A practical look at the prompts, revisions, and human decisions behind my recent short-form videos.
I recently started making short videos with Gemini for my marketing content. My first idea sounded simple enough: picture a chaotic moment, let papers and bills fly across the screen, then bring everything to a dramatic stop. In practice, getting from that idea to a reel I would actually publish took more than one prompt.
That is the part of AI video creation I find most interesting. The tool can make an ambitious concept possible without a production crew. But the marketer still has to decide what the viewer should feel, what the message means, and whether the final cut earns a place on the brand’s page.
I started with the message, then built the visual hook
For an insurance video, I wanted the opening to convey the feeling of being overwhelmed by Medicare information. A stack of papers on a desk would have been too quiet. I asked for papers and bills flying around me, followed by a sudden halt and a calmer moment. That contrast gives the viewer a reason to stop scrolling before I say a word about coverage.
I approached another video from the opposite direction: marketing metrics coming together so a business owner can make a clearer decision. The creative treatment was different, but the strategic question was the same: What idea should someone understand after watching? A visually striking clip is only useful if it moves that idea forward.
Challenge 1: A short clip can only carry so much
I worked in roughly eight-second segments that could be stitched together. That constraint forced me to write for one beat at a time: the hook, the shift, the explanation, and the call to action. When I tried to ask a single clip to carry too much action and dialogue, the pacing became the problem.
My fix was to outline the full reel first, then give each segment a single job. I specified the setting, movement, spoken line, and ending frame. I also treated the transition between clips as part of the script. A strong opening does little good if the next scene feels like a different campaign.
Challenge 2: “Close enough” was not close enough
The first generation often captured the general idea without landing the precise moment I wanted. The stop in my chaotic scene needed to feel more dramatic. A line had to come after “let’s make it clear,” and I did not want the video ending with my eyes closed.
I revised the prompt with concrete instructions about timing and the final frame. Instead of saying “make it better,” I described the visible change: intensify the halt, place the next spoken line after the existing one, and end on an open-eyed expression. These are small details, but they change how finished the video feels.
Using my avatar also meant checking whether the result still felt like me. I could reference it with @me, then review the expression, voice, and scene in context. The avatar helped me create concepts that would have been difficult to film quickly; it did not remove the need for judgment.
Challenge 3: AI sometimes adds things you never requested
One of my marketing videos ended with a stray “Your logo here” placeholder. I removed it, reviewed the revised cut, and found it was still there. That was a useful reminder to watch the entire export after every revision, especially the final seconds. A requested edit does not guarantee that an unrelated defect disappeared.
I also moved the Sweet Impact logo from the upper corner to the bottom of the frame, where it fit the composition better. On another reel, I wanted the Ask April Mai logo visible throughout and the call-to-action screen held longer. Those adjustments happened after generation, during the same kind of brand and editorial review I would give any creative asset.
My final check is simple: watch with sound, watch without sound, inspect the first and last frames, read every word on screen, and make sure the right brand and call to action appear. If a placeholder, awkward expression, or mistimed line survives, the clip is still a draft.
Where AI belongs in my marketing process
Gemini helps me explore a creative direction, produce scenes, and test variations faster. The useful workflow is a loop: define the audience and message, write a focused prompt, generate a short segment, review it, revise one specific problem, then assemble and quality-check the complete video.
I apply the same thinking to marketing data. Organizing information makes it easier to spot what matters; organizing a video concept into clear beats makes it easier to create and evaluate. In both cases, AI can speed up the process, while a person remains responsible for the decision.
I am still experimenting. The encouraging part is that I can now turn an idea like “make the chaos stop” into something people can see, refine it, and learn from what the finished piece does in the real world. That is how I want to use AI at Sweet Impact Marketing: to give good ideas more room to become useful work.
Have a creative concept you have been sitting on? Let’s talk about the message, the audience, and a practical way to bring it to life.
