Last week, Dartmouth researchers analyzed 146,000 patient portal conversations and reported that AI-drafted replies often take physicians longer to fix than writing the reply from scratch. A tool you have to babysit is not leverage. It is added work with extra steps.
I made a 23-second video about that study. Claude scripted it, Higgsfield animated it, ElevenLabs narrated it in a clone of my voice, and Blotato staged it for Instagram. My total contribution was one paragraph of direction and two style corrections. Both corrections were written into the system’s standing instructions, so the next video applied them without my asking.
That last detail is the goal of this article. Everyone knows AI can generate video. The hard problem is consistency and style: making video number forty match video number one in voice, palette, logo treatment, and caption style, while you are in the OR. This is a systems problem, and it has a systems answer.
Why this matters beyond content creators
Across major platforms, roughly 85% of mobile video is watched with the sound off, and on LinkedIn the figure runs near 80%. Captions are not an accessibility afterthought; for most viewers they are the video. Short-form video also earns about 2.5 times the engagement of long-form on social platforms, and marketers consistently rank it their highest-ROI content format.
The case for clinicians and academics is more specific than “build a brand.” In the TSSMN randomized trial in the Annals of Thoracic Surgery, 112 published articles were randomized to be tweeted or not; at one year, the tweeted articles had accumulated significantly more citations and higher Altmetric scores than controls. Dissemination is now part of scholarly impact, measurably. The same logic applies to health systems: service lines compete on visibility for referrals, clinical trial recruitment depends on reach, and policy influence follows the people who explain the policy first. My own numbers are consistent with the pattern. Posts that demonstrate a working tool chain convert views to paid subscribers at roughly 2%, five to ten times the rate of commentary posts, and that audience is where consulting, speaking, and partnership inquiries originate.
The constraint has never been whether visibility is worth building. It is that a single well-edited reel costs two to four hours of skilled labor: scripting, recording, editing, captioning, exporting, posting. At three reels a week, that is a part-time job. The pipeline below reduces it to minutes of review time per video.
The stack
Two design decisions steward the system to operation. First, the style key: a single reference image attached to every video clip so the visual register never drifts between generations. I don’t do this manually, I have higgsfield generate the image from my prompt then instruct consistency and bake it into the skill. Character-consistency studios use the same technique for faces; here it is applied to brand aesthetics. Second, brand memory: every correction gets written into the instructions the next run reads. Taste accumulates in the system instead of depending on my attention that morning. Once you’ve made a branding skill for your particular company logo or project, the reels can get made in your style, just mention in your prompt or your scheduled task that you want claude to utilize [inert your brand here] brand kit in the output.
I also highly recommend Blotato and the MCP from Sabrina Ramonov 🍄 . I was not sure it was something I would use, but embedded into Claude Code and Claude CoWork it has helped me in a short period automate much more of my social media and med ed endeavors across a few businesses and projects now. Check out instructions in her article.
Same pipeline, different register
The visual register or “style” is a parameter that you can adjust easily. The reel above uses my default: bright, glassy, product-forward. The one below is the same pipeline pointed at a handcrafted felt-diorama aesthetic, for a story that benefits from a lighter touch. The subject is the Nature Medicine benchmark study in which general-purpose models outperformed the AI tools built specifically for medicine: Gemini 3.1 Pro scored 97.4% on MedQA to OpenEvidence’s 89.6%, and the gap on HealthBench was wider (88.0 for ChatGPT vs. 62.6 for OpenEvidence). OpenEvidence disputes the methodology and has asked for a retraction; the dispute is worth reading in full before you cite either side. As long as you know what style you’re looking for you can create videos in a myriad of styles (I’ll publish my carousel and styles index later, although most AI tools have references you can use. I have one style index below).
🔒 The operator’s playbook continues for paid subscribers
Below the line: the six-step setup, the two installable skill files, the caption spec including the two bugs I shipped so you don’t have to, the scheduled-task structure, per-reel costs, and the single prompt that runs the whole workflow.
Not interested in subscribing but still want the skills? The complete Carousel & Reel Video Skills Package is available on its own at techysurgeon.com/skills for $79, including the video walkthrough for connecting Higgsfield to Claude.
Sample looks styles index from Techy Surgeon carousel kit.
Before we get into the tools and skills, I have no affiliation but I do recommend folks check out Blotato, which has its own in tool AI video generation capabilities and reel and carousel styles. The MCP has been great, and I’m only now getting started with the tool. I I was actually able to eliminate Higgsfield from the flow and, with a single prompt from within CoWork, generate a branded reel, post it to YouTube, Instagram, and TikTok in one fell swoop for someone who doesn’t need as much control over their video and doesn’t mind having another voice. Blotato may be all you need.
