Details
Related
v1
Aikuros (hyoon) | Korean Instagram Model & Streamer

Aikuros (hyoon) | Korean Instagram Model & Streamer

313
76
474
#Character
#GIRL
#person
#Photorealistic
#????
#Female
#Asian
#lora
#Vtuber
#model
#?????
#woman
#beautiful
#celebrity
#girls
#Photography
#portraits
#Realism
#real person
#Character
#GIRL
#person
#Photorealistic
#????
#Female
#Asian
#lora
#Vtuber
#model
#?????
#woman
#beautiful
#celebrity
#girls
#Photography
#portraits
#Realism
#real person

Hold up! ⚠

This model (v1) is still a work in progress.

I apologize for the big file, still learning how to train LoRa's efficiently 😉

Would love some pointers if anyone sees a glaring mistake or error in my first training.

The Model

Tools:

  • Custom web scraper for HQ photos off social and Reddit

  • birme.net for bulk image cropping

  • Pre-processed w/ BLIP

  • bmaltais' kohya_ss for the actual training

Parameters:

121 photos of varying poses and angles, 80% cropped to face, rest to body

\\ hyoon_v1 parameters

ss_mixed_precision:  "bf16",
ss_network_alpha:  "128.0",
ss_network_dim:  "128",
ss_network_module:  "networks.lora",
ss_noise_offset:  "None",
ss_num_batches_per_epoch:  "6050",
ss_num_epochs:  "1",
ss_num_reg_images:  "0",
ss_num_train_images:  "12100",
ss_optimizer:  "torch.optim.adamw.AdamW",
ss_output_name:  "hyoon",
ss_prior_loss_weight:  "1.0",
ss_resolution:  "(672, 672)",
ss_sd_model_name:  "runwayml/stable-diffusion-v1-5"

Recommended Settings:

I probably messed up somewhere, because the best results I have gotten so far are from the following triggers:

<lora:hyoon_v1:0.8> (hyoon:1.2)

Prompt used:

<lora:hyoon_v1:0.8>,(8k, RAW portrait,photorealistic,masterpiece:1.2), (full body) portrait of (hyoon:1.2) posting on a grassy hill, detailed cute face, ??????????????, low-cut shirt, cleavage, slender body, looking at the viewer, smiling, <lora:fashionGirl_v40:0.1>


Negative prompt: easynegative:0.4, illustration, painting, drawing, art, sketch, deformed, ugly, mutilated

Model used: doctorDiffusionV08


WARNING: this model seems to be on the naughtier side, so please respect the person and use negative prompts - ????, ?????, ???????, ?????

Very oddly, Euler a seems to be the only 'particularly ?????' sampler. No idea why or how that happens, but please feel free to enlighten me.

DISCLAIMER: Please respect the person of the model, use common sense, and don't get in trouble. What you do during fap time is your private business, but don't ruin it for everyone else by being reckless and damaging the reputation of the person or the website and the great services they are providing.

View Translation

Rating & Review

-- /5
0 Ratings

Not enough ratings or reviews received yet

no-data
No data available
avatar
Chat with the model
Notice
2023-04-01
Publish Model
2026-01-02
Update Model Info
Model Details
Type
LORA
Publish Time
2023-03-20
Base Model
SD 1.5
Trigger Words
<lora:hyoon_v1:0.8>
hyoon:1.2
Copy
Version Introduction

First release, lots of optimization to do. Some pointers welcome!

Potentially could turn ???? easily, please use negative prompts.

License Scope
Model Source: civitai

1. The rights to reposted models belong to original creators.

2. Original creators should contact SeaArt.AI staff through official channels to claim their models. We are committed to protecting every creator's rights. Click to Claim

Creative License Scope
Online Image Generation
Merge
Allow Downloads
Commercial License Scope
Sale or Commercial Use of Generated Images
Resale of Models or Their Sale After Merging
QR Code
Download SeaArt App
Continue your AI creation journey on mobile devices