LORA Models and SID LORAS

Glamourshotz, is a AI image augmentation startup that, has been exploring innovative ( and fast ) techniques for image generation. The app focuses on creating vintage-style photos of individuals from different decades using a combination of Single Image Dataset (SID) Loras and style Loras.

SID Loras are a type of low-rank adaptation (LoRA) that allows for the generation of multiple variations of an individual’s photo from a single input image. By using SID Loras, Glamourshotz can create diverse images of a person with different characteristics and styles.

To add a vintage touch to the generated photos, Glamourshotz employs style Loras. These Loras are trained on datasets of photos from the 60s, 70s, and 80s, capturing the distinctive visual styles of each era. By training separate style Loras for each decade, Glamourshotz can generate images that resemble photographs from those specific time periods.

The process of generating the vintage-style photos involves running inference with both the SID Loras and the style Loras simultaneously. This combination allows Glamourshotz to create multiple photos of an individual that appear as if they were taken in different decades. The generated images exhibit the color palettes, fashion styles, and photographic characteristics of each era while maintaining the individual’s likeness.

We process an save a 1 shot model then execute an inference using both the created LORA model and one of the LORA style models that we previosly created by fin tuning on LORA per decade, on a variety of different image styles.

Here is the original Image

The strength to this approach is that we are able to quickly generate image on the fly without an long wait times or pre-training