Face-Swap LoRA for FLUX 2 Klein 9B

A LoRA adapter for FLUX 2 Klein 9B fine-tuned for high-quality face swapping. Given a face source image and a body/scene image, the model produces a natural face swap preserving identity, lighting, and skin tone.

Prompt

Use the following prompt for best results:

face_swap: take the face from Picture 1 and swap it onto the person in Picture 2. the result should show the exact face and features from Picture 1 on the body, pose, and scene from Picture 2, with natural skin tone blending and matching lighting.

• Picture 1: Face source image (the identity to transfer)
• Picture 2: Body/scene image (the target pose and environment)

How to Use

Try it on fal

Open on fal.ai

Via fal.ai API

import fal_client

def on_queue_update(update):
    if isinstance(update, fal_client.InProgress):
        for log in update.logs:
           print(log["message"])

result = fal_client.subscribe(
    "fal-ai/flux-2-klein/9b/base/edit/lora",
    arguments={
        "prompt": "face_swap: take the face from Picture 1 and swap it onto the person in Picture 2. "
                  "the result should show the exact face and features from Picture 1 on the body, pose, "
                  "and scene from Picture 2, with natural skin tone blending and matching lighting.",
        "model_name": None,
        "loras": [{
            "path": "https://huggingface.co/ilkerzgi/face-swap/resolve/main/FS2_V2_r32_800_pytorch_lora_weights.safetensors",
            "scale": 1
        }],
        "embeddings": [],
        "image_urls": [
            "https://example.com/face_source.png",
            "https://example.com/body_scene.png"
        ],
        "num_images": 4
    },
    with_logs=True,
    on_queue_update=on_queue_update,
)
print(result)

Evaluation Overview

9 LoRA variants were trained with different hyperparameters and evaluated against 3 image models across 25 diverse test subjects.

Training Configurations

Image Models

Latency Comparison

All LoRA variants average ~6-7s, while image models average ~29-33s.

Model Weights

Test Cases (25)

#1 — middle-aged woman with short stylish grey hair and blue eyes, elegant bone structure

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#2 — young woman with long straight brown hair parted in the middle and brown eyes

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#3 — older woman with short white curly hair and warm brown eyes

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#4 — young African American woman with braided hair and warm brown eyes

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#5 — young woman with short blonde bob haircut and bright blue eyes, Scandinavian features

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#6 — young Hispanic man with wavy dark hair and brown eyes, trimmed goatee

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#7 — young South Asian man with dark thick hair and brown eyes, short neat beard

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#8 — young man with a buzz cut and hazel eyes, tanned skin

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#9 — young Middle Eastern man with dark wavy hair and olive skin, prominent eyebrows

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#10 — young man with curly black hair and deep brown eyes, clean-shaven

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#11 — young African American woman with short natural TWA hairstyle and dark brown eyes, high cheekbones

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#12 — young woman with tight blonde curls, dark brown skin, bright smile, dimples

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#13 — young woman with tight blonde curls, dark brown skin, bright smile, dimples

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#14 — middle-aged woman with dark skin, close-cropped natural hair, elegant gold jewelry

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#15 — young Nordic man with platinum blonde hair, ice blue eyes, sharp cheekbones, clean shaven

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#16 — middle-aged Hispanic woman with dark wavy hair streaked with grey, warm brown eyes, round face

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#17 — young Pacific Islander woman with long wavy dark hair, brown eyes, strong features

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#18 — older Eastern European man with a thick grey mustache, lined face, piercing grey eyes

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#19 — older African American woman with silver locs, deep brown eyes, and laugh lines, regal posture

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#20 — older Chinese woman with jet black dyed hair, round face, gentle brown eyes, pearl earrings

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#21 — middle-aged woman with fiery red curly hair and freckled pale skin, green eyes, strong jawline

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#22 — young Southeast Asian man with black hair, smooth brown skin, sharp jawline

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#23 — middle-aged woman with natural grey afro, dark skin, reading glasses on nose

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#24 — middle-aged Hispanic woman with dark wavy hair streaked with grey, warm brown eyes, round face

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

#25 — older East Asian man with thinning grey hair, wire-rimmed glasses, kind wrinkled eyes

LoRA — 300 Samples

LoRA — 800 Samples

Image Models

Training Details

• Base Model: FLUX 2 Klein 9B
• Method: LoRA (Low-Rank Adaptation)
• Trainer: fal-ai/flux-2-klein-9b-base-trainer
• Datasets: Two custom face-swap paired datasets (300 and 800 samples)
• Rank variants tested: 16, 32, 64
• Training steps: 3,000 – 5,000
• Learning rates: 1e-5 to 5e-5
• Guidance scale: 1.0 and 3.5

Author

Ilker Izgi — Creative Engineer at fal