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.

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.

Execution Flow
- 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.
- 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.
- 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.
- 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.
- 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.

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.

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.

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.

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.
---