Every lesson

What the machines are doing, said in words. Diffusion, latent space, attention, seeds and why none of them can count.

A prompt is a brief. Camera language, light, reference images, and the structure that survives a model being replaced.

Making images somebody will pay for. Reference-led workflows, composition control, holding a face and a product, and the manual last ten per cent.

Direction when the making is free. Choosing four from four hundred, leaving the generic middle, and writing the quality bar down.

Why generated work drifts, and the structures that stop it. Trained products, colour fidelity, reference locks and an automatic gate.

Midjourney, Adobe Firefly and ChatGPT Images as they work in October 2026. Plans and prices, Midjourney's parameters, references and editor, Firefly's models and commercial use, and ChatGPT edits on your own product photo, ending with one brief for No. 4's bottle walked through all three.

The camera-first way to make a generated photograph. Light before sharpness, skin, the four things to check, and when to shoot it for real.

Where video generation actually is. Image to video, camera moves, character drift, sound, and a thirty-second film start to finish.

The accounts, keys and terminal every later module needs. Where a key lives, how to cap the spend, and what to do when one leaks.

Veo, Sora, Runway, Kling, Luma and Higgsfield, used for real. Working API code, the settings that matter, and which model for which shot.

MiniMax H3, LTX-2.5, Wan and HunyuanVideo, and whose country each licence covers. Hugging Face, diffusers, ComfyUI, the GPU you need and how to rent one.

Claude Code and Codex as a video crew. Install, brief, batch render, ffmpeg, MCP servers, and checking what the agent did.

Voice, music, lip sync, upscaling and the edit. Hosted and open tools, with the ffmpeg commands that finish a film.

What the people who ship this every day do. Prompt formula, shot templates, seeds, versioning, QC, cost, licences and disclosure.

Designing the rule rather than the artwork. Parameters, identities that generate themselves, pattern, and authorship afterwards.

The pipeline that replaces a chat window. Human gates, node graphs, batching, versioning images and controlling the cost.

Turning a workflow into a system. APIs, queues, caching, automatic checks, provenance, and the bill that surprises you.

What Figma, Canva, Framer and Adobe actually added. Tokens, the Model Context Protocol, and a stack a two-person studio can hold.

Screens, systems and code. Generated first drafts, design systems as the thing that makes them usable, and evaluating what an agent built.

The vocabulary, grouped by where you will hear it. Models, training, generating, controlling, agents, cost, rights, and the words that mean nothing.

What the models were trained on, where the cases have got to, and how a studio commissions honestly while using all of this.