TL;DR - strength around .6 to start, can go all the way up to 1.5 or so (with other LoRA's together, seems to do the trick as low as .15 at times), trigger is optional (Grit rush reality) but in short, it sometimes can tweak realism, grit, enhance prompt adherence and grade color a little differently than without. Seems to work best at the end of your prompt after the period/full stop for some real juice.Trained on 200 or so images that I manually collected, cropped, and heavily color graded with LUTs over the last year, then mixed and matched parts of the set over a few training runs, this LoRA is designed to do a few things, some more successfully than others (I need feedback!): <ul><li>Make more controversial and realistic topics accessible to Flux and increase context understanding without needing to ramble. It was trained on subjects like:<ul><li>Riots</li><li>Sex workers (not porn, but documentary like and real-life scenes)</li><li>Civil rights</li><li>Civil unrest</li><li>Poverty</li><li>Racism</li><li>Bleakness</li><li>Subcultures</li><li>Time periods</li></ul></li><li>Address the default Flux styling that sort of overrides a lot of things to do with diversity, both stylistically and with regard to subject appearance.</li><li>Make people look real, especially body types</li><li>Allow for more complex prompts with multiple subjects</li><li>Make things by default look like a photo with some color grading and grain and no need to prompt about it, but also retaining the ability to style and prompt away like normal.</li><li>Easy lighting control</li><li>Understand locations better</li></ul>In general, start at about .6 strength. Trigger is optional, but usually enhances lighting and realism a hair, but there is a bit of stylistic bias.Known issues to address if there is enough interest to warrant me spending another 20 hours on updating the dataset:<ul><li>Some stuff sometimes happens with floating limbs or fingers at higher strengths with close-up images that have multiple subjects interacting.</li><li>I went out of my way to not use NSFW pics in the dataset. This is not designed for it. Occasionally, if the situation you describe would have nudity, it may pop up if not careful</li><li>Increase understanding of subcultures and historical figures.</li></ul>Tips:<ul><li>See samples for examples of all of the above, including use of trigger and what it does.</li><li>Give me feedback please. I can only address it if I know and have motivation.</li></ul>I started making this because I was mad that I couldn't just call Lady Elaine Fairchilde (or anyone) a whore in prompts without explaining, so I trained something and noticed some stuff it was picking up, and went into this folder of gritty pics I've been amassing. Being a childish jackass makes me innovate :)

Files with SHA256: d39763ea57f594bda487f1846754d09fc769d401e0bb0d73b18a3a71f2b67c88