Part 1: LoRA With Your Eyes Open — mflux Dreambooth

Chapters 01–03: Build calibrated intuition for LoRA hyperparameters using image generation, where failure modes are visible instead of inferred from a loss curve.

TUTOR WITH THEFOCUS.AI

Agent Integration

Copy this prompt into Claude, ChatGPT, or any external AI assistant. It points the assistant to the course instructions and links it to your student profile to track your progress and customize observations.

Please tutor me in this lesson using the following context. First, read the instructions at: https://courses.thefocus.ai/llms.txt My Student ID is: <none> The lesson markdown source is at: https://courses.thefocus.ai/fine-tune-local-agent/01-lora-eyes-open/01-why-images-first.md

You are not enrolled yet. Enroll to generate a Student ID to track lesson completions and store learning notes.

Part 1: LoRA With Your Eyes Open

mflux Dreambooth adapter · ~1 day

You’re about to spend many hours staring at training loss numbers and guessing whether rank 16 or rank 32 was right. Before that, build intuition in a modality where you can see the failure modes: image LoRA. Same math, same knobs, visible results.

Chapter	What You’ll Do
01 · Why Images First	Understand what LoRA actually does and why image training teaches text training
02 · Train an Adapter	Build a Dreambooth dataset and run your first training with mflux
03 · The Experiments	Sweep rank, steps, dataset size, and target modules — and see what each one does

What You’ll Have After Part 1

A working mflux training setup on Apple Silicon
A trained Dreambooth LoRA of a subject you know well
A contact sheet: one grid image, same prompt, every adapter variant labeled
The ability to look at a generation and say “that’s overfit” or “that’s underbaked” without checking any numbers

This module is deliberately throwaway. The artifact is in your head.

Chapter 01 →