Workflow Name: [Consistency] Generate Character Material Set
[Workflow Introduction]
This workflow is extensive, and running it entirely on RH will consume 200 tokens at once. If you only want to train headshots, please disable the half-body and full-body parts of the workflow.
You need to upload 2 images: a portrait and an outfit image. Looking at this workflow from the far left:
Step 1: Perform face swapping using Apply PuLID Flux;
Step 2: Use Runninghub Captioner to reverse-generate prompt words for character appearance and outfit, then manually add some prompt words;
Step 3: Control skeleton shapes via Controlnet. I have embedded 3 images here: headshot, half-body, and full-body;
Step 4: Use SD to upscale the generated images, enhancing clarity and detail;
Step 5: Segment the upscaled images through this node;
Step 6: Since the LoRA training set requires specific materials, I processed the images with overlays to ensure the final output is in the correct orientation;
Final Step: Select an angle you like and use the Expression Editor node to create facial expressions, making the material set richer.
If you want to further process the images for lighting effects, I recommend switching to another workflow: IC Light Custom Lighting. URL: https://www.runninghub.cn/post/1856994862955483138

[Usage Scenarios]
If you want to generate a LoRA training set from a single image, you can try this workflow.

[Key Nodes] PuLID Flux, Controlnet

[Model Version]
FLUX
Model Name: fux1 dev Q8_0.safetensors

[LoRA Model]
None

[ControlNet Application]
FLUX.1 dev ControINet Union Pro Shakker Labs.safetensors
DW Pose Preprocessor
Start Time: 0
End Time: 0.85
Strength: 0.85

[Custom Sampler (Advanced)]
Sampling Method: deis
Scheduler: beta
Denoising: 1