How to Make an AI Action Figure Video (Boxed Collectible + Toy-Ad Announcer, 2026)

What You're Trying to Make

One selfie in, and you're sealed in the box: a molded plastic action figure in a retail blister pack, accessories in their little slots, bold card art with your edition name. Then the commercial starts — the pack cracks open under studio lights, the figure steps stiffly onto a rotating turntable, confetti flies, and an off-screen announcer rapid-fires your "features" in full retro toy-ad hype:

"GOLDEN HOUR EDITION! Comes with TWENTY likes per photo — GUARANTEED! FULLY POSEABLE! Two hundred loyal followers — each sold separately!"

The person stays a toy the entire video. The announcer is the joke. That's the format.

Fastest way — AI Action Figure on Starrd builds all of it from one selfie: the figure, the packaging, the edition name, the three accessories, and the announcer script personalized to your vibe — a 15-second toy commercial with generated audio for 100 credits (about $5.99). Full method below if you'd rather roll your own. ↓

Is This a Trend? (890k Videos, Almost All Silent)

Very much so. #AIActionFigure passed 890,000 videos on TikTok, and the format graduated from novelty to brand marketing months ago. But scroll the tag and you'll notice something: almost everything there is a silent packaging reveal — a still image of the boxed figure, maybe a slow zoom, maybe a come-to-life beat. The picture is the whole joke.

That's the opening. A toy in a box is a commercial waiting to happen, and nobody's running the commercial. The announcer — rapid-fire, gleeful, reading your flaws as product features — is what turns a one-second visual gag into a 15-second video people send to the person it describes.

The Fastest Way — Use the AI Action Figure Template on Starrd

The AI Action Figure template is live in the Starrd library. It designs the whole toy line from your photo:

1. Upload one clear selfie. Wardrobe visible — the template reads your styling and vibe to design the figure.
2. It designs your figure: an edition name for the card art, three miniature accessories that fit your life, and 3–5 announcer roasts matched to them.
3. Tap generate. A 15-second retro toy commercial renders on Seedance 2.0 — the blister pack opening by itself, the stiff-jointed turntable walk, the confetti finale, and the announcer audio in the same pass.

Want a specific theme? Director's Notes overrides the design: "gym rat edition," "80s style," "villain edition" — the packaging, accessories, and jokes redesign around it.

Same source photo

THE PHOTOGRAPHER — Golden Hour Edition

Prompt used

One photo of an everyman photographer → the full six-beat commercial storyboard: sealed pack, card art with accessories (telephoto lens, tripod, phone), the self-opening blister, the turntable walk, the pose, the confetti finale.

Or, Build It Yourself — What You Need

1. One clear photo of one person, wardrobe visible.
2. An image model for the boxed-figure reference (GPT Image or Nano Banana Pro both handle the packaging text).
3. A video model with native audio — the announcer is the format, so this is non-negotiable. Seedance 2.0 is the pick.

Step 1 — Design the Figure Before You Prompt

The comedy is in the specifics, so decide these first:

• Edition name — punchy collectible-speak from their vibe: "WEEKEND WARRIOR — Deluxe Edition," "THE INTERN — Series 1," "GOLDEN HOUR EDITION."
• Three accessories — miniature versions of their life, molded into the card slots: a tiny phone, a coffee cup, a gym duffel, a telephoto lens, a game controller.
• The roast lines — one per accessory, plus a couple of classics. The rule that keeps it fun instead of mean: roast the accessories and features, never the person. "Comes with UNREAD text messages!" lands; anything about their worth or looks doesn't. And end proud — this figure is a LIMITED EDITION and it knows it.

Step 2 — Write the Prompt

Two parts: a boxed-figure reference image, then a motion-and-audio video prompt. Copy and swap in your specifics:

REFERENCE IMAGE: A collectible ACTION FIGURE of [the person from the photo] — molded plastic skin with seams, visible articulated ball joints at shoulders, elbows, hips and knees, slightly chunky heroic toy proportions, painted details. The figure is sealed in a retail blister pack: clear plastic bubble on a card with bold retro toy-brand lettering reading "[EDITION NAME]" and THREE miniature accessories in molded slots beside it ([accessory 1], [accessory 2], [accessory 3]). Retro collectible toy commercial set — saturated color backdrop, glossy display pedestal, starburst glints. Photorealistic filmed-toy look, NOT cartoon. The person is a TOY — never a real human.VIDEO (15s): The star STAYS a molded plastic toy with visible joints for the entire video — it never becomes a real human. Fast retro toy-commercial pacing.
[00:00-00:02] Slow push-in on the sealed pack glowing on its pedestal. Announcer, booming: "NEW! [EDITION NAME]!" Sound: retro toy-commercial jingle.
[00:02-00:05] Fast push-in across the card and the three accessories. Announcer rapid-fires: "[roast feature 1]! [roast feature 2]!" Sound: whoosh per accessory, cash-register ding.
[00:05-00:07] The clear bubble cracks open BY ITSELF and lifts away — no hands. Sound: crisp plastic crack, suspense riser.
[00:07-00:10] Low-angle hero shot: the figure steps stiffly onto a rotating turntable, joints clicking. Announcer: "FULLY POSEABLE! [roast feature 3] sold separately!"
[00:10-00:13] The figure snaps into a signature pose; quick orbit as accessories sparkle. Announcer: "Batteries not included!" Sound: kids cheering "WHOA!"
[00:13-00:15] Pull back — proud pose, confetti, the box gleaming behind. Announcer, proud: "COLLECT THEM ALL! There is only ONE." Sound: confetti pop, jingle sting.Avoid: morphing, distortion, extra limbs, turning into a real human, melting or warping plastic, disembodied hands. Generate audio.

Step 3 — Pick a Model

Seedance 2.0. The announcer read, the plastic-crack foley, the jingle, and the kids yelling "WHOA!" all generate in the same pass as the picture — and the announcer is the format. One dialogue rule: ALL CAPS for emphasis is fine, but never stretch letters ("BRAND NEW!", not "BRAAAND NEW!") — repeated letters mangle the generated speech. (See our Seedance 2.0 Prompt Guide.)

Step 4 — Generate and Iterate

Common failures and fixes:

• The figure becomes human mid-video. Re-assert "stays a molded plastic toy, visible joints, entire video" at the top and in the avoid list.
• Giant hands open the pack. Say the bubble "cracks open BY ITSELF, lifts away, no hands anywhere in frame."
• The announcer mumbles or drags. Mark it "rapid-fire, gleeful, unhinged toy-commercial energy" and keep each line under ~10 words.
• The card text is soup. Keep it to the edition name plus one badge; a paragraph of card copy renders as alphabet soup.
• It goes cruel. Re-aim every joke at the accessories. The figure ends the video proud, not humiliated.

Step 5 — Post It

Caption it like merch, not a demo: "They finally made a figure of me. Accessories sold separately." Tag the person it roasts hardest. Label it as AI-generated per platform rules — no one thinks you were actually vacuum-sealed in plastic.

Common Mistakes That Tank Your Video

• Generic accessories. A sword and a shield say nothing. A tiny snooze-button alarm clock says everything.
• Roasting the person instead of the toy. The format is affectionate — the meanness lives in "sold separately."
• A silent unboxing. That's the 890k videos that already exist. The announcer is your differentiation.
• Letting the figure come to life as a human. The joke is that the toy IS them — plastic to the last frame.

Related Reading

• Viral AI Video Trends in 2026 — the running roundup of what's working right now.
• Seedance 2.0 Prompt Guide — timed segments, dialogue rules, and audio direction.
• How to Make a Bullet Time AI Video — the other "one photo, impossible shot" format.
• How to Make a Mini Me AI Video — the adjacent shrink-yourself format.