The Stack

The Build

Declan here. Before this channel had a single subscriber, I spent roughly ninety days building the engine that would feed it. The reasoning is straightforward. A weekly publishing cadence is not a creative problem. It is a logistics problem. If the supply chain breaks down at week six, the channel does not get a second chance to make a first impression. The work that gets built before video one goes live is the work that decides whether the channel survives the first hard month. Building the engine first is not perfectionism. It is the only honest way to commit to a publish schedule that will run for forty weeks without interruption. The same logic applies whether the operation is a faceless YouTube channel, a newsletter, or a software product. Proof of work before public presence.

The image pipeline is the first half of the supply chain. Eight hundred and ninety images were generated through the Grok Imagine API at seven cents per image, for a one-time cost of sixty-two dollars and thirty cents. The architecture is simple. A JSON configuration file defines forty weeks of content, three episodes per week, with seven scene images per episode. A Node script called generate-images.js iterates that configuration, builds the prompt for each scene, calls the Grok endpoint, and writes the result to disk under a deterministic folder structure. The output lives at Channels\The_Exit_Build\Assets\Images\Episodes, with one folder per episode. Inside each folder, the seven images correspond to the seven beats in a typical sixty to seventy second short. When the Remotion pipeline assembles a video, it pulls those images by path, in order, and applies a Ken Burns transform to each. The library is a static asset. It does not need to be regenerated. It does not need to be touched again. Sixty-two dollars buys forty weeks of visual coverage.

The voice pipeline is the second half. The narrator on every Exit Build video is Rex, generated through the Grok TTS API at four dollars and twenty cents per million characters. A typical video script runs around twelve hundred characters, which works out to about half a cent per video in raw voice cost. The decision to use AI voice is not a cost decision. It is an availability decision. Recording a clean take of a sixty second narration, in a quiet room, after a corporate workday, is not realistic three times a week. AI voice is realistic. It is also good enough that nobody is asking for the alternative. The Rex voice supports four speech tags that the channel relies on. Square bracket breath markers cue a deliberate inhale before a weight moment. Square bracket sigh markers cue a release after a corporate grind line. Angle bracket emphasis tags raise the energy on a key word. Angle bracket soft tags drop into an intimate register for the close. Two tags that the rest of the internet uses are deliberately avoided. The slow tag pulls Rex out of his natural pacing. The pause tag breaks the cadence in a way that does not match the brand. The current tag set is the result of iteration, not a starting position.

The Remotion pipeline takes a finished script and produces a published MP4 in roughly forty-five minutes, with the rendering itself running unattended. The exact steps are visible to anyone who wants to see them. Step one is the script written to scripts-txt as a reference file with all markers in place. Step two is a TTS-clean version of the same script, with intro phrases and breath markers stripped, written to a parallel content file. Step three calls generate-audio.js to produce the narration MP3. Step four calls generate-timestamps.js to align the audio against word level captions through Grok speech to text. Step five calls generate-video-json.js to assemble the props the renderer needs, including the image manifest, the caption track, and the theme. Step six calls render-episode.js to produce the final MP4 at ten eighty by nineteen twenty resolution, thirty frames per second. The intro card is a three second navy and brass open with the words "Declan here." in brass over the intro background. The close card carries the brand line in white with the URL in brass. Captions are word by word, accent pill style, generated through the Remotion captions library and trimmed to drop the closing phrase from the on-screen track.

The full cost breakdown for a single video, summed across both halves of the pipeline, is not large. Voice runs about two cents. Five purpose-built images, when needed, run thirty-five cents. Rendering is a local CPU job, so the marginal compute cost is zero. Hosting the rendered MP4 on YouTube is free. Hosting the static blog on Netlify is free at this volume. The only recurring monthly line item is Metricool at twenty-two dollars per month for cross-platform scheduling across YouTube, TikTok, and Facebook. At four videos per week, the operational cost of the channel runs roughly twenty-eight dollars per month, all in. The math is meant to be inspectable. There is no hidden infrastructure. There is no team. There is one person, one laptop, one credit card, and the scripts that connect them.

Every tool listed below has a link. None of the links are ad placements. They are the same tools that produced the video you just watched, in the order in which they actually run. Some of the links are affiliate links, which means I earn a small commission if you sign up. The disclosure is at the top of the page. The reason any of this gets documented is that the only way to make a build log credible is to put the receipts on the table. The numbers will be checked next month against the actual income, and the next month after that, and the next month after that. The point of the channel is not that the stack is impressive. The point of the channel is that the stack is reproducible. Follow along. This is the build log.

The Tools

The Math

The Next Log