An applied assignment for a mobile-game user-acquisition studio: a 37-second vertical performance ad for TikTok and YouTube Shorts, concepted and produced end to end with a custom AI-and-Python pipeline — from narrative hook to generated footage, dialogue QC, and a final assembled cut with captions and a download CTA.
The studio's assignment was to create a short-form user-acquisition ad for a survival-shooter mobile game, demonstrating how AI tooling could be used across the creative and production process. The deliverable was an actual finished video — not just a concept — built for vertical placements on TikTok and YouTube Shorts, with the reasoning behind every content and tooling choice documented alongside it.
Built a narrative concept following three ordinary friends in a zombie outbreak. The hook lands in the first few seconds: one confidently makes a terrible survival decision, nearly gets everyone killed, then the group escapes — an instantly readable joke with a jump-scare for punch.
Sequenced a clean short-form rhythm: mistake-driven opening hook, escape and hiding, progression through shelter-building and weapon prep, a wave warning, then a direct gameplay reveal ending on a download CTA with store badges.
Chose a relatable, story-based angle over the usual direct-gameplay or fail-minigame ads — giving the ad a stronger reason to be shared, more room to test creative variety, and a clearer tie back to the core game loop of survive, build, and defend.
Used AI coding and writing assistants to structure the concept, refine the beat flow, write scene prompts, and develop the production pipeline — defining character setup, transitions, camera direction, and continuity for each scene.
Generated the 9:16 scene clips with an AI video-generation model, writing detailed character locks and negative prompts to keep the cast, tone, and visual continuity consistent across every beat.
Used the model's generated audio, then ran speech-to-text transcription as a quality-control step to verify spoken lines matched the intended script and to catch unclear or mismatched dialogue.
Assembled the generated footage with a non-AI Python pipeline — video editing and encoding in one library, rendered captions, UI-style overlays, store badges, and the end-card CTA in another — for full control over the final format.
AI-generated clips drift between scenes, so character locks and negative prompts were essential to keep the same cast and tone across the full ad.
Generated audio can be inconsistent, so a transcription-based QC step was built in to confirm the spoken lines actually matched the script.
The ad had to stay funny and shareable while still setting honest expectations about the real game — survival, base progression, and wave defense.
This was a spec assignment completed during an interview process. The studio name and game title have been anonymized. The concept, prompts, production pipeline, and final edit are my own original work. The embedded video is a de-branded edit of the finished ad.
This project shows that I can run AI-assisted creative end to end — from a sharp short-form concept through a controlled generation-and-assembly pipeline to a finished, deliverable performance ad.
