Intended to create cyberpunk style environments and characters, trained on 1000 screenshots from Cyberpunk 2077. The images were auto tagged and run using CivitAI default settings.
This is just a first test run and I think it came out prety good. Im new to making loras and ai images so Im not really know what Im doing here just winging it. I have some ideas about how to improve the quality and get the lora to follow the prompt more closely and consistently but it will be a lot of work separating tagging the images by hand. Any advice on how to improve would be greatly appreciated.
Description
FAQ
Comments (5)
This is a really incredible direction. Really hope there's a v2 that nails the quality down just a little more
Thanks! Id like to improve this one and Ive got some ideas on how to do it but it will take some time.
@TVapor very interesting. Any update OP?
I too am a novice to making LoRas. I don't publish what I create. I can't handle rejection but I did just get through making an Emma Kotos Pony model with 410 HR pics at 1.25 repeats , training it once, which brought my steps up to 5125, 5 out of the 10 epochs were publishable. Took 4hrs to process. I used 3 sample prompts without supporting LoRas. Only one epoch successfully achieved a good picture in all 3 samples.
I don't know if that helps but it's all I can offer. Good luck. Also, I like your LoRa. I love the Blade Runner genre.
Thanks! Im reading up on training and Ill keep this in mind.