Flux.1 In-Context Learning

This repository contains experiments and training resources for Flux.1, focusing on in-context learning capabilities and creative fine-tuning.
It includes workflows, datasets, and example code for applying Flux.1 to custom domains such as pixel art characters.

---

🔥 Overview

• Base Model: black-forest-labs/FLUX.1-dev
• Adapter: LoRA (Low-Rank Adaptation) applied to attention layers
• Custom LoRA Model: milliyin/pixel_art_characters_lora_flux_nf4
• Goal: Extend Flux.1 beyond realism into stylized domains (e.g., pixel art, retro sprites, custom avatars)
• Training Hardware: NVIDIA A100

This repo demonstrates how Flux.1 can adapt to new artistic styles using in-context learning with small, curated datasets.

---

✨ Features

• LoRA training scripts for Flux.1
• Example dataset structures (image + caption pairs)
• Sample inference notebooks for text-to-image generation
• In-context editing & style transfer examples
• Integration with Hugging Face Diffusers

---

📦 Installation

Clone the repo:

git clone https://github.com/milliyin/Flux.1-In-Context-Learning.git
cd Flux.1-In-Context-Learning

Install dependencies:

pip install -r requirements.txt

---

🚀 Usage

Inference with Pixel-Art LoRA

import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev",
    torch_dtype=torch.float16
).to("cuda")

# Load LoRA weights from Hugging Face
pipe.load_lora_weights("milliyin/pixel_art_characters_lora_flux_nf4")

prompt = "pixel art RPG adventurer, green cloak, facing forward, clean outline"
negative = "blurry, low quality, distorted"

image = pipe(
    prompt=prompt,
    negative_prompt=negative,
    height=512,
    width=512,
    guidance_scale=7.5,
    num_inference_steps=50
).images[0]

image.save("sample.png")

---

📜 License

This project is released under CC BY-NC 4.0.
You may use it for research and personal projects, but commercial usage is not allowed.

---

📬 Contact

Created by Muhammad Illiyin (@milliyin)
For inquiries or collaborations: milliyin.vercel.app