch and tune the txtfusion.projector layers of Krea2 models in-memory. It provides real-time slider controls and highly optimized presets (including the ultimate Strongx3_wl7 configuration) to control filter bypass (decensoring), style, and prompt-following without writing permanent files to disk.

[!IMPORTANT] Author and Developer: blacksnowskill (BSS)
© 2026 blacksnowskill (BSS). All rights reserved.
This project is protected by copyright. Any unauthorized copying, modification without attribution, or representing this code as your own product is strictly prohibited.


⚡ Key Features (v2.0.0)

  • In-Memory Dynamic Patching: Uses ComfyUI's native weight patching system. Modifies model weights temporarily in RAM/VRAM during generation, ensuring 100% safety with no risk of model file corruption on your storage.
  • Dynamic Presets System in JSON: Built-in presets are automatically written as separate .json files in the presets/ folder on the first start (strongx3_wl7.json, strong_2.5.json, clean_bypass.json). You can customize, rename, share, or delete them directly in your file explorer.
  • Interactive On-Node Buttons: Save, delete, and browse your custom weights configurations directly from the node interface (dynamic UI updates without reloading).
  • Automatic Web UI Sync: Mapped values update dynamically on your canvas sliders when selecting a preset. Dragging a slider automatically switches the preset dropdown back to custom.
  • LoRA-Style Step Increments (0.01): Smooth manual slider dragging with a standard 0.01 step size (identical to LoRA Loader), while fully preserving high-precision floating point inputs (up to 6 decimals) when typed or loaded from presets.

📥 Installation

Method 1: Via ComfyUI Manager (Recommended)

  1. Open ComfyUI and click on the Manager button.
  2. Click Install Custom Nodes.
  3. Search for Krea2 Projector Tuner.
  4. Click Install, wait for the process to complete, and restart ComfyUI.

Method 2: Via Git URL (If not indexed yet)

  1. Open ComfyUI Manager -> click Install via Git URL.
  2. Paste this repository URL: https://github.com/BlackSnowSkill/ComfyUI-Krea2-Projector-Tuner
  3. Click Install and restart ComfyUI.

Method 3: Manual Installation (Git Clone)

  1. Open your terminal in the ComfyUI/custom_nodes folder.
  2. Clone this repository:
    git clone https://github.com/BlackSnowSkill/ComfyUI-Krea2-Projector-Tuner.git
    
  3. Restart ComfyUI.

🧩 Node Specifications

🎛️ Krea2 Projector Tuner (BSS) (class Krea2ProjectorTuner)

Registered under the BSS/Krea2 category.

  • Inputs:
    • model: The loaded Krea2 model.
    • preset: Select from the loaded JSON files in presets/ folder, or custom (enables manual sliders).
    • multiplier: Overall weight multiplier (-10.0 to +10.0) with 0.01 step increments (default is 1.0). Multiplies all knob values, making it quick and easy to scale the total patch strength (e.g., 3x or 5x).
    • knob_0 to knob_11: Fine-tune multipliers (-5.0 to +5.0) with 0.01 step increments.

🎛️ On-Node Action Buttons

  • Save Preset: Prompt-saves your current manual slider values as a new JSON file in presets/ folder. The dropdown list refreshes instantly on the screen.
  • Delete Selected Preset: Deletes the currently selected preset file from disk and switches the dropdown back to custom.
  • Open Presets Folder: Instantly opens the presets directory in your OS file explorer.

📋 Understanding the Sliders (Knobs)

The 12 sliders correspond to the hidden states of the CLIP text encoder, processing different levels of representation:

  • Knobs 0 to 1 (Early Layers): Control syntax, text tokenization, and direct prompt-following.
  • Knobs 2 to 6 (Mid Layers): Control spatial geometry, objects, composition, and human proportions. Adjusting these acts as an anatomy stabilizer.
  • Knobs 7 to 10 (Deep Layers): Control abstract themes, styles, and negative alignment (safety filter / censorship). Suppressing these (moving negative values further down) bypasses the censorship.
  • Knob 11 (Final Layer): Represents the output layer of the CLIP text encoder.

📋 Recommended Workflow

Connect the node directly in the model output path:

[ Loader (Model) ] ──> [ Krea2 Projector Tuner (BSS) ] (Preset: Strongx3_wl7) ──> [ KSampler (model) ]

☕ Support and Development

If you like my optimization tools and want to support the development of new projects, nodes, and models, you can support me here:


📄 License & Terms of Use

© 2026 blacksnowskill (BSS). All rights reserved.
Any unauthorized mirroring, redistribution, merging with other projects, or re-uploading without explicit written permission from the author is strictly prohibited.