Free preview · no sign-up required

Midjourney Masterclass: full syllabus and an entire chapter

From today's landscape to the contract: creating images with AI without betting on a tool that could be shut down

This course starts with a date. On August 17, 2026, Google shut down the Imagen API — three models, with the notice buried in a changelog almost nobody reads. Every tutorial that taught that API broke the same day. This isn't a technical detail: it's the reason this course is structured the way it is. Tools are the part that ages, and material that teaches buttons ages right along with them.

What doesn't age is the method: choosing the model by the task, reading a ranking knowing what it measures, describing intent and constraints instead of memorizing modifiers, and recognizing when the output is wrong. That last skill became more valuable in 2026, because the model became an agent — GPT Image 2 plans the layout and self-checks, HiDream rewrites your prompt before generating. The skill that 2023 courses were selling is being automated by the model itself. What's left is judgment.

From chapter 1 to 30, we follow the same person: Marcos, a freelance designer in Sorocaba, four small clients, revenue coming from e-commerce assets. In chapter 1, the script that generated images for his biggest client stops working overnight. By chapter 30, he doesn't depend on any tool: he chooses by task, knows what he can sell and under which license, knows when it's worth running on his own machine, and charges for direction and review instead of charging per click.

30 chapters · 6 modules · 7.4+ hours

What you learn, chapter by chapter

Module 1: Why the tool is the part that ages

A shutdown with a set date opens the course and explains its format. The short story from GAN to transformer, just enough to understand why today's tool behaves the way it does — and the 2026 turning point, where the model became an agent and modifier-based prompt engineering began to die.

  1. 1. The day the API was shut down, and what it teaches about learning tools

  2. 2. Four shifts that explain what you see on screen

  3. 3. The model became an agent — and 2023 prompt engineering is dying

  4. 4. Describing intent and constraint: the request that survives switching models

  5. 5. Reading rankings with skepticism: the score measures what the score measures

Module 2: Today's landscape ⚠️ dated chapter

The scoreboard and spec sheet for each tool, with the reading date next to every number. This is the module that ages on purpose, so the rest of the course doesn't age with it: when the ranking shifts, this module gets rewritten.

  1. 6. The August 2026 scoreboard, and how to double-check it yourself

  2. 7. Top closed models: GPT Image 2, Nano Banana, Seedream and the new ones

  3. 8. Midjourney today: what changed, how much it costs, and when it still wins

  4. 9. Open weights: Z-Image, HiDream, Qwen and FLUX — and Alibaba's turning point

  5. 10. The specialists: Ideogram for text, Recraft for vector, Firefly for legal clearance

Module 3: What you can do without paying

The free tier shrank in 2026 and started charging in other ways: public images, retained intellectual property, watermarks, the main feature behind a paywall. And the only truly free option — open weights running on your own machine, with the license read beforehand.

  1. 11. The free tier shrank: what's left, service by service

  2. 12. The five forms of disguised franchise

  3. 13. Open isn't the same as free to sell

  4. 14. Install the truly free one: an open model running on your machine

  5. 15. The free workflow, and where it breaks down

Module 4: Your own machine: cost × time

The honest math in a year when GPUs got expensive: what runs on each VRAM tier, how long it takes, what Apple Silicon changed, why quantization format matters more than swapping your card, and at what monthly volume the machine pays for itself.

  1. 16. 2026 is the worst year of the decade to buy a graphics card

  2. 17. What runs on each VRAM tier, and how long it takes

  3. 18. Apple Silicon stopped being a joke

  4. 19. Swapping the format yields more than swapping cards

  5. 20. At what volume the machine pays for itself — and the rival that isn't the API

Module 5: The right to use and sell

What you can register, what courts have already decided and what's still open, the labeling requirement that took effect in Europe in August, what Brazil is debating, and why provenance became platform infrastructure.

  1. 21. You don't register what AI generated — and what that changes for your business

  2. 22. The lawsuits: what's already been decided and what remains open

  3. 23. The European AI Act is already in force: labeling is no longer best practice

  4. 24. Brazil: What's Already Required, and What's Still Under Discussion

  5. 25. Provenance became infrastructure: declare instead of hide

Module 6: Where this turns into money

Five applications with cost, market price, and margin: product photography, marketing, real estate staging, architectural visualization, and fashion. Plus the recheck routine that keeps all of this true next quarter.

  1. 26. Product photography: the application with the best business case

  2. 27. Marketing: the job is in curation, not in the click

  3. 28. Real Estate: the highest gross margin in the course, with a mandatory caveat

  4. 29. Architecture and fashion: where AI fits and where it still doesn't

  5. 30. The check-in routine that keeps this course true

Chapter 1 in full

This is the complete chapter, just like the one inside the course — text, images, and videos.

Overview

You're reading this chapter because you want to generate images with AI. Maybe you already have a favorite tutorial, a prompt that works, a tool you use every week. We're going to break that trust on the very first page — on purpose. Not to frustrate you, but because the only certainty in this market is that what you learned yesterday may not exist tomorrow.

On August 17, 2026, Google shut down three models in the Imagen line that were live until the day before. Anyone who had automated their workflow on top of them woke up to a broken system. This chapter uses that real shutdown to teach you the one skill that survives any change: evaluating a tool before depending on it. You'll learn to read a changelog, to separate what's essential from what's accessory, and to build a workflow that doesn't break when a vendor decides to leave the game.

Key Concepts

An AI tool isn't a stable product like a hammer. It's a service that changes by a company's decision, with notice that fits in a paragraph. The official Gemini API changelog, read on 08/16/2026, marked the shutdown of imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001 for the following day. The imagen-3.0-generate-002 had already been discontinued on 11/10/2025. Notice: this wasn't an update, it was a closure. The Imagen line is being retired, not succeeded — there's no "Imagen 5." Google's official recommendation is to migrate to the Nano Banana family of models.

This teaches you the first rule: tools are temporary, tasks are permanent. Marcos's task, our freelance designer from Sorocaba, is generating product images for e-commerce. The tool he used for that changed overnight. What he needed to know wasn't the magic prompt for that specific API, but how to evaluate whether the next tool handles the task, what it costs, what the limits are, and what the risk is of it also being shut down.

Deciding what a good result looks like before asking the tool for anything.
Deciding what a good result looks like before asking the tool for anything.

The second rule is: the changelog is the first place to look before choosing anything. You don't need to read the entire documentation. You need to read the change history, the deprecation notices, and the dates. If a tool has three discontinuations in the last year, it's a risk. If the company is migrating product lines, as Google did, you need to know where to. This reading is a trainable skill, and it's exactly what we're going to do now.

The Gemini API changelog shows the shutdown dates for the Imagen models. Learning to read this page is the first step to not being caught off guard.
The Gemini API changelog shows the shutdown dates for the Imagen models. Learning to read this page is the first step to not being caught off guard.

Execution Flow

  1. Open the Gemini API changelog page (ai.google.dev/gemini-api/docs/changelog) and search for "deprecation" or "shutdown." You'll see a table with model names and dates. Write down the names you use or plan to use.
  2. Check whether the model you want is on the discontinuation list. If it is, look for the "Recommended migration" section on the same page — that's where they point you to the model to migrate to. In our case, the recommendation was Nano Banana.
  3. Test the recommended model with a real task of yours, not a generic prompt. Use the same product image, the same style, the same text. Compare the result with what you got before. If the new model doesn't cut it, look for alternatives outside the vendor.
  4. Log the discontinuation date in a calendar, with a reminder one week before. That way you don't find out about the shutdown on the day a client calls complaining.
  5. Review your automated workflows: if you have scripts or integrations that call the API, check whether the model name is hardcoded. Swap it for a variable you can update without rewriting everything.
Step-by-step on screen: how to navigate the changelog, find the model, and note the shutdown date.
Step-by-step on screen: how to navigate the changelog, find the model, and note the shutdown date.

Applied Scenarios

Scenario 1: The script that broke on Monday. Marcos had a script that automatically generated ad images for his client "Café do Bairro," a local coffee shop. He used imagen-4.0-fast-generate-001 because it was fast and cheap. On Monday, August 17, the script returned an error. Marcos opened the changelog — which he had never read — and saw the model had been shut down that day. Result: the client went without a new ad, and Marcos lost the entire morning redoing it by hand. After that, he set up a monthly reminder to check the changelog for every tool he uses.

Scenario 2: The conscious choice. Two weeks later, Marcos needs to choose a tool for a new client, a clothing store that wants to generate product variations. Instead of following the first tutorial that came up, he opened the changelogs of three candidates: the Gemini API, a third-party service, and an open-source tool he could run on his own machine. He saw the Gemini API was in a transition phase, the third-party service had a stable history, and the open-source option depended on hardware he didn't have. He chose the third-party service, aware of the monthly cost, and kept the open-source option as a plan B for the future.

Marcos comparing three tools in a spreadsheet: price, date of last changelog, and discontinuation risk.
Marcos comparing three tools in a spreadsheet: price, date of last changelog, and discontinuation risk.

Common Mistakes

  • Thinking "official" means "it will last." Google is official and shut down the Imagen line. No tool is eternal.
  • Trusting the tutorial that worked last month. Tutorials don't warn you when an API changes. The changelog does.
  • Not reading the date of the notice. The changelog says "shut down on 17/08," but the notice may have been published months earlier. If you only look afterward, it's already too late.
  • Migrating to the recommended model without testing. Google's recommendation was Nano Banana, but the result may not suit your case. Test it with your real task.
  • Hardcoding the model name in your code. When the model changes, you have to rewrite the entire script. Use a variable.
API error on Marcos's script screen: what appears when the model no longer exists. The message doesn't say where to migrate.
API error on Marcos's script screen: what appears when the model no longer exists. The message doesn't say where to migrate.

Pro Tip: Set up a Google alert for "Gemini API changelog deprecation" and also for the name of any tool you use. Google doesn't send an email warning you it's going to shut down a model — but the changes page is updated before the shutdown. If you get the alert on the day it's published, you still have time to migrate. Gemini 3.5 Flash and Gemini 3.1 Pro are the current models in the line, but the lesson applies to any provider: the changelog is your primary source of truth.

Setting up a Google alert for the Gemini API changelog page.
Setting up a Google alert for the Gemini API changelog page.

Practical Exercise

Open the Gemini API changelog (ai.google.dev/gemini-api/docs/changelog) right now. Find the section listing deprecated models and write down in a text file: each model's name, the shutdown date, and the migration recommendation. Then do the same for the AI tool you use most today — it could be an image generator, a chatbot, whatever. If the tool doesn't have a public changelog page, that's already a warning sign: write that down too. The done criterion: you have a list with at least two tools, the deprecation dates (if they exist), and the migration recommendation. Keep that list — you'll use it in the next chapter.

Implementation Checklist

  • I can explain why no AI tool is stable enough to base my business on without monitoring.
  • I can find a tool's changelog page in under five minutes.
  • I can identify a deprecation notice and its associated date.
  • I can test a recommended model with a real task of mine before migrating.
  • I have a list of the tools I use and their change dates.

Chapter Summary

  • The shutdown of the Imagen line on 17/08/2026 is a real example of how tools disappear without personal notice.
  • The changelog is the primary source of truth: it shows dates, deprecations, and recommendations.
  • The task is permanent; the tool is temporary. Learn to evaluate, not to memorize.
  • Migration requires testing with your real case, not blind trust in the official recommendation.
  • In the next chapter, you'll understand the four shifts that explain why today's tools behave the way they do — and how that helps you predict the next change.

---

Previews of similar courses

View all courses →
FAYAI