AI generated video ads for Geepas, made from existing product photography
Project 54, led by Michael S.U. Hudson, produced AI generated video ads for Geepas UK from the brand's own product photography, with no shoot. The client approved the ads. Rollout through paid media was not taken forward.
Outcome
5 days
Planned production window for the ads, about 22 hours of production, with no photo or video shoot.
Market
United Kingdom, English
Formats
9:16 for TikTok and Instagram Reels, with YouTube Shorts specifications prepared
Source
Project 54 production plan and delivery, April 2026
Situation
Geepas sells more than 1,500 small appliance, homeware, heating and cooling, and personal care products. Its UK business sells through its own online store and major retailers. It wanted short form video ads without the cost and time of a product shoot.
The problem
AI video tends to redesign products: invented buttons, drifting logos, wrong proportions. For an appliance brand the product must be exactly right in every frame, or the ad cannot run.
Role
Michael led the work as Chief AI Officer of Project 54: the research, the concepts, the production method and the quality control.
What was built
Research: UK competitive intelligence across 18 competitors and 16 channels, and research on AI video advertising.
Six concepts: three brand ads pairing a kettle and toaster, and three user generated style ads for a tower fan, each with a shot list and a creative rationale grounded in short form ad research: hook timing, delayed product reveal, problem and solution.
Production: start frames generated from the client’s own product photography, with prompts that lock the product to the reference photos; image to video animation with motion only prompts; then edit, film grain, sharpening, colour grade, a British voiceover and music.
Accuracy method: the product photography is treated as ground truth. Prompts require the product to stay identical, body, controls, display, wordmark and proportions, and forbid redesign. Brand ads use hands and silhouettes instead of faces, removing the highest risk failure. Every output was checked against the reference photos before delivery.
Human in the loop: people selected the source photography, approved every start frame, chose the keeper clip from three to five generations per shot, and approved the final edit before the client’s own approval.
Outcome
The client approved the ads. Rollout through paid media was not taken forward.
What it shows
AI can turn existing product photography into platform ready video ads in days, if product accuracy is engineered into the method rather than checked at the end.