About This Model

Trained on YoreSpot (https://yorespot.com) - a free AI image generation and Video platform with built-in LoRA training, 30+ workflows, and no content restrictions.

Training Specs

NVIDIA H200 GPU, Kohya sd-scripts, AdamW8bit optimizer, LR 1e-4 (text encoder 5e-5), Cosine w/ Restarts scheduler (3 cycles), 15 epochs, Batch size 2, 1024px resolution w/ aspect-ratio bucketing (512-2048), Noise offset 0.0357, Min SNR Gamma 5, bf16 training / fp16 save, Alpha = Rank / 2, Caption shuffling, Gradient checkpointing, Latent caching to disk

Why This Matters

A 20-image dataset trains for around 3,000 steps across 15 full epochs. A 50-image set reaches around 7,500 steps. Most online trainers cap around 950 steps - enough for a rough likeness, but not enough to learn varied poses, expressions, and outfits. More steps with cosine warm restarts means a LoRA that is flexible, prompt-responsive, and does not lock into one stiff pose.

Visual Epoch Selection

YoreSpot generates sample images at every epoch using 3 prompts and seeds. After training you pick the best epoch visually - choosing the sweet spot between underfitting and overfitting instead of being stuck with the last checkpoint.

Base Models: Illustrious XL, NoobAI XL, Pony Diffusion V6 XL, SDXL 1.0

Use It

Browse models: https://yorespot.com/#models
Generate: https://yorespot.com/#generate
Train your own: https://yorespot.com/#train

Full ComfyUI-powered generation. NSFW fully supported - no keyword restrictions. Free to use with credits. All trained models downloadable as .safetensors for use anywhere.