Helps render scenes that would fit the vice city nightlife, in the clubs and out on the streets. Trained on a mix of GTA VI preview material and actual Miami imagery.
Description
FAQ
Comments (3)
Hey! This looks really appealing, I'm going to play with it...
I've been meaning to do something a bit similar for "life & atmosphere in Paris": do you have advice or suggestions for the dataset or the training process?
A few. Feel free to hit me up in chat, short versions:
1: I use civitai model gen and found their "concept" settings more effective than the "style" settings. (I also set the Epochs to 30).
2: Don't overdo scope. Pick a handful of types of scenes (in my case, the "in the clubs and on the night streets" bit) and try to pick good, clear examples of the aesthetics you want your lora to give those scenes. (Splitting my dataset into "daylife" and "nightlife" was one way to control that scope. I could have gone further, separating the indoors and outdoors scenes.)
3: Also try to ensure the elements of said aesthetic have clear, consolidated tags that are at the forefront of the associated tags for each image. For me, I preprended tags like "indoors, club, outdoors," and then systematically checked each image to delete the ones that didn't apply there.
4: for Pony and (I believe) Illustruous, the healthiest thing you can do for the dataset is have a mix of visual styles and mediums- digital painting, photos, 3d, etc. That's when it's viable, of course.
@Thranatrax 🙏🙏🙏