Kazi Yousuf — Dhaka, Bangladesh

Character-consistent AI video, built shot by shot.

Voice to prompt to generated shot to final cut — one person, start to finish.

Selected work

Four projects, two disciplines — AI-generated video work and a weekly explainer series, both written, produced, and edited solo.

In Production

The CEO's Forbidden Assistant

A two-character vertical microdrama — Lena and Damien, one Executive Office, one night.

Episode 01: nine shots, ~62 seconds of raw footage, cut toward a 60–70 second final that ends on a cliffhanger. Hero references locked in Nano Banana Pro, each shot generated in Veo 3.1 with dialogue voiced natively in-shot. Currently in edit.

Veo 3.1Vertical 9:16MicrodramaOpen

Completed

UGC Talking-Head Ad Campaigns

A locked photorealistic spokesperson character carrying two full campaigns.

Character locked once in Google Flow, then generated per shot in Veo 3.1 Lite and cut in CapCut. Video 1: a four-shot DTC spec ad, ~29 seconds, hook to problem to product reveal to CTA. Video 2: a fourteen-shot gig intro, ~74 seconds. Both fully produced and edited.

Veo 3.1 LiteUGC2 CampaignsOpen

In Production

Spec Ad Portfolio

One locked 2D-illustrated character running across six spec ads, three SaaS and three DTC.

FlowStack, PitchBird, and LedgerLoop for SaaS; SLUMBR, LUMENA, and FetchBox for DTC. Five shots per ad, generated with the Wan 2.2 14B image-to-video pipeline, vertical 9:16. Ads one and two have shots and stills generated, with animation underway — four more to go.

Wan 2.2 14BVertical 9:166-Ad SeriesOpen

In Production

Blindside Effect

A weekly psychology and cognitive-bias explainer channel — written, voiced, and edited solo.

Five episodes covering biases and behavioral effects — Dunning-Kruger, the Halo Effect, Murphy's Law, and more. Separate from the AI-generated character work above: scripted, recorded, and cut by hand, released on a weekly cadence.

YouTubeExplainer Series5 EpisodesWatch on YouTube

About

Script to voice to generated shot to final cut — one continuous pipeline, run solo.

Education

Higher Secondary Certificate

National Ideal School & College, Dhaka — 2025

I build video without a camera. Every project starts as a script and ends as a cut — voiceover written and timed, references generated and locked, each shot generated frame by frame, then edited into a final sequence. I run all of it myself: character design, prompt writing, generation, and the edit.

Two things I hold to on every project. The character doesn't drift — same face, same outfit, shot to shot. And the visual matches the words: if the line says she opens the drawer, the shot shows her opening the drawer, not a mood board standing in for it.

I'm based in Dhaka, Bangladesh, building toward freelance AI video production work. Everything on this site is real, current work — flagged honestly by status, finished or not.

Pipeline

ComfyUI (local): Wan 2.2 14B & MiniMax H3 image-to-video, self-hosted
MiniMax H3: Primary generator — ~3–4 min per 5-second shot
Wan 2.2 14B I2V: Fallback generator, 24fps native duration
Google Flow: Character locks and reference images
Veo 3.1 / 3.1 Lite: Hero shots and native in-shot dialogue
ElevenLabs: Voiceover generation
CapCut / Premiere Pro: Editing and final assembly

CineFlow

Self-built, still evolving

I built my own tool for the repetitive parts of this pipeline — it turns a script into timed voice tracks, generation prompts, and a draft assembly automatically, so multi-shot projects stay organized instead of managed by hand. It's a working personal tool, not a finished product — I keep extending it as new projects need it.

Want the short version?

Everything above, condensed to one page.

Download résumé (PDF)

Preview not loading? Open the PDF directly.

Start a project

Available for freelance AI video production work. Reach out directly.

AI UGC / talking-head adsProduct & spec adsCharacter-driven short-form videoExplainer & narrative video

kaziyy999@gmail.com

linkedin.com/in/kaziaiops

GitHub

github.com/kaziaiops

+880 1410-216644