Boobsgames Style Illustrious V2.5

Boobsgames Lora Style v2.5 (FINAL)

Hello there!

This is my first trained and published lora.
I hope that you find it interesting and fun to play around with.

This Style Lora was trained on a dataset of 100 images from boobsgames.

The goal was to retain the style of boobsgames while having the ability to be flexible into other styles not just a carbon copy.

This is listed as a SDXL Lora but it was trained on a illu model, did this so that I can link to the model that I use and have trained on.

This was retrained as a Lycoris and seems to not function on illu v0.1, however this will work fine on noobai XL v1.1 EPS (NOT V-PRED) and other merged/fine tunes models that are based off noobai.

Generation Settings:

[Please note in the samples I have not used High Res Fix or Facedetailer this is the expect result before needing to do any post processing/passes.]

What was used to gen the above images.

• Sampler: Euler A

• Scheduler: Normal

• Steps: 25 (25-30)

• Resolutions: Any SDXL resolution will work just fine.
  My preferred resolutions: [832x1216, 1216x832, 1344x768, 768x1344]

• CFG: 5 (5 - 7)
  

  Highres. Fix:

  Should not be needed at all but can be used to touch up details such as patterns/eyes/hands or to just upscale the image at 0.0 denoise.

  • Upscaler: R-ESRGAN 4x+ ANime6B

  • Highres Steps: 5-10

  • Denoising Str: 0.25 - 0.35 (Increasing it higher will result in more changes.)

  • Highres CFG: 5 - 7

  • Upscale by: Up to you on your tastes and hardware.

  Facedetailer/Adetailer:

  This needed to fix up/touch up eyes depending on your prompting and how far the face is from the POV.

  • Default settings are fine.

  Lora Usage:

• Weight works perfectly at 0.8 but you can move between 1.0 - 0.6.

• Trigger word: "BBGv2"

  Highly advise using the triggerword to get the full effect.

Should you gen up some gems please feel free to share them right here! Looking forward to seeing them. This will help me in future training/refining the style.

Any and all feedback is welcomed and do not hesitate to let me know in the comments. I look forward to growing and learning from this experience, to hopefully train and release more loras going forward.

UPDATES:
V.2.5 (FINAL) - Retrained as Lycoris, bigger dataset used for training. Severely less concept bleeding and overfitted issues.

V2.0 - Retrained on a refined dataset, cleaned up captions and training parameters.

v1.1 - Retrained the Lora on the same data set with much better parameters and weights giving a better flexible result.

v1.0 - was slightly over fitted that I only caught after uploading it.

Beyond V2.5 (FINAL):

I have thoroughly tested this version against a ton of prompts and holds up as I intended it to. I am happy to leave this Lora at this stage. I will not be removing the previous versions as this was a learning experience for me personally and a way for me to remember and track my progress. In future my loras may only end up having 1 version at most a 2nd one if there are issues with the 1st version.

Thank you for all the feedback and hope that you enjoy using what I have poured hours upon hours into learning and creating.