Film Portrait

Introduction

FilmPortrait is a LoRA model finetuned on FLUX.1-dev, designed to enhance film texture with a focus on low-saturation color palettes reminiscent of classic Japanese cinema. It offers a unique aesthetic experience by capturing characters, still lifes, and landscapes with a modern approach, subtly biased towards Asian features.

Architecture

The model builds on the FLUX.1-dev base, leveraging the capabilities of the diffusers library. It requires specific trigger words, such as "filmfotos," "film grain," and "reversal film photography," to generate images with the desired film texture effects.

Training

The model was trained by DynamicWang, utilizing a non-commercial license. It is optimized for generating images that evoke a sense of nostalgia through its refined portrayal of film textures.

Guide: Running Locally

1. Install Dependencies: Ensure that you have the necessary libraries installed, including torch and diffusers.

2. Model Setup: Use the following code to set up the model:

   import torch
   from diffusers import FluxPipeline
   
   pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)
   pipe.load_lora_weights('Shakker-Labs/FilmPortrait', weight_name='filmfotos.safetensors')
   pipe.fuse_lora(lora_scale=0.9)
   pipe.to("cuda")
   

3. Generate Images: Use the following code snippet to generate an image:

   prompt = "a young girl, filmfotos, film grain, reversal film photography"
   
   image = pipe(prompt, 
                num_inference_steps=24, 
                guidance_scale=3.5,
                width=768, height=1024,
               ).images[0]
   image.save(f"example.png")
   

4. Hardware Recommendations: For optimal performance, it is advisable to use a cloud-based GPU service such as AWS, Google Cloud, or Azure.

License

The model is distributed under the flux-1-dev-non-commercial-license. For more details, refer to the license link.