Start with 1 task or 10 hours — scale when ready. View pricing →
AI video & visuals · 3 projects · jewelry, publishing, a viral idea

AI where a shoot is too expensive, impossible or not needed.

Three very different clients, one approach. I build the key frames by hand and let AI do only what it does well: animate, extend and scale. The result looks crafted, not generated.

February to October 2026Three projects, two still running
Static + videoSocial posts, trailers, teasers, short-form
My scopeVisual system, key frames, prompts, generation, editing, sound, copy
Not my scopePublishing and ad buying: clients post on their own accounts
one video from each project
6-7months of monthly jewelry content, still running
5books animated for a US publisher
12k+likes on just six of the book posts
0photo or video shoots
Key frame = a still I compose by hand that the video must hit.Start → end = two key frames; AI animates the move between them.Hold = a pause where the camera stops so the text can be read.
Q1 · Why AI at all?

Because some content costs more to shoot than it will ever earn.

A Paris shoot every month for a jewelry feed. Hand-animated trailers for every new book. A crowd scene where two strangers mirror each other's steps. Each of these is real work for a production team. With AI and enough craft, it fits a normal content budget.

The rule I work byAI never decides the frame. Composition, details and transitions come from me; generation fills the motion in between.

What usually goes wrong
  • Plastic skin and perfect symmetry
  • Subject dead centre, studio light
  • Text and details that drift
  • Colour that jumps between clips
What I do instead
  • Rules that make frames imperfect on purpose
  • Key frames built by hand from real assets
  • Short clips joined frame to frame
  • Text, sound and holds added in the edit
1Q2 · Fine jewelry brand · social content

Parisian quiet luxury, every month, without a single photoshoot.

The whole brand lives in a slightly retro European world: old-money interiors, soft daylight, silk and skin. Shooting that for real every month means locations, models, styling and a photographer. I build it with AI instead, simply and at a fraction of the cost.

6-7 monthscontinuous work, still running
12 + 6posts and videos every month
10+stone categories with their own prompts
1 style"rich iPhone": shot on a phone, in an expensive life
Jewelry post 1
Jewelry post 2
Jewelry post 3
Jewelry post 4
Jewelry post 5
Jewelry post 6
Jewelry post 7
Rule 1

Never centred

A centred subject is the first tell of a generated image. The jewel sits off-centre with generous empty space.

Rule 2

Real, matte, imperfect

No "hyper-realistic". Skin is unretouched, hands are asymmetric, faces have character, clothes look worn.

Rule 3

Light from the side

Soft raking daylight shows texture and metal. Flat studio light is banned.

copy too: lowercase, old-money tone
2Q3 · US book publisher · trailers and reveals

Book trailers that read as hand-made art, not AI.

A major US publisher of romantasy and dark fantasy needed trailers, teasers, cover reveals and deluxe-edition showcases for each new release. The audience is book lovers who notice every detail of a cover. The videos had to feel like the book's own artwork come to life.

For marketersEvery frame starts from the publisher's real artwork. Nothing is invented by the model, so covers, edges and endpapers stay exact.

  1. Take the book's real art. Cover, stencilled edges, endpapers, map, character cards.
  2. Split it into key frames by hand. Each scene is composed as a still before any generation.
  3. Animate between frames. Start and end frames go into Seedance in 4-6 second clips, so colour holds and any weak clip can be redone alone.
  4. Finish in the edit. Titles, transitions, holds for reading, music and sound effects.

Teaser, jungle series. Eight key frames cut from the book's own illustrations: eyes in the dark, a hawk, a python, flowers. Seedance only moves between them.

Jungle teaser storyboard

Teaser, lilac series. Snakes, a sword, a wing, a dragon: each object placed by hand in the same frame so the cuts feel like one illustration coming alive.

Lilac teaser storyboard

Trailer, sea series. Titles sit on calm frames; the ship and the lightning arrive between them.

Sea trailer storyboard

Storyboards scroll sideways. Each frame here is a still I built before any animation.

Trailer
Deluxe edition showcase
Tagline trailer
Teaser

All book artwork, covers and titles belong to the publisher, the authors and their illustrators. I show this work only to describe my role in the project.

Q4 · Did readers buy it?

12,000+ likes on six posts, and the comments praise the art.

The videos go out as joint posts by the publisher and the authors. Here are six of them; it is only part of what I made for this client.

likes12,354across six posts, from 430 on a teaser to 3,220 on a trilogy follow-up.
reposts551on the two posts where shares are visible: readers send these to friends.
comments449on three posts with visible counts, mostly about the artwork.
PostLikesCommentsReposts
Teaser, second book of a trilogy3,220··
Cover reveal, new trilogy2,66394·
Deluxe edition announcement2,508173209
Book trailer1,893182342
Cover reveal, book in an animated frame1,640··
"Something big" teaser430··
"Oh my god… That ART""THE ART INSIDE AHH""Gorgeous""I can't handle this""Already ordered this and begging for an early copy"
Trailer post, 1,893 likes
trailer · 1,893 likes
Teaser post, 3,220 likes
teaser · 3,220 likes
Cover reveal post, 1,640 likes
cover reveal · 1,640 likes
the author: "such a pinch me moment!"
3Q5 · Shuggle · short videos for a new word

A moment everyone knows, and almost nobody can film.

My client invented a word, "shuggle": when two people walk toward each other, step to the same side, then the other, trying to pass. He wanted short videos to bring the word into everyday language. It is hard to stage with real people and one of the hardest things to generate: two characters, perfectly mirrored steps, a busy street.

Problem

Two bodies, one rhythm

AI handles one character well. Two that react to each other step by step fall apart.

Approach

Many iterations

Scenes rebuilt again and again, kept only where the dodge reads instantly.

Craft

Key frames by hand

Every beat of the dodge is fixed as a still first, so the viewer sees exactly what happens.

Seven key frames, drawn before any animation. Two people walk in, meet, both step the same way, stop face to face, then pass. Each beat is fixed as a still so the dodge reads in one look.

Shuggle storyboard
Bike path
Gala night
Red carpet

Also testedInstead of one prototype I suggested comparing formats: AI animation, edits of existing scenes and a real creator, so the client could see which one people react to.

Q6 · What would this look like for you?

Content that's hard to shoot, made to look crafted.

For brands with a strong look, publishers and anyone whose idea is too costly or impossible to film. Monthly content or one launch.

step 1Visual system

Your references turned into rules and a prompt library.

step 2Key frames

Built by hand from your real assets.

step 3Animation

Short clips, frame to frame, colour kept.

step 4Edit

Text, holds, music and sound effects.

step 5Delivery

Posts and videos ready for your feed.

Start a project · okad.agencystart here