Workflow name: Train face LoRA-generate portrait
[Workflow introduction]
This process requires training the LoRA model based on the face image you uploaded, and combining the two groups of nodes InstantID and SUPIR to generate images.
The following is a detailed introduction:
Upload portrait image: First upload a portrait photo, preferably a frontal image without watermark.
RunningHub Portrait node training: Use the RunningHub Portrait node to train the model. The first training usually takes about 5 minutes to complete.
InstantID facial feature migration: After training, use InstantID to perform facial feature migration to extract facial features.
Local redraw optimization: Use the trained model to perform local redraw optimization on the facial image to improve facial details.
SUPIR texture upgrade: Finally, use SUPIR to upgrade and refine textures such as skin to further improve image quality and details.

[Use scenario]
Using the trained face generation model, you can apply your facial features to different environments and wear different styles of clothing. This process combines face synthesis and image processing technology to allow you to see yourself in different ways in virtual scenes. With this method, you can try to show your facial features in various backgrounds and explore how different styles of clothing combine with your facial features.

[Key nodes]
RunningHub Portrait, apply InstantlD, SUPIR,

[Model version]
SDXL
Model name: dreamshaperXL v21TuroDPMSDE.safctensors

[LoRA model]
None

[ControlNet application]
None

[K sampler]
CFG: 2
Sampling method: dpmpp_sde
Scheduler: karras
Noise reduction: 1