Before 2027, practise editing an existing image, controlling motion in a short scene and archiving a project. Client work also calls for checking material rights and agreeing how AI use will be disclosed. These preparations are useful regardless of the next release date.
If you work in Midjourney, a new version usually brings curiosity and a little disruption. You want to try its features, but you also have an image to finish. You do not have to drop everything and start learning from scratch.
For 2027, I would prepare a few practical things in the way you work. The broader context is in my post about the direction of AI art. Here, the focus is on exercises you can do with your own files.
Finish one image
Choose a piece that is close to what you want but still feels wrong. Before opening the editor, name the problem. Is the background too busy? Does the subject blend into its surroundings? Is the light drawing attention to the wrong place?
Change one thing and compare the versions. Check the parts you did not intend to alter, too. This tells you more about editing than ten random instructions. Midjourney opened its V8.2 editing model in August 2026, so there is already a tool to try this with.[1]
Start with a short scene
You do not need to plan a film straight away. Take one frame and decide what should move. A simple gesture or a slow camera movement is enough to begin with. Fewer events make it easier to see what works.
Watch for continuity: does the face remain the same face, do objects keep their shape, and does the movement make sense? Composition and lighting skills help, but video also asks you to think about time. Not everything carries over automatically from still images.
Save what you want to return to
The closure of Sora's app in April 2026 is a reminder that access to a service is not permanent.[2] Rather than trying to predict every platform's future, put your own materials in order.
For an important project, keep the source images, selected results, prompts, settings and a short note about your changes. Name the files so they will still make sense in a few months. A sequence of generation numbers will not tell you much then.
Try another application on one small task, too. You do not have to move your entire workflow. It helps to know where to start if your current tool stops being suitable for a project.
Before you hand an image to a client
Find out where the work will be used and what the client requires. Look at the reference materials, recognisable characters and logos. The fact that a model generated the image does not settle the rights to everything in it.
These disputes are real: Disney and Universal sued Midjourney in 2025.[3] You do not need to wait for the outcome to ask whether your client has permission to use the materials they supplied.
Check what disclosure is needed
The AI Act distinguishes between providers, users and content types, including deepfakes.[4] Saying “I use AI professionally” is not enough to establish your obligations for a particular publication.
Discuss how the image's origin will be described before delivering the files. If the work could be mistaken for a photograph of a real event, agree with the client how it will be presented. Check the European Commission's materials rather than relying only on short summaries.
Video, disclosure and rights: things to consider when working on a project.
One exercise for this week
Take an unfinished image. Write one sentence about what you want to improve, make one change and assess it. Save both versions with your notes. If you want to explore motion, use the same frame for a short animation.
Then ask yourself whether the exercise helped you describe more precisely what you want from the image. That skill makes it easier to work with whatever version of the software comes next.
Sources checked on September 20, 2026. Forecasts for 2027 are the author’s assessment.