ComfyUI-Krea2Edit is an advanced image editing tool designed for use within ComfyUI, specifically tailored to enhance the Krea 2 model's capabilities. It allows for instruction-based, identity-preserving edits by leveraging dual conditioning, which ensures that both the visual and semantic aspects of the source image are effectively utilized during the editing process.
- Enables dual conditioning by injecting the source image into both the VAE latent tokens and the Qwen3-VL text encoder for improved semantic grounding.
- Features specialized nodes for model patching and grounded encoding, allowing for precise edits based on user instructions.
- Supports two-input edits, enabling complex transformations that can modify both subjects and scenes simultaneously.
Context
This tool serves as a node pack for the Krea 2 Identity Edit LoRA within ComfyUI, facilitating a streamlined workflow for image editing. Its primary purpose is to allow users to perform detailed edits on images while preserving the identity of subjects, making it particularly useful for tasks that require subtle modifications without losing the original characteristics.
Key Features & Benefits
The tool includes two essential nodes: Krea2EditModelPatch and Krea2EditGroundedEncode. The former enhances the diffusion model by incorporating the source image as clean in-context tokens, which is crucial for maintaining visual fidelity during edits. The latter enables the text encoder to process the image while interpreting user instructions, ensuring that edits are contextually relevant and accurate.
Advanced Functionalities
One of the standout features of this tool is its ability to handle two-input edits, allowing users to manipulate both a subject and a scene in a single workflow. Additionally, the ref_boost parameter allows for fine-tuning the fidelity of the reference image, offering users greater control over how closely the output adheres to the desired appearance.
Practical Benefits
By integrating these advanced functionalities, ComfyUI-Krea2Edit significantly enhances the editing workflow within ComfyUI. Users benefit from improved control over the editing process, higher quality outputs, and increased efficiency when making complex edits, all while ensuring that the identity of the subjects remains intact.
Credits/Acknowledgments
This project is licensed under Apache-2.0 for the nodes, while the Krea 2 Identity Edit weights are distributed under the Krea 2 Community License Agreement. Acknowledgments are given to contributors such as stablellama, CeciliaXCIX, akashzeno, SubtleShader, and Mark from sogni.ai for their support and contributions to the development of this tool.
