Logoiconemojimoe V0.2 FLUX.1 Dev LoRA

🚀 LogoIconEmojiMOE-v0.2-FLUX.1-dev-LoRA

This project utilizes a mixture of experts dataset to fine - tune a LoRA model for generating various logo, icon, and emoji images. It offers a unique way to create diverse visual content with specific styles.

🚀 Quick Start

This LoRA model is designed to generate images based on specific prompts. To get started, you need to use the trigger word LogoIconEmojiMOE in your prompts.

✨ Features

• Mixture of Experts Dataset: Utilizes a dataset of 100 unique logos, icons, and emojis, including Microsoft Fluent UI emojis and some synthetic image examples.
• Fine - tuned on NVIDIA Titan XP: The model was fine - tuned on an NVIDIA Titan XP with 12GB of VRAM using Fluxgym.
• Style - based Generation: By using the trigger word and referring to the training data's image annotations, you can generate images in specific styles.

📦 Installation

First, ensure you have FLUX.1 [dev] installed:

# Install
pip install -U diffusers

# Loading with Diffusers
import torch
from diffusers import FluxPipeline

# Load base model
pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.enable_model_cpu_offload()

# Load and apply LoRA weights
pipe.load_lora_weights("Borcherding/LogoIconEmojiMOE-v0.2-FLUX.1-dev-LoRA")

# Generate image
prompt = "HuggingFaceIcons a yellow smiley face sun with a book in its hands, surrounded by planets and stars in a dark background."
image = pipe(
    prompt,
    height=1024,
    width=1024,
    guidance_scale=3.5,
    num_inference_steps=50,
    max_sequence_length=512,
    generator=torch.Generator("cpu").manual_seed(0)
).images[0]
image.save("autumn-scene.png")

# Unload LoRA weights if needed
pipe.unload_lora_weights()

# Optional: Merge LoRA weights permanently
pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.load_lora_weights("Borcherding/LogoIconEmojiMOE-v0.2-FLUX.1-dev-LoRA")
pipe.merge_lora_weights()
pipe.save_pretrained("LogoIconEmojiMOE-v0.2-FLUX.1-dev-LoRA")

💻 Usage Examples

Basic Usage

To generate an image, use the following basic steps:

import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.enable_model_cpu_offload()
pipe.load_lora_weights("Borcherding/LogoIconEmojiMOE-v0.2-FLUX.1-dev-LoRA")

prompt = "LogoIconEmojiMOE, a simple blue circle icon"
image = pipe(prompt).images[0]
image.save("example_image.png")

Advanced Usage

For more advanced usage, you can adjust parameters such as height, width, guidance_scale, etc.

import torch
from diffusers import FluxPipeline

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
pipe.enable_model_cpu_offload()
pipe.load_lora_weights("Borcherding/LogoIconEmojiMOE-v0.2-FLUX.1-dev-LoRA")

prompt = "LogoIconEmojiMOE, a detailed red star icon with a golden border"
image = pipe(
    prompt,
    height=2048,
    width=2048,
    guidance_scale=5.0,
    num_inference_steps=80,
    max_sequence_length=512,
    generator=torch.Generator("cpu").manual_seed(42)
).images[0]
image.save("advanced_example_image.png")

📚 Documentation

Trigger Word

Use the trigger word "LogoIconEmojiMOE" in your prompts to activate the LoRA. For the best results, place the trigger word at the start of the prompt. You can refer directly to the image annotations in the training data for optimal prompting. The style variations, similar to the trigger word, will trigger the model to generate in that style type.

Here is the training data which provides a direct insight into the prompting techniques you will want to apply: LogoIconEmojiMOE

🔧 Technical Details

• Base Model: The base model is FLUX.1 [dev] by Black Forest Labs.
• Fine - tuning: The LoRA was fine - tuned on an NVIDIA Titan XP with 12GB of VRAM using Fluxgym.
• Dataset: A mixture of experts dataset of 100 unique logos, icons, and emojis was used for fine - tuning.

📄 License

This LoRA falls under the same licensing terms as FLUX.1 [dev]. Please refer to the base model's license for usage terms: flux - 1 - dev - non - commercial - license

Credits

• Base model: FLUX.1 [dev] by Black Forest Labs
• LoRA training and development: [Borcherding at BorchInk]

📦 Download model

Weights for this model are available in Safetensors format. Download them in the Files & versions tab.