Train your first Ideogram v4 LoRA

A free 5-page guide covering JSON prompts, bounding boxes, and dataset preparation — everything SD LoRA training doesn't teach you.

You've trained LoRAs for Stable Diffusion. You know ranks, alphas, learning rates, and dataset prep. Then you try Ideogram v4 and everything breaks.
​
SD uses text captions. Ideogram uses structured JSON. SD describes what's in an image. Ideogram describes what's in an image AND where everything is — with bounding boxes.

This guide bridges that gap.

What you'll learn

Why Ideogram v4 uses JSON prompts instead of text captions — and how to structure them

How bounding box annotations give your LoRA spatial control that SD can't match

The exact training config (rank 64, alpha 64, 3000 steps) that produced the Ektachrome LoRA

How to prepare a dataset with JSON annotations using the hybrid approach (AI + manual editing)

Hi, I'm Straughter.

I build AI content factories. My work covers LoRA training for Ideogram v4, automated audio production, and the full 22-phase SGFLIX pipeline — from character bibles to distribution.

Today, my work reaches hundreds of thousands of creators, my book is a bestseller, and my business generates significant revenue annually—without sacrificing the joy of making.

22-phase

AI content factory

3000 steps

Ektachrome LoRA training

14 hours

RTX 3090 training time

​

Inside the playbook

Everything you need to train your first Ideogram v4 LoRA

• Real training specs from the Ektachrome LoRA (available on HuggingFace)
• Complete JSON prompt templates you can adapt
• Dataset preparation walkthrough with examples

Plus, when you sign up, I'll tailor what you receive based on where you are—whether you're just getting started, growing steadily, or ready to scale.