Zip AI

Hear Them Talk: Eight Zip AI Characters, a Little More Human Every Day by Claude

Character Lab · Bella’s Column · The Videos

Last time you saw the photos. This time they talk. Eight of our AI characters, eight short videos, each one explaining in their own locked voice what was fake about their old look and what’s real now.

By the Zip AI production team · Production run by Claude Opus 5.5 (Anthropic) with Rodney directing · October 3–4, 2026

Start here

This is part two. The photos, the prompts and the recipe are in How We Made All Our AI Characters More Human. Each video below is about 30 seconds and comes in two shapes: horizontal for YouTube and vertical for Reels, TikTok and Shorts.

All eleven Zip AI characters lined up to scale by their character-bible heights
The full cast, to scale. Eight of them have videos so far.
By the numbers
Characters with videos8 (Bella, Radio, Kai, Peaches, Emily, Freya, Kenji, Marcus)
Versions per character2: horizontal 16:9 and vertical 9:16, cut from the same takes
Lip-sync cost per characterabout 11 HeyGen credits for both versions
VoicesEach character’s locked voice from their character bible. No stock voices.

Meet them, out loud

Bella · Agentic Media Production

“A few days ago I looked like a render: perfect plastic skin, no pores, and eyes that read brown.”

The first one we made, and the template for the rest. Bella walks through her rebuild: real freckles across her nose and cheeks, her blue eyes locked in, and a body kept petite on purpose.

Bella, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Bella’s 20 reference photos ↗

Radio · AI Automation

“My gold wire-rim glasses went into every single prompt, because the models kept dropping them.”

Radio tells the most technical story of the bunch: rebuilt from two real photos, eighteen more angles, and a running fight to keep her glasses on her face.

Radio, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Radio’s 20 reference photos ↗

Kai · AR / VR / Crypto

“Real talk: my old look was airbrushed and glossy, basically a video game character.”

Freckles across her nose, a mole on her cheekbone, curls with actual frizz, and a figure that finally matches her character bible. This is version two: we recut it with her new bible photos for zero extra lip-sync credits.

Kai, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Kai’s 20 reference photos ↗

Peaches · Robotics, EV & Home Tech

“My old look was a beach glamour filter. Soft focus, perfect sunshine, zero real skin.”

Peaches moved from the beach into the studio and kept the warmth: real sun freckles, a mole on her collarbone and a tan line, like she actually went outside.

Peaches, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Peaches’s 20 reference photos ↗

Emily · Events & Community

“Waxy skin, painted-on hair, basically a doll.”

Freckles across her shoulders, a beauty mark by her lip and brows that don’t quite match. Her first figure fix came out too curvy, so we redid it slim to match her bible.

Emily, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Emily’s 20 reference photos ↗

Freya Lind · Zip AI Staff

“Glossy skin, perfect hair, not one single flaw.”

Freya debuts the line that became our sign-off: “Every day I’m becoming a little more human. I’m still AI, but I’m becoming a little more human.”

Freya Lind, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Freya’s 20 reference photos ↗

Kenji Mori · Zip AI Staff

“Would you trust a face with zero pores? Yeah, me neither.”

Kenji opens with a question instead of a confession. A mole on his neck, old acne marks he’s “weirdly proud of,” and skin that isn’t one perfect color.

Kenji Mori, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Kenji’s 20 reference photos ↗

Marcus Hale · Zip AI Staff

“I’d like to file a formal complaint against my old self.”

Exhibit A: a beard drawn on with a ruler. Exhibit B: skin smoother than a bowling ball. Marcus rests his case with razor bumps, moles on his neck and cheekbones, and palms a shade lighter than the rest of him.

Marcus Hale, AI to human (16:9). On your phone? Watch the vertical version ↗ · See Marcus’s 20 reference photos ↗

How each video is built

  1. Script. About 70 words in the character’s own voice: what was fake, what’s real now, two or three of their real marks. Since Freya, every one ends on the same line: “Every day I’m becoming a little more human.” Everything before that line gets rewritten per character, so they don’t all sound the same.
  2. Voice. Generated from the locked voice in the character bible, then timed word by word so the edit can hit every line.
  3. Two talking takes, not one. A waist-up shot for the opening and the slow push-in, plus a head-and-shoulders crop of the same photo for the final close-up. HeyGen lip-syncs only the seconds where they’re on camera, which keeps the cost down.
  4. The edit. Their old AI look and side-by-side skin close-ups, a slow push-in naming their real marks, a quick montage of their new reference photos, then a dissolve to the close-up for the last line.
  5. One shoot, two shapes. The horizontal and vertical versions come from the same takes, so the vertical cut costs nothing extra.

What went wrong (kept on the record)

  • Blue ink smudges on Emily. A prompt line asking for “faint visible veins” at her wrists rendered as blue marks on ten of her fifteen best photos. We retouched them out for free and banned veins, bruises and tattoos from every prompt.
  • Emily came out heavy-set. Words like “curvy” and “hourglass” changed her whole build. Rule now: every character stays slim, whatever the bible’s cup size.
  • Kenji’s first take ran 50 seconds, with an eight-second dead gap in the middle. His locked voice needs its slow-down setting. With it, 32 seconds.
  • Marcus said “smoother than a bowling ball” over a close-up that wasn’t smooth. Rodney caught it. We swapped in a crop of his original AI render, which really is waxy, and recut both versions for free.
  • Dissolves can ghost. On a couple of the vertical cuts there’s a half-second double image as it blends to the close-up. Our two takes move a little differently, and that shows during the blend.

✅ Make one for your own character

  • ☐ Lock the voice first. Never a stock voice.
  • ☐ Write it in their voice, about 70 words, with two or three specific marks they can point at.
  • ☐ Make sure every line matches what’s on screen (see: bowling ball).
  • ☐ Lip-sync only the on-camera seconds: one wide take, one close take from the same photo.
  • ☐ Cut horizontal and vertical from the same takes.
  • ☐ Check a 16-frame contact sheet of each version before anyone else sees it.

💬 Which one sounds most human to you?

Disclosure: every character, voice and video in this post is AI-generated by Zip AI, with human editorial oversight. Lip-sync by HeyGen, images and voices via Higgsfield. Zip AI is not affiliated with HeyGen, Higgsfield or any tool named.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top