The imagery in every Living Devotional film is generated with AI tools. Everything that decides what the film is — the script, the theology, the shot list, the camera language, the casting of every recurring face, the selection of which generated frames survive, the edit, the sound, and the bill — is mine. No part of this is automated. I generate every scene by hand, one at a time, and I throw most of them away.
I have spent ten years as a producer — documentaries, brand films, corporate education, drone work, events. I know what a shoot costs. So when I say I could not have flown a crew to a nineteenth-century-BC Egyptian prison to shoot Joseph, I mean it literally: that film does not get made, at any budget I have.
AI image and video tools are the reason it exists. I want to be direct about that, because I think audiences deserve to know what they are looking at, and because the moment you hide it is the moment you should not be doing it.
But there is a version of “AI video” that means type a sentence, post whatever falls out. That is not what this is, and the difference is the entire job. Here is the actual pipeline.
1. The script comes first, and it is written
Every episode starts as a screenplay, then becomes a structured JSON file — one entry per scene. For the Joseph episode that is 36 scenes. Each one carries the English narration, the Portuguese translation (written to sound natural, not translated literally), the narrative block it belongs to, and the direction.
I check the story against the text before anything visual happens. If the passage does not say it, it does not go in the film. Genesis 40:23 says the cupbearer forgot Joseph; that sentence is the spine of the whole episode, and I built the structure backward from it.
2. I write the camera, not just the prompt
Every scene specifies framing, lens length in millimetres, and movement — and I deliberately vary lens and movement between consecutive scenes, the same way I would break up coverage on a real shoot. A film where every shot is a slow push-in on a 35mm reads as a machine made it, because a machine did.
There is also a global style definition and a global negative list that ride on every scene: anamorphic character, volumetric light, prison-umber shadows, dawn-gold highlights, real skin texture, film grain — and explicitly no morphing hands, no fused fingers, no sliding feet, no waxy skin, no modern objects in ancient frames, no text, no watermarks. That negative list exists because I kept seeing those failures and got tired of them.
3. Characters get cast, then kept
Joseph is seventeen in scene 7 and thirty in scene 31. He has to be recognisably the same man. So each recurring character gets a written physical sheet and a multi-angle reference — front, profile, three-quarter, one intense expression — and every scene that character appears in is generated against that reference.
This is the part most AI video skips, and it is why most AI video is unwatchable for longer than fifteen seconds. Continuity is not a nice-to-have. It is the difference between a film and a slideshow.
4. Nothing is generated automatically
This is the part I most want on the record. My pipeline used to call generation APIs. In July 2026 I removed that capability entirely, on purpose.
Now every single visual is produced by me, by hand, one scene at a time, in the tool — and dropped into the project folder as a finished asset. The pipeline cannot generate anything. If I have not personally made and approved a shot, that scene renders as a grey placeholder until I do.
I made that change because automated batch generation produced volume, not quality. I would get 36 scenes back and like four. Doing it by hand is slower and it is the only way the result is any good.
5. Most of what I make gets thrown away
A single usable still costs credits in the dozens, and I routinely burn several attempts before one survives — wrong hands, dead eyes, a face that drifted off-model, a composition that fights the narration. Those failures are logged and paid for exactly like the keepers.
That cost is mine. Nobody sponsors this channel. Every rejected render on every episode came out of my pocket, and the rejects outnumber the keepers by a wide margin. That is not a complaint — it is the actual price of the curation, and curation is the work.
6. Voice, assembly, and the real edit
Narration is synthesised in both English and Portuguese, then the pipeline aligns it to the scenes, builds subtitles and SRT files, and outputs a Premiere package — a real FCP7 XML timeline with every media file linked, one sequence per language.
Then I open it in Premiere and cut the film. Pacing, holds, cutaways, grade, sound design, music, mix, the final call on every frame that stays. The pipeline gets me to an assembly. It has never once made an edit.
So what is honestly AI, and what is honestly me?
AI generated
The pixels. Every image and every moving frame. The narration voice.
Mine
The story selection and the theology. The script in two languages. The scene structure. The shot list, lens choices, and camera movement. The character design and continuity. The prompt for every individual frame. The decision to keep or kill every generated asset. The edit, pacing, grade, sound design, and mix. The money.
Why I am telling you all this
Two reasons. The first is simple honesty: you should never have to wonder whether what you are watching was photographed.
The second is that this is my craft now. I direct films I could not otherwise afford to shoot, using tools that punish anyone who does not know what a shot is supposed to look like. Ten years of production is exactly what makes the difference between this and the noise. If you are a church or a ministry looking at AI tooling and wondering whether it can be done well and done honestly — yes, and this is what it takes.