flux.1 lite 8 B alpha

Introduction

Flux.1 Lite is an 8B parameter transformer model, derived from the FLUX.1-dev model. It is designed for text-to-image generation while optimizing performance by using 7 GB less RAM and running 23% faster.

Architecture

Flux.1 Lite maintains the same precision (bfloat16) as the original FLUX.1-dev model. The model features a distilled architecture that reduces resource usage without compromising on image generation quality.

Training

The model's development involved analyzing the mean squared error (MSE) between input and output blocks. This analysis revealed that skipping certain early or late blocks impacts performance, while skipping blocks in between does not significantly affect image quality. These insights have guided the distillation process of the model.

Guide: Running Locally

To run Flux.1 Lite locally, follow these steps:

1. Install Dependencies: Ensure you have PyTorch and the diffusers library installed.

2. Prepare Environment: Use a CUDA-enabled device for optimal performance.

3. Load Model: Use the following code snippet to load and run the model:

   import torch
   from diffusers import FluxPipeline
   
   base_model_id = "Freepik/flux.1-lite-8B-alpha"
   torch_dtype = torch.bfloat16
   device = "cuda"
   
   pipe = FluxPipeline.from_pretrained(
       model_id, torch_dtype=torch_dtype
   ).to(device)
   
   prompt = "A close-up image of a green alien with fluorescent skin in the middle of a dark purple forest"
   guidance_scale = 3.5
   n_steps = 28
   seed = 11
   
   with torch.inference_mode():
       image = pipe(
           prompt=prompt,
           generator=torch.Generator(device="cpu").manual_seed(seed),
           num_inference_steps=n_steps,
           guidance_scale=guidance_scale,
           height=1024,
           width=1024,
       ).images[0]
   image.save("output.png")
   

4. Cloud GPU Recommendation: For enhanced performance, consider using cloud GPU services like AWS, Google Cloud, or Azure.

License

Flux.1 Lite is released under the FLUX.1 [dev] Non-Commercial License. All model weights and usage are subject to the terms outlined in this license. For detailed license information, visit the license page.