Flux Product Ad Backdrop LLM Model

Introduction

The Flux-Product-Ad-Backdrop is a text-to-image model designed for generating high-quality product advertisement images. Built using the diffusion model architecture, it features fine-tuning through LoRA (Low-Rank Adaptation) for specific use cases such as product advertising.

Architecture

Base Model: Utilizes the "black-forest-labs/FLUX.1-dev" as its core model.
LoRA Fine-Tuning: Enhances the model's capabilities to generate targeted advertisement images.
Optimizer: AdamW is employed for optimization.
Learning Rate Scheduler: Uses a constant scheduler.
Noise Parameters: Noise offset of 0.03 with multires noise discount of 0.1.

Training

Training Images: 19 images were used to train the model.
Image Dimensions: Best results are obtained with dimensions of 768x1024 and 1024x1024.
Epochs and Iterations: The model was trained over 15 epochs, with specific iterations for noise adjustments.
Labeling: Images were labeled using florence2-en for natural language processing.

Guide: Running Locally

Setup Environment: Install torch and the necessary Hugging Face pipelines.
```
import torch
from pipelines import DiffusionPipeline
```

Load Base Model:

base_model = "black-forest-labs/FLUX.1-dev"
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16)

Load LoRA Weights:

lora_repo = "prithivMLmods/Flux-Product-Ad-Backdrop"
trigger_word = "Product Ad"  
pipe.load_lora_weights(lora_repo)

Use GPU: Transfer the pipeline to a CUDA-enabled device for better performance.
```
device = torch.device("cuda")
pipe.to(device)
```
Cloud GPU Suggestion: Consider using cloud-based solutions like AWS EC2 with NVIDIA GPUs for efficient processing.

License

The model is released under the CreativeML OpenRAIL-M license, which allows for certain commercial uses while imposing restrictions on redistribution and derivative works.

More Related APIs in Text To Image