FLUX.2-Klein-LoRA-Studio

FLUX.2-Klein-LoRA-Studio is an experimental, high-performance image-to-image manipulation and style-transfer platform built on top of the black-forest-labs/FLUX.2-klein-9B model. The application integrates specialized LoRA adapters—such as Klein-Consistency, Klein-Delight-Style, Best-Face-Swap, and Ghost-Mannequin—to deliver precision edits including consistency-preserved transformations, lighting neutralizations, identity-locked face swaps, and 3D apparel volume creation.

The workspace is served via a single-page web app built with a FastAPI backend server (gradio.Server) and a dark-mode frontend interface featuring a dual-view canvas, A/B comparison slider, history filmstrip, and interactive prompt suggestions.

Key Features

• Lazy-Loaded Adapter Registry: On-demand downloading and weight-fusing for specialized FLUX.2 Klein LoRA adapters (e.g., Klein-Consistency, Klein-Delight-Style, Best-Face-Swap, and Ghost-Mannequin).
• Identity-Preserving Face Swap: Incorporates a dual-input configuration (Alissonerdx/BFS-Best-Face-Swap) designed to seamlessly combine the background environment and micro-expressions of a base image with the target facial structure of a source portrait.
• Text-Guided Image Editing: Enables style transformations, camera angle shifts, lighting neutralization, and apparel volume generation in 4-step sampling windows.
• Studio SPA Interface: An interactive single-page application built with modern vanilla web components—featuring an A/B image comparison slider, history filmstrip, quick prompt chips, and drag-and-drop file support.
• Smart Aspect Ratio Snapping: Automatically resizes uploaded images to stay within 1024px while snapping width and height to multiples of 16 to prevent shape mismatch errors during inference.

Repository Structure

├── examples/
│   ├── 1.jpg
│   ├── 2.jpg
│   ├── C.jpg
│   ├── cloth.jpg
│   ├── face-swap.jpg
│   ├── image.webp
│   ├── mc.png
│   └── Snow-Klein-consistency.png
├── qwenimage/
│   ├── __init__.py
│   ├── pipeline_qwenimage_edit_plus.py
│   ├── qwen_fa3_processor.py
│   └── transformer_qwenimage.py
├── app.py
├── index.html
├── LICENSE.txt
├── pre-requirements.txt
├── pyproject.toml
├── README.md
├── requirements.txt
└── uv.lock

Installation and Requirements

To set up the FLUX.2-Klein-LoRA-Studio environment locally, configure your system according to the specifications below. A modern CUDA-enabled GPU is required.

• Python Version: Minimum Python 3.12 is needed; Python 3.12 or 3.14 is recommended.
• PyTorch Version: torch==2.11.0 or above is required for best compatibility.
• CUDA Version: CUDA 13.0 is recommended (--extra-index-url [https://download.pytorch.org/whl/cu130](https://download.pytorch.org/whl/cu130)), matching the environment used on the live Hugging Face demo.

Running with uv (Recommended)

uv is an ultra-fast Python package and project manager written in Rust. It ensures rapid virtual environment setup and exact dependency synchronization based on the uv.lock file.

Step 1 — Install uv

• macOS / Linux: curl -LsSf https://astral.sh/uv/install.sh | sh
• Windows: powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

Step 2 — Clone the repository

git clone https://github.com/PRITHIVSAKTHIUR/FLUX.2-Klein-LoRA-Studio.git
cd FLUX.2-Klein-LoRA-Studio

Step 3 — Initialize the project and install dependencies

uv sync

Step 4 — Run the script

uv run app.py

Standard PIP Implementation

1. Update Package Manager Upgrade your local package manager:

pip install pip>=26.1.2

2. Install Core Dependencies Install the primary deep learning stack, transformer libraries, and core computing utilities listed in requirements.txt:

pip install -r requirements.txt

Core Requirements List (requirements.txt)

--extra-index-url https://download.pytorch.org/whl/cu130
torch==2.11.0
torchvision==0.26.0
transformers==5.14.1
accelerate==1.14.0
diffusers==0.39.0
peft==0.19.1
gradio==6.22.0
av==17.1.0
spaces==0.51.1
huggingface-hub==1.24.0
kernels==0.16.0

Usage

Once the web server initializes, open your browser to the local address output in your terminal (typically http://127.0.0.1:7860/).

1. Upload Asset: Drag and drop an image into the main canvas workspace, paste an image from your clipboard, or click the upload icon in the left rail.

• Note: Standard editing/consistency tasks use a single base image. Selecting Best-Face-Swap requires uploading two images: Image 1 acts as the base environment, and Image 2 acts as the facial identity source.

2. Select Style / LoRA: Choose your target editing task from the Style / LoRA dropdown menu in the right inspector panel. The adapter weights will download lazily on first use.
3. Refine Instructions: Type your instructions inside the prompt field, or click one of the Quick Prompts chips to instantly fill it. Press ⌘/Ctrl + Enter or click Edit Image.
4. Compare & Chain: Use the Compare tool on the left rail to view an A/B slider of the before and after states. Click Use as Input to chain multiple edits sequentially.

Links and Source

• GitHub Repository: https://github.com/PRITHIVSAKTHIUR/FLUX.2-Klein-LoRA-Studio
• Hugging Face Live Space: https://huggingface.co/spaces/prithivMLmods/FLUX.2-Klein-LoRA-Studio
• License: Apache License 2.0