Quality-preserving per-layer conditioning control for Krea 2 in ComfyUI enables users to adjust the influence of 12 Qwen3-VL taps without introducing artifacts or likeness drift. This tool is a refined version of the original conditioning node, improving stability and output quality.
- Provides RMS-renormalized per-layer rebalancing to maintain consistent conditioning magnitude.
- Includes presets for different conditioning profiles, allowing users to quickly switch between settings.
- Compatible with any multi-layer tap conditioning, making it versatile for various diffusion models.
Context
This tool is designed to enhance the Krea 2 conditioning process within ComfyUI by allowing users to control the influence of individual layers in the text conditioning. Its main purpose is to prevent quality degradation that can occur when using a global multiplier, ensuring more stable and accurate outputs.
Key Features & Benefits
The primary feature of this node is its ability to preserve quality through RMS renormalization, which maintains the overall conditioning magnitude while rebalancing the ratios between layers. This functionality is critical because it allows users to avoid common issues such as artifacts and likeness drift that can arise from excessive conditioning magnitudes. Additionally, the inclusion of presets simplifies the adjustment process, catering to different artistic needs without requiring extensive manual configuration.
Advanced Functionalities
The tool features advanced capabilities such as the ability to customize per-layer weights through a "custom" preset option, which allows users to define their specific adjustments for each of the 12 taps. This flexibility is particularly useful for experienced users who want to fine-tune their outputs based on individual project requirements. The node is also structure-safe, meaning it integrates seamlessly with existing ComfyUI setups without altering the underlying data structure.
Practical Benefits
By utilizing this conditioning control node, users can significantly enhance their workflow efficiency and output quality in ComfyUI. The ability to maintain a consistent conditioning magnitude while adjusting layer influence leads to more predictable and visually appealing results. This tool ultimately allows for greater creative control, enabling artists to achieve their desired outcomes without the risk of introducing unwanted artifacts.
Credits/Acknowledgments
This tool is a fork of the original work by nova452, who developed the per-layer weighting technique in the ComfyUI-ConditioningKrea2Rebalance project. The current version builds upon that foundation by incorporating quality-preserving defaults and additional functionality. The project is licensed under Apache-2.0, with Krea 2 weights released under the Krea 2 Community License.
