How I produced educational videos and an e-book for lawyers at the WLN Conference in Dubai
Client: Warwick Legal Network
Event: WLN Spring Conference 2026, Dubai, 12–14 February 2026
Scope: 2 educational videos with an AI avatar + e-book The Legal Promptbook
Time: ~70 hours (Video 1: 35h, Video 2: 15h, e-book: 5h, website: 15h)
Costs: AI tool subscriptions I use regularly anyway
Website hosting: Own VPS (OVH) in a Docker environment
AI avatar video production is the process of creating video content in which the narrator is a digitally generated character with speech synthesis. In this article, I describe how I single-handedly produced 2 educational videos and an e-book for an international legal network for a conference in Dubai — in 70 hours, without a studio, without a crew, and without an actor.
It didn't come through active client prospecting or a portfolio website. It came through LinkedIn. Someone in my network, whom I'd met through earlier educational initiatives, recommended me to Warwick Legal Network as someone working with AI in the context of legal practice.
The WLN Spring Conference 2026 carried the tagline "Be Bold. Be DubAI. Why Not?" and featured a panel dedicated to AI in law firms: Between Fear and Frustration: The Use of AI in the Legal Business. For that session, they needed two educational videos and an e-book.
The timing was right. I had just finished consolidating everything I'd built throughout 2025: certifications, first commercial projects, technical infrastructure on my own VPS. I described all of that in the article 2025: Year of Transformation — From Student to AI Architect. This project was the first real test of whether what I'd built actually worked under commercial conditions.
The scope was clear: two educational videos and an e-book. Nothing more. I added the website on my own initiative — I wanted all materials in one place and an easy way for people who watched the videos to reach me. That was my idea, not part of the brief.
The video had to answer the one question lawyers ask themselves most often: does AI threaten me, or help me? To show this concretely — not through slides full of generalities — I built the narrative around two fictional characters.
In scenes featuring Microsoft Copilot, I played the role of a law firm client — Juliusz Tomeczek, a project director at Dubai Marina Developments — who uses AI to manage correspondence and prepare negotiation briefs. In subsequent scenes, with Google Gemini, Claude, and Perplexity, the same character transforms into a lawyer working on documentation for complex corporate disputes.
The narrator is an AI avatar — a woman in business attire, with a British accent. The backdrop: Dubai skylines, the Museum of the Future, and the DIFC financial district. Everything recorded in English.
The script was developed iteratively: Gemini helped shape the strategy and initial structure, Claude handled language refinement and legal nuances, Perplexity served for fact-checking. Every line passed through several models before reaching HeyGen, which gave the avatar its voice.
Production of the first video took 35 hours.
🎬 Film 1 — „From Fear to Strategy"
Not because the material was particularly complex — but because Canva Video Editor was a freshly released beta tool at the time, and I was learning it mid-project on a commercial brief. I restarted the edit from scratch three times. Twice I scrapped everything because unexplainable bugs appeared in post-production: Canva was randomly clipping audio at the beginning and end of every scene. Only the third attempt reached an acceptable level. Each time, I knew more. And each time, I started the edit from zero.
If Video 1 answered "what can AI do for a lawyer?", Video 2 answered "how do you actually talk to these tools to get results?"
This was a particularly meaningful task for me — I finally had the chance to bring together everything I'd been building about prompt engineering over the past year. I've written about it extensively on the blog: about the anatomy of a great prompt, about the template technique (Few-Shot Prompting), about Chain of Thought — thinking step by step — and about prompt chaining in complex projects. Each article was written separately, each around a single technique. Video 2 was the opportunity to connect them into a coherent whole.
The result is the 5 Pillars of Effective Prompting framework, which grew out of my AI training programme for law firms. Pillar by pillar: Persona, Context, Task, Style, and Format. The video explains each one through concrete legal examples — not abstract definitions. We also covered generative AI fundamentals that are often skipped in this type of content: what a prompt is, where hallucinations come from, and what grounding means — anchoring responses in verifiable sources.
The same avatar as in Video 1, this time in a different business outfit. The same tech stack. Production took 15 hours — not 35 — because I didn't restart the edit three times. Everything that cost me time and frustration on the first video, I could anticipate and avoid.
🎬 Video 2 — „5 Pillars of Effective Prompting"
The e-book brought everything together into one reference document. It's not a summary of the videos — it's an extension of them. Video 1 shows what the tools can do. Video 2 teaches how to talk to them. The e-book delivers a ready-to-use prompt matrix for immediate use in a law firm, built on the P.C.T.S.F. framework. Designed in Canva, illustrations generated with NanoBanana. Five hours of work.
I built it using a vibe coding tool — Google AntiGravity. As a civilian developer — someone without a formal programming background — I deployed the site on my own VPS in a Docker environment, the same infrastructure I built in 2025. I integrated it with HubSpot (contact form), Google Analytics, and Bing Webmaster Tools, and optimised it for SEO and SGE. The whole thing took 15 hours.
| Stage | Tools | Purpose |
|---|---|---|
| Strategy & script | Gemini, Claude, Copilot, Perplexity | Iterative scriptwriting and fact-checking |
| Avatar & voiceover | HeyGen | British-accented avatar, speech synthesis |
| Video editing | Canva Video Editor (beta), Movavi Video Editor 25 | Cutting, layers, subtitles, audio |
| E-book & graphics | Canva, NanoBanana | E-book design, AI illustrations |
| Website & hosting | Google AntiGravity, VPS OVH, Docker | Vibe-coded landing page, own hosting |
| CRM & analytics | HubSpot, Google Analytics, Bing Webmaster Tools | Contact form, traffic tracking, SEO/SGE |
| Element | Time | Notes |
|---|---|---|
| Video 1 (Be Bold. Be DubAI.) | ~35 hours | 3 edit restarts, learning the tool from scratch |
| Video 2 (5 Pillars of Effective Prompting) | ~15 hours | No restarts — learning cost already paid |
| E-book (The Legal Promptbook) | ~5 hours | P.C.T.S.F. framework + graphic design |
| Website + integrations + SEO/SGE | ~15 hours | Vibe coding, HubSpot, Analytics |
| Total |
~70 hours |
The difference between Video 1 and Video 2 is the key takeaway here: identical scope, half the production time. The learning curve is a real cost — and you pay it once.
Let me come back to the production time on the first video, because that number needs explaining.
35 hours for an 11-minute video sounds absurd. But this wasn't one edit — it was three complete attempts from scratch. Twice I scrapped everything and started over, because the result wasn't at the level I considered acceptable for a commercial project.
Canva Video Editor was a freshly launched tool, still in beta. Layer behaviour was unpredictable, and some features simply didn't work as intended. The HeyGen avatar was recorded on a green screen — it had to be cut out and placed so that other image layers remained visible beneath the figure. On the first attempt, I thought I understood how it worked. After several hours of editing, I didn't.
Then came the audio problem, which I only discovered halfway through. Canva was clipping the beginning and end of the avatar's lines during export — audio cut off at random points, making scene transitions sound like technical errors. There was no single fix: I had to export, listen, go back, correct, and export again, breaking it into multiple stages to be sure the problem wouldn't return. In the end, the solution wasn't in the editor at all — I had to go back to HeyGen and regenerate the avatars with adjusted parameters. Only then did the audio behave predictably in the edit.
The third attempt was informed. I knew the tool's limitations, knew how to work around the layers, and knew where to check the audio. The video got made — but only after two previous versions went in the bin.
Video 2 took 15 hours instead of 35. Not because it was simpler — the script was similar in length, the materials comparable. It happened because I didn't restart the edit three times. Everything that cost me time and frustration on the first video, I could anticipate and avoid.
For anyone who wants to replicate this process — here's a simplified workflow:
The e-book The Legal Promptbook and Video 2 are both built around the P.C.T.S.F. methodology. It's my own prompt structure, developed over a year of working with various AI models. It didn't come from any course. It emerged from trial, error, and observing what actually produces consistent results.
| Element | What it means in practice |
|---|---|
| P — Persona | Tell AI who to be: senior partner, DIFC law expert, legal translator |
| C — Context | Describe the situation: case, document, parties, what you already know |
| T — Task | Define the task: what specifically to do, not what to talk about |
| S — Style | Set the tone: formal, analytical, concise, persuasive |
| F — Format | Specify the output format: list, table, memo, email, quotable excerpt |
Download the e-book The Legal Promptbook
a ready-to-use prompt matrix for law firms.
All materials free at bedubai.aiforeveryone.com.pl
The conference took place in Dubai, 12–14 February 2026. The videos were screened for panel session attendees. Full feedback hasn't arrived yet — these things take time. But I already know a few things.
An AI avatar in business attire, speaking English with a British accent, against a Dubai skyline backdrop, held up in a demanding legal environment. There was no uncanny valley effect — the psychological phenomenon where a character that looks almost human, but not quite, triggers unease and discomfort in the viewer rather than connection. I had worried the avatar would distract rather than engage. It didn't. The format delivered.
The website is collecting contacts through the form and generating organic traffic. SEO and SGE are long-term investments — the full results will show in a few months.
The most important thing I took from this project: AI avatar video production at a corporate client level is possible solo — without a studio, without a crew, and without a production budget. It's not easy or fast the first time. But it is possible. And with every subsequent project, it becomes more repeatable.
The videos, e-book with a ready-to-use prompt matrix, and contact form are all available on the project website. The materials were prepared for WLN conference attendees, but they're open to everyone.
If you work in a law firm and are wondering how to start working with AI — visit bedubai.aiforeveryone.com.pl. Video 1 shows what AI does. Video 2 shows how to talk to AI. The e-book gives you ready prompts to start with. All free.
If your clients are already using AI to prepare for meetings with you — and they increasingly are — the question is no longer "whether to implement AI in the firm." It's: who will do it well, professionally, and in line with professional ethics?
This project shows it can be done solo — without a studio, without a crew, and without a production budget. The first time isn't easy or fast. But it is possible. And with every next project, it becomes more repeatable.
Be Bold. Be DubAI. Why Not?
In this project, the cost was AI tool subscriptions (HeyGen, Canva Pro, Claude Pro, Gemini via Google Workspace) that I pay for regularly anyway. No studio, camera operator, or voice actor costs. Time: 35 hours for the first video (including learning the tools), 15 hours for the second. On repeat projects, time drops by 50–60%.
Yes — provided the style is right: business attire, professional accent, neutral backdrop. In this project, a HeyGen avatar with a British accent against the Dubai skyline was presented to lawyers at the international WLN conference with no uncanny valley effect.
P.C.T.S.F. is a proprietary prompt structuring methodology: Persona (who AI should be), Context (case context), Task (specific task), Style (tone of voice), Format (output format). The framework was built from a year of hands-on practice with various AI models and forms the basis of the Legal Promptbook e-book.
Minimum stack: HeyGen (avatar + speech synthesis), a video editor (Canva Video Editor or Movavi), an AI model for the script (Claude, Gemini). Optionally: Perplexity for fact-checking, Canva and NanoBanana for graphics, own hosting for the landing page. The entire project can be delivered solo.