Example Gallery

This is a gallery of compositions created with the help of artificial intelligence. Each character style and model was trained by me with the aim of achieving the highest fidelity possible. The images are the result of multiple studies, processes, and post-production edits in third party software.

ML Model Training

Each result was obtained by combining checkpoints. Additional tweaks were incorporated into the model to rectify color imperfections. The optimal approach involved identifying an appropriate camera setting to adjust the overall composition, along with a few embeddings. Additional networks, such as textual inversions and loras, were trained to transfer styles and characters. Each dataset underwent training with a minimum of 60 images, up to a maximum of 250. Working with an open-source platform like Stable Diffusion poses numerous challenges; most base models are plagued with content not suitable for work. It was necessary to utilize additional third-party plugins and extensions to circumvent such content. Training on large datasets often results in overtrained work, which I addressed by taking additional steps to ensure quality.