← All work
AI video · Social spot

Wakey Bean

An 8-second vertical social ad for a fabricated coffee brand. AI built the clay world and animated the mascot. I directed the character and the packaging, and finished it by hand.

My roleConcept, character and packaging art direction, animation
ToolsClaude · ChatGPT · Runway · Photoshop · After Effects
Format9:16 · 0:08
Year2026

The idea

An empty, sleepy mug slumps on the counter at sunrise. Coffee pours in, it wakes up so happy the coffee jumps out with it, and it lands without spilling a drop. Then: the bag. Built for a muted, scrolling feed, so the story works without sound or captions.

Sleepy empty clay mug slumped on a kitchen counter at sunrise
1. Empty and sleepy
The mug jumps for joy as its coffee flies up in one glossy blob
2. Coffee, then joy
End card: the Wakey Bean coffee bag on orange with the line Wake up happy
3. The product

Finding the character

The first mascot was a clay coffee bean. Two rounds in, a small brown oval with legs wasn’t working, however I styled it. So I changed the character instead of fighting the shape.

Clay coffee bean character turnaround, round one
AI outputRound 1: a lumpy brown bean
Clay coffee bean character with steam, scarf and sneakers, round two
AI outputRound 2: glossier, still not right
Clay coffee mug mascot turnaround: front, three-quarter and side
FixedThe pivot: a mug. Not brown, instantly coffee
Three expressions: sleepy and empty, surprised mid-hop, happy and waving
FixedLocked: one expression per story beat

Style frames

I locked one kitchen first, then reused it for every shot so the set, the light and the camera never changed. These stills are what Runway animated.

Style frame 1: the sleepy empty mug sitting on the counter
1. Sleepy and slumped
Style frame 2: the carafe floats up and pours coffee into the mug
2. The carafe pours itself
Style frame 3: the full mug jumping for joy above the counter
3. Full, and jumping for joy

Social posts

Two Instagram posts to carry the spot onto the brand’s feed: a product still and a looping wave from the mascot. Runway only animated the mug; the logo and type were added in After Effects so the text stays sharp.

How I made it

  1. Step 1Character expression sheet

    Concept

    Claude helped shape the story, the beats and every prompt, and kept the budget on cheap stills instead of video.

    Claude
  2. Step 2Style frame of the pour

    Design

    Character sheet, expressions and three style frames, all made in one ChatGPT thread so the mug stayed the same.

    ChatGPT
  3. Step 3The coffee splash moment

    Animate

    Runway turned each frame into motion. A first-and-last-frame bridge stitched the clips into one continuous shot.

    Runway
  4. Step 4The end card

    Finish

    12 fps stop-motion timing, a circle-wipe into the product and an end card built around the redesigned bag.

    After Effects

What went wrong, and how I fixed it

Same rule as every AI project: it's fast, but it doesn't follow directions. Here's where it drifted and how I steered it back.

Performance

The wake-up went backwards

In the pour clip, the mug's eyes closed blissfully, opened, then drooped back to sleepy before finally waking up. Fix: I cut the step backwards, so the face goes straight from smelling the coffee to sparkly-eyed.

Mug with eyes closed, smelling the coffee
Smells the coffee
Mug drifting back to sleepy, half-closed eyes
AI outputFalls back asleep
Mug wide awake with sparkly eyes
FixedCut straight to awake
Continuity

The jump started mid-air

The second clip began with the mug already in the air, so the cut from the sitting mug popped. Fix: a bridge clip. I gave Runway the exact last frame of clip 1 and the exact first frame of clip 2, and it animated the stand-up and launch in between.

Last frame of clip 1: mug sitting, carafe still pouring
FixedFirst frame: end of clip 1
First frame of clip 2: mug mid-jump, carafe on its coaster
FixedLast frame: start of clip 2
Packaging

The clay bag didn't say "coffee"

AI's clay version looked like a toy, not something you'd find on a shelf. Fix: I redirected it toward a realistic shelf product, with bold marker lettering in the brand colors, and did a draw-over in Photoshop to guide ChatGPT's final version. The world around it stays playful.

AI-generated clay coffee bag on a pedestal
AI outputClay bag: cute, but a toy
The final Wakey Bean bag with orange marker lettering and a sunrise
FixedRedirected: a real shelf product
Motion

AI motion felt too smooth

Runway's movement was floaty and digital, with small morphs between frames. Fix: I conformed every clip to 12 frames per second, the rhythm of real stop-motion. The stepped motion reads as handmade and hides the morphing.

Why clay? I picked the style on purpose. Handmade clay is supposed to wobble a little, so the frame-to-frame drift that ruins photoreal AI reads as charm here.

What I learned

Test the silhouette first

If a shape reads wrong at thumbnail size, no amount of detail will save it.

Pick a style that forgives AI

Clay's handmade wobble hides AI's frame-to-frame drift.

Plan the cuts, not just the shots

First and last frames make separate generations play as one shot.

Real product, playful world

The story can be clay. The thing people buy should look real.