Flux Isometric 3 D Lo R A

Introduction

The Flux-Isometric-3D-LoRA is a model designed for generating isometric 3D images using text prompts. It leverages the capabilities of the LoRA (Low-rank Adaptation) framework to enhance diffusion models for specific text-to-image generation tasks. This model is particularly focused on creating detailed and vibrant 3D scenes in an isometric view.

Architecture

The model is built upon the base model black-forest-labs/FLUX.1-dev and uses LoRA to adapt the diffusion process. Key parameters include:

• LR Scheduler: Constant
• Optimizer: AdamW
• Network Dim: 64
• Network Alpha: 32
• Epochs: 20

Training

The model is currently in the training phase, utilizing a total of 20 images. It is optimized using AdamW with a noise offset of 0.03. Training is conducted with multiresolution noise iterations and discounts to enhance image quality. The model is designed to save progress after every epoch, allowing iterative improvements.

Guide: Running Locally

To run the Flux-Isometric-3D-LoRA model locally, follow these steps:

1. Setup Environment: Ensure you have PyTorch installed with CUDA support.

2. Import Required Libraries:

   import torch
   from pipelines import DiffusionPipeline
   

3. Load Base Model and LoRA Weights:

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

4. Load LoRA Weights:

   lora_repo = "strangerzonehf/Flux-Isometric-3D-LoRA"
   pipe.load_lora_weights(lora_repo)
   

5. Set Device:

   device = torch.device("cuda")
   pipe.to(device)
   

6. Trigger Image Generation: Use the trigger word "Isometric 3D" in your text prompts to generate images.

For optimal performance, consider using cloud GPUs such as those offered by AWS, Google Cloud, or Azure.

License

The Flux-Isometric-3D-LoRA is distributed under the CreativeML Open RAIL-M License, which allows for open and responsible AI usage. Please refer to the license for specific terms and conditions.