Workflow name: Adding a character model to the scene
[Workflow introduction]
This is a workflow for adding a character model to the scene. There are still many areas that need to be improved. Suggestions are welcome.
After opening the workflow, first upload two images in the lower left corner of the screen, one for the scene and the other for the character. Draw a mask in the scene to determine the approximate position of the character. Then process the mask of the image, and use SegmentAnythingUltra V2 for keying many times during this process. Then, process the prompt words through the Runninghub Captioner and Midjourney Style Prompter nodes. If you want to use it locally, you can replace it with other reverse word nodes. Next, come to the first sampler, first use the kolors model for partial redrawing. I personally prefer to use the kolors model for partial redrawing. If you have other better options, you can replace it. Then use the sdxl model for local refinement, then redraw with the flux model, replace the background elements and refine the image, finally optimize the face, and then use SUPIR for overall high-definition zoom processing.

[Usage scenario]
This workflow can add character models to the scene, starting from uploading scene images and character images, and gradually performing a series of operations such as drawing masks, matting, and prompt word processing. In this way, the characters can be cleverly integrated into the scene naturally. It really saves the tedious process of hiring models, brings great convenience to work, and greatly improves work efficiency, allowing creators to focus more on creativity.

[Key nodes]
SegmentAnythingUltra V2, Runninghub Captioner, Midjourney Style Prompter

[Model version]
FLUX
Model name: flux1-dev-fp8-Comfy-Org.safetensors
kolors
Model name: Kolors-Inpainting.safetensors
SDXL
Model name: realvisxlV50 v50Bakedvae.safetensors

[LoRA model]
Model name: xlab flux Realism2 lora.safetensors

[ControlNet application]
None

[Custom sampler (advanced)]
Sampling method: euler
Scheduler: simple
Noise reduction: 0.7