Introduction

Hi, this is an experimental LoRA that I've trained using Google Colab and the AnyLora model. The name for this LoRA model is a twist on Logan Paul's "99 Originals", where he took 99 Polaroids over 99 days. The difference is I only used 30 images for the dataset. The whole purpose of making this LoRA was so that I have a more consistent generation of Polaroid-framed images that could be useful for someone else.

I used 30 well-detailed, quality images and captioned them (to the best of my ability, haha) that are of different Polaroids and that have different Polaroid frames/text. (Yes, I know the text is random when you generate; I wanted to keep the aesthetic despite it being random nonsense).

Example Image:

Prompt:
<lora:Unoriginal_88-10:0.8>, Overgrown Forrest, (Collapsed Tree in foreground), (Beautiful Business Man in a suit holding a human skull in his hand and smiling:1.2), (foggy), full Moon in the distance, Night time, Dynamic Lighting, smudged, foggy, vintage, old, realistic, sharp focus
Negative prompt: bad_prompt_version2:0.8, cartoon, cartoonish, lowres, blurry, low quality, bad body anatomy, letterbox, amputee, ((watermark)), (((text))), signature, badly drawn, jpeg artifacts, ((logo)), out of frame, incoherent, cartoon, childish, error, worst quality, username, artist name
Steps: 30, Sampler: Euler a, CFG scale: 7, Seed: 2613144170, Face restoration: CodeFormer, Size: 512x512, Model hash: 9aba26abdf, Model: deliberate_v2

How to use and Install?

Download the LoRA and put it into your models/lora folder

When prompting put this towards the front of it "<lora:Unoriginal_88-10:0.8>".

Recommended weights are 0.8-1.0, but just experiment.

Note I've only done a lot of testing with text2img. I do believe it struggles a bit when doing img2img, so like I say experiment with the weight of the lora. As I learn more, I will try my best to improve this.