Video Production
How to Create an FPV Feel Without Flying a Drone
We produced the ATASAY Tema World video without a drone shoot, using real photos of the location. Every decision in the process, plus a free 24-page guide.
Reference-Based FPV Training Guide
A 24-page PDF: photo selection, camera design, copy-paste prompts, the real revision text and a pre-delivery checklist. Free, no sign-up required. The guide is in Turkish.

Everyone who watched the ATASAY Tema World video we shared on Instagram asked the same first question: which drone was this shot with?
The answer satisfies no one: there is no drone in that video. No new location shoot either. The video was produced with a reference-based AI workflow built on real photographs of the store.
In this post we walk through the decisions behind that process. At the end you can download the 24-page training guide that shows the process step by step, free of charge.
What Was in the Video?
The goal was an uninterrupted ad narrative resembling an FPV flight: a controlled approach from outside, a slow close-up along the ATASAY sign, entry through the open door, a short interior tour and a final on the angled corner composition. All of it in 15 seconds, vertical 9:16, with a single-shot feel.
Shooting this with a real drone inside a mall means permits, insurance, crowds and a full production day. On top of that, the FPV pilot has to clear that narrow doorway cleanly in one take. With reference-based production, the same route is designed from existing photos of the location.
The Result Is a Video, the Real Work Is Decisions
The most important part of the process is not writing the prompt but the decisions made before it. We use six steps: brief, reference selection, camera design, production settings, revision and delivery checks.
The critical decision is this: write down what must not change first. The ATASAY lettering and the 1937 digits, the white corner column, the window and door positions, the counter layout inside. These are not aesthetic preferences but accuracy criteria. If the model breaks any of them, the video is not delivered no matter how smooth it looks.
One Job per Photo
We used four main references in production and gave each one a single job:
- R1, the corner: the processed exterior composition. The video's final framing.
- R2, the sign: a current photo of the metal letters and the window. The authority for the logo close-up.
- R3, the door: a photo of the real open entrance. It defines where the camera passes through.
- R4, the interior: the central aisle, counters and ceiling. It draws the limits of the interior tour.
Uploading images to the interface in order is not enough. The prompt names each image and its job explicitly, and states which one wins when references disagree. Our rule was clear: R3 to align with the doorway, R1 for the final.
A 15-Second Choreography
We wrote the timing plan in second ranges: four seconds of exterior approach, then a three-second logo close-up, the entry through the door, a short interior tour and the exit settling into the final over the last two and a half seconds. Each range has one main movement, one target and one speed decision.
The lesson here is time discipline: squeezing a long mall route, a slow logo pass and an interior tour into the same 15 seconds is not physically convincing. If the route does not fit, you shorten the approach instead of stitching the space together with jumps.
A Beautiful Frame Is Not Enough
A generation being accepted does not make the result correct. Before delivery we do two viewing passes: first the rhythm at normal speed, then frame-by-frame checks at the sign, the door threshold and the turning points.
The checks are concrete: counting the A-T-A-S-A-Y letters one by one, verifying the 1-9-3-7 digits, watching that the number of counters does not change and the window geometry does not melt. Revision follows a single rule too: fix one main problem at a time and write a concrete request with a timestamp instead of "it did not work".
Locking the final onto the corner photo we wanted was exactly that kind of revision: the previously generated video was given as the motion reference, the R1 photo as the end-frame target, and only the last section was re-described. The technical details of that attempt, including the first method that was rejected, are covered openly in the guide.
The Full Guide Is Yours
This post is a summary of the process. The kitchen work is in the 24-page PDF: which photos to take on site, the reference manifest, copy-paste English prompts, the exact revision text that was used, production settings, quality control questions and budget notes.
The guide is free and requires no sign-up. Download it and try it on your own location. If you send your attempt to our Instagram account, we will look at the part you struggled with most together.
- AI video
- FPV
- promo video
- Seedance
beyworks team
A creative agency bringing strategy, production and digital growth under one roof.
Related Posts
An FPV Prompt That Builds the Building as You Fly
The effect architecture promos love: the building assembles as golden wireframe while the camera climbs. A copy-paste 15-second FPV prompt plus adaptation notes.
Read the PostWhy Presenter-Led Content Builds More Trust
Two videos, same budget, very different results. How content with a face actually works, three formats worth using, and how to start without a presenter.
Read the PostYour Videos Are Lost in the First Three Seconds
When a video goes unwatched the problem is usually the opening, not the edit. Five openings that work and the fastest way to test your own.
Read the Post