I trained a LoRA to capture the nostalgic 90s / Y2K movie aesthetic. Go make your own Blockbuster-era film stills.
It's trained on stills from a bunch of my favorite films from that time. The goal wasn't to copy any single film, but to create a LoRA that can apply that entire cinematic mood to any generation.
You can use it to create cool character portraits, atmospheric scenes, or just give your images that nostalgic, analog feel
Settings i use: 50 steps, res2s + beta57, lora strength 1-1.2
P.S.: doesn't replicate actors
Description
FAQ
Comments (10)
Lora is very good. Did you erase or hide the characters in the image in white when you created the dataset? I'm a beginner, so I'm asking
No, I didn't. Anyway, with such a low number of steps per image, the model doesn't learn the faces.
And which models like illustriuns or Flux is it compatible with? Can it be used with a wide variety of SD models?
Sorry man, it's for Qwen only atm
a tip @babpeb11551 - on every single model page there's a section called Base Model that will list exactly which model this is for.
Oh yeah. Knew you were working on something. Can’t wait to get home .
I try to release content constantly. But if I don't release anything, it means I'm working hard on training. Waiting for your generations 😏
@Danrisi that motivates me to make more 😋😁
Wow, this is impressive. It really takes Qwen to a new level.
Great Work! How many images did you use here? And what type of caption? And did you crop them or use the native aspect ratio? Thanks !