Two frames, two prompts: how AI transitions actually work for brands

You’ve probably seen at least one AI video clip recently that stopped your scroll. A transition that looked physically impossible, footage that seemed to move through a surface or a lens or a window into another place entirely. Your immediate thought was probably some version of: how did they do that, and could we do that for our brand?

The honest answer is yes. But the more useful answer is: it depends entirely on whether the person making it understood what they were doing, or whether they got lucky once and can’t repeat it.

What the process actually looks like

Here is the specific example worth understanding. A creator took two frames: a real photograph of a camera lens taken on location, and an AI-generated image of what you’d see from inside that lens looking out at the Azores. He dropped both into Gemini and used Veo 3 to generate an animated clip bridging them. Then he reversed the direction: the AI frame became the new start, the real location shot became the new end. Second prompt, second clip. Stitch the two together and you have footage that appears to fly through the lens into a real place.

Two frames. Two prompts. One effect that looks like it required a camera rig, a drone, and a post-production team.

What makes this work is not the tool. Veo 3 is powerful but it doesn’t make editorial decisions. What makes it work is the structural logic:

• The start and end frames are chosen with the transition in mind, not just as standalone images.
• The AI frame is designed to serve as both an end point and a reversible start point.
• The stitch is the actual creative problem being solved, and everything before it is in service of that.

That is a workflow, not a one-off. It can be applied to a product shot, a brand film opener, a campaign asset that needs to feel distinct from anything competitors are doing.

Why most AI video still disappoints

The majority of AI video content that brands are producing right now has a recognisable quality: it looks like a series of experiments. Different visual logic in each clip. Characters that shift between frames. A colour palette that drifts. Transitions that feel random rather than intentional.

This is not a tools problem. The tools available now, including Veo 3, are genuinely capable of producing work that holds together. The gap is in the thinking that goes in before a single prompt is written.

The craft is in knowing which frames to choose and how to prompt the join so it holds together.

A brand director who has worked through this kind of process dozens of times has developed a feel for it that a first-time user cannot shortcut. They know when an AI-generated frame will serve as a credible bridge and when it will break the illusion. They know how to write a prompt that preserves the visual logic of the real footage rather than fighting it. They know the stitch point is where the work either convinces or falls apart.

What this means when you’re hiring

If you’re commissioning AI creative work for your brand, the question to ask is not “have you used Veo 3” or “can you show me AI video you’ve made.” Those are table stakes.

The question is: can you show me a process? Can you explain why you chose those frames, what you were solving for in the prompt, and how you’d run this again next month with different source material and get a consistent result?

A creative who can answer that is bringing you a system. A creative who can’t is bringing you a clip.

The transition in this example looks impossible because the process behind it was completely deliberate. That is the standard worth holding AI creative work to.