01
what I am building
A ChatGPT workflow with a set of skills that takes a topic through script creation, visual planning, narration, image generation, captions, assembly, and validation. I use it to produce reels for my Facebook page while keeping the editorial decisions in my hands.
02
script creation
The script skill turns a topic into a short, concrete story for the audience I am trying to reach. Most of the work happens in the opening, the order of the causal events, and the point where the story pays off.
For recent Gutenberg reels, I have been testing consequence-first openings: show the visible result, create a clear question, then reveal the chain that caused it. The approved script is locked before production begins so an older idea cannot leak into the reel.
03
causal scene plan
Another skill translates the script into a complete panel plan before generating images. Each panel records who appears, what they look like, where they stand, what changes from the previous panel, and whether background people are allowed.
The panels follow the causes in the story. Every image needs to make the next event easier to understand, even with the sound off. This plan carries most of the visual reasoning and keeps characters and locations consistent across the reel.
04
local voice and assembly
The narration is generated locally with Qwen3-TTS using Apple silicon. The voice never needs to leave my computer. The workflow then combines the narration, images, timed captions, and vertical layout into the final video.
The assembly step aligns the spoken words with the captions and keeps them inside the safe area used by Facebookâs interface.
05
validation
The workflow reviews small contact sheets during image production and checks character identity, clothing, extra people, background continuity, and whether each panel still tells the intended story.
Before delivery, it checks the file, duration, narration, caption timing, exact word alignment, and the complete video. It also runs a silent-story check: if the causal sequence cannot be followed without narration, the visual plan is not finished.