Your AI Model Was Retired: What Happens to Your Images, Prompts and Work

Model retirements used to be rare. In 2026 they are monthly. Every lab is shipping new versions fast and switching old ones off, and if you build a look on one model, that model can be gone by the time the client asks for a second batch. This guide covers what has been retired recently, what that does to your existing work, and what to keep so you can get close to a result again.

Recent retirements

Two things to note. API retirements don't always mean the model has gone from the apps: Veo 3.1 is still listed in Google Flow's credit table after its Gemini API preview shut down. And Google says its listed shutdown dates are the earliest possible dates. The original Nano Banana API model (gemini-2.5-flash-image), for example, has an earliest shutdown date of 15 March 2027, not 2026 as some reports said.

What happens to your existing work

Files you already made

In the cases above, retirement switches off generation, not your history. Midjourney says V8.0 images "are still available in your website gallery" even though the version can't be used. The bigger risks are account-level: deleting an account, or a workspace being removed. Download anything that matters regardless of model news. For Midjourney, our backup guide covers the official routes.

Your prompts

Prompts don't break, but they drift. A new model reads the same words differently: it may follow them more literally, style them more boldly, or ignore phrases the old model relied on. Midjourney describes V8.2 images as "more creative, bold, sophisticated, and edgy" than earlier versions, which is great for new work and a problem when you need a match.

Seeds and exact repeats

A seed only reproduces a result on the same model. Even then, Midjourney warns that seeds "shouldn't be relied on for the same results over different prompting sessions". Once the model is gone, exact reproduction is gone. Our Midjourney seeds guide explains why.

Plan and feature changes

Models also disappear from plans without being retired. Runway's help centre notes that the models included in its Free plan "can change", and Midjourney's V8.2 replaced Omni Reference, Character Reference and Retexture with its new Edit Model. A workflow can break even when the model name survives.

How to protect your work

1. Keep the full recipe with every keeper: the prompt, the exact model and version (not just "Midjourney" or "Veo"), seed, aspect ratio, stylise or motion settings, and any reference images.
2. Save the reference images themselves, not only links to them. A new model can often match an old look if you give it the old outputs as style references.
3. Pin a version while a job is live. Where a platform lets you choose (Midjourney's --v parameter, for example), set it for the whole project rather than following the default.
4. Download finals at full size as you approve them.
5. Re-test early. When a retirement is announced, run your five most important prompts on the replacement straight away, while the old model is still there to compare against.
6. Watch the notices. OpenAI and Google both publish deprecation pages with dates and replacements, and OpenAI says it gives at least six months' notice for generally available models.

Apps versus APIs

Most retirement notices are written for developers, and an API shutdown doesn't always mean the model has gone from the apps on the same date. Veo 3.1 is the clearest example: its preview models were switched off in the Gemini API on 22 October 2026, yet Veo 3.1 Lite, Fast and Quality are still in Google Flow's credit table. The reverse also happens. Apps change their default model when a new version ships (Midjourney moved its default from V8.1 to V8.2 on 24 July 2026), and third-party tools built on an API have to switch when that API model is retired. If your results suddenly look different, check the tool's model picker or changelog before you rewrite the prompt.

Try it: a migration test

Pick one finished image or clip you would need to match for a client. Take its saved prompt and settings and run them on the current default model. Then run them again with your old output attached as a style or image reference. Compare both with the original: what changed (colour, contrast, framing, faces) tells you which words to adjust. Write down the new working version next to the old one.

Where Dreamdrive fits

Step 1 is the one people skip, because copying prompts and settings by hand is slow. Dreamdrive is a free Chrome extension that saves each image and video automatically as you make it, with its prompt, model and settings, at full size, in one searchable library across your AI tools. When a model is retired you can search for everything you made with it and start your migration test from the real recipe. For older work, see how to find the prompt behind an old AI image.

Related: seeds and repeatable results and downloading everything you made.