cubiq/ComfyUI_IPAdapter_plus
ComfyUI IPAdapter Plus
ComfyUI_IPAdapter_plus 是 ComfyUI 的一个节点式扩展，它实现了 IPAdapter 模型以使用参考图像引导图像生成。它作为一个图像提示工具和 Stable Diffusion 图像适配器，允许参考文件作为视觉提示，用于控制风格、构图和主体身份。
该项目提供了在生成的肖像中保持面部身份和高保真特征的专门功能。它支持从参考图像迁移视觉特征和艺术风格，以及提取空间布局以引导新生成中对象的排列。
该扩展涵盖了广泛的功能领域，包括 AI 图像调节、一致的角色生成和图像构图控制。
Features
- ComfyUI Custom Node Suites - Provides a custom node suite for integrating IPAdapter models into ComfyUI workflows.
- Image-Prompted Generation - Implements a system that treats reference images as visual prompts to control style and composition.
- Cross-Attention Conditioning - Injects reference image embeddings into the UNet via cross-attention layers to guide the generation process.
- Image-Conditioned Generation - Enables the generation of new images using reference files as structural or stylistic baselines.
- Style Transfers - Applies the artistic style and visual characteristics of a reference image to control the generated output.
- Image Encoder Embedding Extractions - Uses pretrained CLIP vision models to extract numerical embedding representations from reference images.
- Stable Diffusion Ecosystem - Extends the Stable Diffusion ecosystem by providing image-based conditioning adapters.
- Generative Character Consistency - Ensures visual continuity of a person's identity across multiple high-fidelity generated portraits.
- Identity Consistency - Maintains high-fidelity facial features and subject identity across different generated portraits.
- Attention Masking - Implements attention masking to precisely control which regions of a generation are influenced by reference images.
- Compositional Layout Transfers - Extracts spatial layouts from reference images to guide the arrangement of objects in new generations.
- Latent Layout Mappings - Extracts spatial layout information from reference images to align the structure of the generated output.
- Image Composition Controls - Provides frameworks for managing the spatial layout and arrangement of subjects using reference images.
- Adapter Projection Layers - Provides linear projection layers to align image encoder outputs with the dimensionality of model attention layers.
- Weight Scaling Integrations - Merges IPAdapter weights into the base diffusion model by scaling layer parameters during the loading process.
- Portrait Identity Preservers - Provides a specialized implementation for maintaining facial identity and high-fidelity features in portraits.
