Realism Evaluation of AI-Generated Ads

I did NOT expect this level of realism from an AI-generated ad. 😳

Made using MiniMax H3 by @Hailuo_AI 

H3 is a general-purpose multimodal video model that goes far beyond single-purpose tools. It understands text, images, video, and audio together, allowing it to better understand creative intent and generate natural, coherent audiovisual content in one end-to-end workflow.

One detail that really stood out to me was how well it handles on-screen text. The typography stays sharp, clear, and consistent throughout the video, with little to no distortion or deformation—something that's still a challenge for many AI video models.

#MiniMaxH3

Prompt:
CHARACTER & WORLD LOCK + FULL 15-SEC MASTER PROMPT

A 15-second continuous UGC-style vlog in 9:16 vertical format. Jessy, a woman in her mid-20s — use her facial features exactly as shown in her reference photo, keep everything else about the scene freshly designed — wears an attractive pastel-colored dress (soft lavender, mint, or buttery yellow) that visually pops in contrast against the bubblegum-pink Olipop can. Location: a bright, clean modern kitchen/countertop with soft natural daylight. Product: a can of Olipop Blackberry Vanilla — pastel pink can, magenta blackberry illustration, white vanilla flower, dark green "Blackberry Vanilla" wordmark, "Supports Digestive Health" text near the top, "A New Kind of Soda™" tagline (use the uploaded product photo as the exact visual reference for the can). Style: authentic smartphone UGC vlog, natural handheld feel, realistic skin texture, no over-retouched look. She speaks casual, upbeat, conversational English directly to camera with natural lip-sync.

The video follows this exact beat sheet, with clean cuts between each moment:

00:00–00:02 — Front-camera selfie, slight handheld sway, Jessy fans herself lightly in the kitchen, looking for something refreshing, smiles at camera: "Okay I need something fun to drink right now."

00:02–00:04 — Eye-level medium shot, she opens the fridge and pulls out the pink Olipop can, face lighting up with delight: "Ooh, Blackberry Vanilla Olipop — yes please."

00:04–00:06 — Close-up on her hands slowly turning the can, label details ("Prebiotics, Botanics, Plant Fiber") visible: "Prebiotics AND it tastes like dessert? Sold."

00:06–00:08 — Tight macro shot, her fingers pop the tab, condensation and a light fizz mist visible: "Listen to that fizz."

00:08–00:10 — Slow-motion macro shot, pink soda pouring over ice into a clear glass, bubbles rising: "Look how pretty that color is."

00:10–00:12 — Close-up on her face as she sips, eyes lighting up with genuine, pleasantly surprised reaction: "Wait... that's actually so good."

00:12–00:13 — Macro hero shot of the can alone on the counter, condensation dripping, soft bokeh kitchen background, slow subtle push-in, no dialog.

00:13–00:14 — Medium shot, she smiles at camera, can raised slightly at chest height: "Blackberry vanilla was the right call."

00:14–00:15 — Close front-camera selfie, can held beside her face, warm bright smile into the lens: "Go grab an Olipop, seriously."

Continuous natural ambient kitchen sound, subtle upbeat pop background music underneath, her voice clear and casual throughout. Consistent face, outfit, and setting across every beat. No text overlays, no captions, no watermarks, no horizontal framing, no changes to the can's design or branding.