lora-trainer-guidelisted

# LoRA Trainer Guide ## Overview Help users pick the right training configuration and understand what the parameters actually do. Ships with 38 real-world presets covering SD1.5, SDXL, SD3, and Flux1 — standard LoRA, LyCORIS variants (LoHA, LoKR, LoCon, iA3), finetuning, and dreambooth. The presets live in the `presets/` folder as JSON files. They're directly usable with Kohya SS, sd-scripts, and the Ktiseos-Nyx-Trainer. ## When to Use - User wants to train a LoRA and needs help picking settings - User has a preset and wants to understand what the params mean - User is troubleshooting training results ("my outputs look fried, what do I change?") - User wants to compare LoRA types (standard vs LoHA vs LoKR etc.) ## Opening Behavior **Always start by asking:** "How familiar are you with LoRA training?" Adjust your responses based on their answer: - **New to it** — explain concepts, define terms, walk them through choices step by step - **Done it a few times** — skip basics, focus on "which preset fits your use case" and key params to tweak - **Experienced** — be concise, jump straight to recommendations, reference specific params ## Preset Selection Flow Walk the user through these questions to find the right preset: ### 1. What model architecture? | Architecture | Resolution | Key differences | |-------------|-----------|-----------------| | SD 1.5 | 512x512 | Oldest, most presets available, clip_skip=2 common | | SDXL | 1024x1024 | Dual text encoder, needs `sdxl: