Kamilla18 [Lydia Cheremnykh] (Adult Web Model)

My second public release. Please remember to rate and review if you like the model.
Click "Show More" to continue reading for best results using this LoRA.

Kamilla18 [Lydia Cheremnykh]

Born on Aug. 30th, 1985, Lydia Cheremnykh, aka Kamilla18, is a retired adult web model active from 2003-2008. Lydia appeared on her own site Kamilla18.com along with many others (under various aliases) such as: MET Art, MPL Studios, JustTeenSite, Rigin Studios.

For web model fans, my JenyaD (aka Katie Fey) LoRA can be found HERE and I hope to begin working on Felicity Fey soon!

Example Images: The example images are unmodified. Asides for ADetailer, no negative embeddings, inpainting, enhancement LoRAs, or Photoshop changes were made.

General Notes:

• NOTE: I have included a ZIP file (labeled 'Training Data') containing a Civitai Helper JSON file and a preview image for convenience.

• It takes a LOT of prompting to keep this model's clothes on above LoRA weight 0.7. You've been warned.

• For her signature photoshoot style, a LoRA weight of 0.9 - 1 is recommended. ADetailer for her face is optional.

• For increased flexibility and use with other LoRAs, a LoRA weight of 0.6 - 0.7 is recommended. ADetailer for her face is a must! Using this LoRA in the ADetailer prompt with an Inpaint denoising strength of up to 0.7 produces great result while maintaining flexibility and near perfect likeness.

Triggers:
Mandatory: Kamilla18, 1girl, solo, black hair, brown eyes, long hair
Optional: ??????????????, ???????, ?????
NOTE: The tags 1girl, solo, black hair, and brown eyes can be omitted at LoRA weights lower than 0.9 for increased flexibility, but as mentioned, ADetailer face denoise strength will need to be increased to maintain a strong likeness.

LoRA Usage:
Recommended Models: epiCPhotoGasm X, PicX_real v1.0, or Realistic Vision v5.1 VAE
Sampler: DPM++ SDE Karras
LoRA Weight: 0.9 to 1
or
LoRA Weight: 0.6 to 0.7 (see ADetailer notes above & below)
Steps: 30-40
Clip Skip: 1 or 2
1=more realism but less flexibility & slightly increased chance of body horrors
2=very slight decrease in realism but more flexibility & less chance of body horrors
CFG: 6 or 7
Width x Height: 512 x 768

Upscaler:
Hires.fix Upscaler: 4x_foolhardy_Remacri [HuggingFace direct link]
Upscale by: 2
Hires steps: 10
Denoising strength: 0.15

ADetailer:
ADetailer Model: face_yolov8n.pt
ADetailer Prompt: Kamilla18, black hair, brown eyes, long hair
Inpaint Mask Blur: 6
Inpaint only masked padding: 48
Inpaint denoising strength: 0.5 - 0.7 for LoRA strength lower than 0.9. For 0.9 - 1, use 0.2 - 0.35.

Example Prompts:

Prompt: best quality, <lora:Kamilla18_v1:0.9>a photo of Kamilla18, 1girl, solo, black hair, long hair, brown eyes, petite, extreme detail, detailed face and eyes, perfectly drawn eyes

Negatives: (fat, obese:1.4), (wide ????:1.4), (fat ???:1.4), plastic skin, blurry, (cartoon, anime, painting, drawing, cg, 3d render, airbrushed, photoshop:1), text, logo, watermark, web address

Training Info:
Software: OneTrainer by Nerogar
Images Used: 1,049 (543 body @ 30 repeats, 454 face @ 34 repeats, 26 ??????? @ 4 repeats, 26 vag @ 6 repeats)
Total Training Steps: 15,993
Epochs: 6
Optimizer/Scheduler: AdamW/Cosine
LR: 0.0003
TE LR: 0.00015
Batch Size: 2
Data Type: bfloat16
Training resolution: 512, except body, which as trained at 512 and 768
Training time: 3.7 hours on my Intel Core I9 14900K & RTX 4090 24GB VRAM w/ 128GB of system RAM


License:
This model is NOT to be uploaded anywhere outside Civitai.
No commercial use permitted.
This model, and its created works, may not be sold.