ComfyUI实现老照片修复

环境：Ubuntu 22.04.1 LTS + NVIDIA GeForce RTX 3090

工作流下载地址：https://openart.ai/workflows/datou/old-photo-restoration-xl/vKtihSX1DGBiBjAovlSP

在ComfyUI中打开下载好的工作流，根据提示安装缺失的节点

下载BOPBTL模型文件

链接：https://github.com/microsoft/Bringing-Old-Photos-Back-to-Life/releases/download/v1.0/global_checkpoints.zip

解压后

• restoration/VAE_A_quality/latest_net_G.pth
• restoration/VAE_B_quality/latest_net_G.pth
• restoration/VAE_B_scratch/latest_net_G.pth

放在models/vae/目录下

• restoration/mapping_quality/latest_net_mapping_net.pth
• restoration/mapping_scratch/latest_net_mapping_net.pth
• restoration/mapping_Patch_Attention/latest_net_mapping_net.pth
• detection/FT_Epoch_latest.pt

放在models/checkpoints/目录下

下载controlnet模型

https://huggingface.co/xinsir/controlnet-depth-sdxl-1.0/resolve/main/diffusion_pytorch_model.safetensors?download=true

下载后修改文件名为xinsir-controlnet-depth-sdxl-1.0.safetensors

下载inswapper_128.onnx文件

https://github.com/facefusion/facefusion-assets/releases/download/models/inswapper_128.onnx

放在ComfyUI\models\insightface目录下

下载https://github.com/xinntao/facexlib/releases/download/v0.1.0/detection_Resnet50_Final.pth放在ComfyUI/models/facedetection目录下

出现报错

在系统执行 curl -fsSL https://ollama.com/install.sh | sh 安装ollama服务

再执行 ollama run phi3 下载phi3模型

填入模型名称

出图成功