Project Goal: Fine-tuning the advanced diffusion model Stable Diffusion XL (SDXL) using the lightweight Low-Rank Adaptation (LoRA) architecture to deeply replicate the distinctive aesthetic of film photography, dramatic directional lighting, and urban melancholy. The series of works creates a unified visual narrative stylized after frames from independent auteur cinema (atmospheric/moody cinema).
Source Dataset
Parameter | Description and Characteristics
Dataset Source | Hugging Face Open Repository: akba08/ultra-
realistic-cinematic-photography
Legal Status / License | MIT License (free for non-commercial and
commercial use, modification, and
creation of derivative works)
Sample Size | 15 carefully selected representative
high-resolution images (in accordance with the
10–200 image protocol)
Data Preprocessing | Centered symmetrical cropping to a 1:1 aspect ratio, scaling
using the Lanczos algorithm to the native SDXL resolution of 1024×1024 px, conversion to RGB
Project Code
Images Obtained
