🎭 Flux.2-Klein-9B Smart Character Swap

The ultimate LoRA for high-end, occlusion-aware character and face swapping.

Designed specifically for commercial imaging workflows, this adapter for FLUX.2 Klein 9B excels at handling complex occlusions (hands, veils, foreground objects) while seamlessly matching the exact color grading and directional lighting of your target scene.

---

🚀 MASTER COMFYUI & COMMERCIAL AI WORKFLOWS

Want to build production-ready AI pipelines? Subscribe to the channel for upcoming in-depth tutorials on how to maximize this LoRA, advanced ComfyUI node setups, GPU optimization, and building real-world AI applications (like automated AI photobooths).

@JettHuangAI — Hit 🔔 so you don't miss the deep-dive workflow tutorial for this model!

---

✨ Why Choose This LoRA?

Standard face-swap methods often fail in professional environments by pasting faces over objects or introducing uncanny, plastic looks. SmartCharcterswap solves this by acting as an intelligent compositor:

• Strictly High-End Realism: Engineered exclusively for practical photography and portrait retouching. It aggressively preserves realistic skin textures and structural anatomy while completely rejecting any cartoonish, illustrative, or overly-smoothed AI aesthetics.
• Masterful Occlusion Handling: Understands depth perfectly. It seamlessly integrates the new identity behind occluding elements like fingers, hair, or microphones without breaking the image.
• Perfect Lighting & Color Sync: Automatically adapts to the target's environment. If your base image is a high-contrast B&W portrait, the swapped identity will perfectly adopt those exact tonal values and shadows.
• Identity Preservation: Maintains the distinct bone structure and unique micro-features of the source subject across diverse angles.

---

🖼️ Visual Proof

Case 1 — B&W Color Matching & Hand Occlusion

Case 2 — Complex Occlusion Handling (Veil)

---

🛠️ Usage & Cfg Settings

This LoRA is optimized for node-based workflows, providing maximum stability for both standard and lightning-fast inference pipelines.

Trigger Word: jhuangswap

📋 Recommended Settings

1. Standard Base Model (High-Detail Inference)

• Inference Steps: 20 – 30
• Guidance Scale (CFG): 2.0 – 4.0
• LoRA Strength: 0.75 – 1.0

2. Distilled / Lightning Model (Ultra-Fast Inference)

• Inference Steps: 4 steps (Highly optimized)
• Guidance Scale (CFG): 1.0 (Strictly required for distilled/schnell base models)
• LoRA Strength: 0.85 – 1.0

---

💻 Integration Guide

ComfyUI Setup

1. Download the .safetensors file.
2. Place it in your ComfyUI/models/loras/ directory.
3. Add the LoraLoader node into your workflow and use the trigger word jhuangswap at the beginning of your positive prompt.

Diffusers (Python)

from diffusers import Flux2Pipeline
import torch

pipe = Flux2Pipeline.from_pretrained(
    "black-forest-labs/FLUX.2-klein-base-9B",
    torch_dtype=torch.bfloat16
).to("cuda")

pipe.load_lora_weights("nhathoangfoto/Flux.2-Klein-9B-SmartCharcterswap", adapter_name="smartcharswap")
pipe.set_adapters(["smartcharswap"], adapter_weights=[0.85])

# For Distilled Model usage:
# result = pipe(prompt="jhuangswap, realistic portrait...", num_inference_steps=4, guidance_scale=1.0)