Pillar 2 · Introduction

AI Image Generation

Understanding and mastering the stochastic world — from fundamentals to a complete ComfyUI setup.

Articles + tutorials + examples · continuously expanded

2.1 · Fundamentals

2.1.1

How AI image generation works — from prompt to image

2.1.2

What is a model — checkpoints, LoRAs, VAE, ControlNet models

2.1.3

Model types overview — SDXL, Flux.1 dev, Flux.2 dev, Ideogram 4

2.1.4

Understanding licensing — open source vs. open weights

2.2 · Setup & Infrastructure

2.2.1

Stability Matrix — installing and managing local AI image generation

2.2.2

ComfyUI — philosophy and interface

2.2.3

ComfyUI node types — loaders, conditioning, sampling, ControlNet, output

2.2.4

Building a workflow — SDXL from scratch in ComfyUI

2.3 · Model-Specific Setups

2.3.1

SDXL — setup and parameters — Juggernaut XL in ComfyUI

2.3.2

Flux.1 dev — setup and parameters

Flux.2 dev — what's new, setup, when it's worth using

in progress

Ideogram 4 — strengths, setup

coming up

2.4 · Prompt Architecture

2.4.1

Prompt architecture: the foundation — CLIP, T5, the 8 categories

2.4.2

Prompts for SDXL — keywords, weighting, negative prompts

2.4.3

Prompts for Flux — two encoders, two fields, no negative prompt

Goal → prompt → result: documented examples

in progress

2.5 · Reference Images & IP-Adapter

What reference images do — and when they replace ControlNet

in progress

IP-Adapter in ComfyUI — concept, node setup, strength

in progress

Use case: models without ControlNet support (Flux.2, Ideogram 4)

in progress

Strengths of IP-Adapter that ControlNet doesn't have

in progress

Reference image setups for different models

in progress

2.6 · Extended Workflow: Local + Cloud

The three-stage production chain — Blender · local AI (RAY-L) · cloud model

in progress

Why local and cloud are production stages, not competitors

in progress

Reference image as a bridge between local results and cloud models

in progress

Gemini, Flux.2 Pro, and other cloud models compared

in progress

Local vs. cloud — same prompt, two worlds

test pending