Using controlnet openpose to get the pose.

Use character lora, change details to suit your desired output, and same prompt to help consistency. Simple clothes is best for consistency.

Using ipadater to get consistent face. (Sometimes results are better if you bypass the ip-adapter.)

Using a fixed seed can help get more consistency too.

This workflow was based on: https://openart.ai/workflows/vole_quizzical_60/openpose-sdxl/wDpGhVwUiSzc2c06gYbk

Added in multi lora loader from Comfyroll Studio, for better lora management.

Added in more input images for the ip-adapter.

Added in the advanced apply controlnet tab, to be able to only run controlnet for the first 20%.

(If controlnet is allowed to run for the entire 100%, the output drops in quality and loses the style. Reduce the amount of steps helps just get the pose down, but then allows the prompt and ip-adapter to get the style down.)