Cyberpunk World
This is a cyberpunk style LoRA. That's it.
Unlike a certain game, this LoRA doesn't overpromise and underdeliver.
More like seedy project red, right?
Why 2 Cyberpunk models?
I also have a "Cyberpunk Style" model. They are both trained from the same data. The only difference is the trigger keyword. This one uses the actual word of "Cyberpunk Style" in it, so it "inherits" information about this from the base model you're using. The other one uses a made-up word which should make the training more pure and it should inherit less knowledge about what cyberpunk means, outside of the training data. In theory at least. Both models have the same prompt and seed as their showcase images, so you can compare and pick your favorite, or use them both for all I care! I'm not your parent!
It may be that this model is slightly better at making humans. Neither LoRA has ever seen a human or a cyberpunk human, but if this model's training data is merged with "generic" cyberpunk human training data, there may be more human-skinned people in this one.
There's also an older "0.5-version" of this model. This one is fully trained, but the results are a bit less colorful and a bit more foggy/washed out with some models. I ended up re-training, but this version is still good, so I decided to share it as well.
Usage
Use the trigger word: "CyberpunkWorld" along with the LoRA to activate it.
Example prompt: CyberpunkWorld computer mouse <lora:CyberpunkWorld:1>
I've found that most settings work well with this model. No special samplers or CFG-scale.
Tips and tricks
Here's a list of keywords you can use to enhance the different effects of the model.
cyberpunk style
colorful
glowing
color explosion
hi-tech
sci-fiLike it?
Description
This was an earlier test version of the model. It turned out too dull and lacked contrast, so I increased it on most of the training images for better results.
In case someone wants to try this version, I figured I'd post it.
With some photorealistic models, it still looks very good!
FAQ
Comments (8)
Thx for this beauty LoRA.
I just checked your Training Data and it does not contain caption data.
Did you train it without using captions?
Cheers.
No it's trained by the names of the folder.
There's a script from KonyConi in one of their older posts (somewhere): https://civitai.com/user/konyconi/models
This script basically adds a caption to each image, with the name of the folder the image was in.
Do you want the captions or were you just asking out of curiosity? I can upload them somewhere if you want them.
@mnemic Hi , thx a lot for explaining your workflow which is very interesting.
Currently i am in learning phase and so far i always use a pic and detailed discription of each pic as captions one by one.
Seeing your LoRA Training Data and how you explain you use folders with specific categories and you name the folders matching instead using one by one pic discription captions is really something new to me.
I haven't seen any guide so far using this method.
And the results you show using this method are a blast :)
Can i use this with Kohya or do i have to use a training tool other than Kohya?
@LVNDSCAPE You can totally use Kohya. I do.
I wrote this article (https://civitai.com/articles/2138/lora-training-data-upload-to-civitai)
going over and listing a lot of models with training data accessible. Including captions and images.
Some of my models have the captions included, and some have the images sorted in folders instead.
When it comes to my shared data, if it has a lot of subfolders, like "car", "toaster" or "landscape", this means that this folder name is used as the caption name.
So for example, in this style model, the name is CyberpunkWorld. And this caption is used for each image in the set. On top of this word, it will also use the word from the folder each image is in.
And I use the script I mentioned to do that.
@mnemic Hi, thx again so much for your tips & hints :) I guess it is this tutorial?
https://civitai.com/models/52697?modelVersionId=58720
I saved all informations now in sorted folders and will start my dojo :)
@LVNDSCAPE Yes, it's that guide.
You will find that it has multiple "files" at the top. The third one is called "Basic Captiniong Script". You can use this to auto-caption stuff. Here's a link to it:
https://civitai.com/models/52697?modelVersionId=57118
Looks cool. Can't say I understood your explanation about the differences between 0.5 and 1.0, grabbed the latter regardless.
Edit: Ah, I see you have another model called Cyberpunk Style, now it makes sense.
Ah, yes, sorry for the confusion.
I'll update the explanation.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.