Two Minutes, Built From Everything
Video case studies for MMA Smarties submissions, InMobi
Concept & Intent
Every campaign has a story bigger than its media plan. My job was to compress that story, the strategy, the creative, the real results a brand achieved with InMobi, into something a Smarties judge would watch in under two minutes and actually remember.
That meant these weren’t edits. They were builds.
Each video blends three different sources into one continuous piece: real footage shot on the campaign, the actual campaign creatives as they ran in market, and AI-generated video sequences created to bridge the gaps. The transitions, the moments no camera was there to capture, the connective tissue that turns a set of assets into a story. An AI-generated voiceover carries the narration throughout, letting the visuals stay tightly paced without waiting on studio time.
Nobody watching should be able to tell where the real footage ends and the generated footage begins. That was the brief I set myself.
Why build it this way
Campaign submission videos live and die on two things: how well they tell the story, and how fast they can be turned around before a deadline. Traditional production couldn’t give me both.
AI-generated video
AI-generated video filled in continuity, the shots that never existed but that the story needed, keeping the narrative moving instead of cutting hard between whatever raw footage happened to be available.
AI-generated voiceover
AI-generated voiceover meant scripts could be revised until the last hour without booking a studio, and every video could carry a clean, consistent narration track without the cost or delay of a VO artist per market.
Image Model:
Chat GPT 2
Video Platform:
Kling, Google’s Omni, and Seedance
Voice Over Model:
Eleven Multilingual v3
Editing
Adobe Afer Effects, Adobe Premiere Pro
The process
A single 90-second video moved through several hands before it was one continuous piece of film:
Concept & stills
Base imagery generated to match the campaign’s visual world.
Prompt engineering
Prompts refined with Claude and ChatGPT to translate a still concept into a directable video prompt: camera movement, pacing, mood.
Video generation
Motion generated from those stills across multiple models, Kling, Google’s Omni, and Seedance, each used where its strengths fit the shot.
Assembly
Generated sequences cut together with real campaign footage and in-market creative, scored with an AI voiceover, and paced to land the story in under two minutes.
Every video went through this pipeline differently, shot by shot. Some leaned more on real footage, some needed the AI sequences to carry most of the narrative. Knowing when to use which was most of the job.
Getting the generated shots to pass as real was its own discipline. Lighting, skin tones, camera shake, the small imperfections that make footage feel shot rather than synthesized, all had to be pushed for deliberately across Kling, Omni, and Seedance, then matched against the actual campaign footage in the edit. A shot that looked impressive on its own but broke realism next to real footage got rejected and regenerated. The goal was never “good AI video.” It was footage that disappeared into the story.
A note on craft
None of this was about hiding behind AI tools. It was about using the right tool for a very specific constraint: real campaign footage is finite, deadlines aren’t, and a Smarties judge has about ninety seconds of patience. The craft was in the invisible seams: knowing which model to reach for on which shot, how much AI narration a story could carry before it stopped feeling human, and how to cut real and generated footage together so the story never breaks stride.
If it looks simple, that’s the point.