ComfyUI-semantic-aware-guidance
★ 12
语义感知Classifier-Free Guidance空间一致性修正ComfyUI节点
基于论文“Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance”的ComfyUI节点，提供语义感知的CFG改进，修正空间不一致性以增强局部语义对齐与细节控制。
💡 在图像生成流程中替代CFG以提升语义一致性与局部控制。
🍴 4 Forks💻 Python🔄 2024-08-08
https://pan.quark.cn/s/86a6deb1b5f6
📄 README
Semantic-aware Guidance (S-CFG)
ComfyUI node for Semantic-aware Guidance based on the paper “Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance”
Paper | Code | sd-webui-incantations (SD-WebUI extension)
Dynamically rescale CFG guidance per semantic region to a uniform level to improve image / text alignment.
Other details:
Computationally expensive. Large resolutions (like on an upscale pass) can trigger OOM.
made to work with other resolutions
made to work with SDXL (based on guesswork, since the original paper/code didn’t mention SDXL)
only works for models with a U-Net backbone (i.e. SD1.x – SDXL)
uses sampler_cfg_function (there can only be one node active using this, so it may conflict with other nodes using the same thing)
Installation
You can either:
In a termnial, run: git clone https://github.com/shiimizu/ComfyUI-semantic-aware-guidance.git into your ComfyUI/custom_nodes/ folder.
Install it via ComfyUI Manager (search for custom node named “s-cfg”).