Building two Unity intros with AI and a lot of revisions

I wanted a few wooden letters to fall down, someone off screen to yank their strings, and my logo to come together. In the other intro, a sleep-deprived dad drinks coffee and gets hit in the back of the head by a stuffed rabbit. Neither sounded particularly dangerous as a project.

Then I found myself typing things like “he has no fucking neck” and “am I seeing this right, is he drinking from the mug handle?”

This is how the Keocko and After Bedtime Studios intros came together. Built in Unity, with Codex, my own source material, generated artwork, and quite a few requests to try again.

The creative direction and the nagging stayed with me

I didn't want a generic AI animation. Upload a logo, let something flashy happen around it, call it done. I wanted to see how far I could take my own idea: the homemade puppet theatre, the fumbling puppeteer, the dad with his coffee, and that tiny pause that makes the joke work.

I wanted to use Codex as a team. Get the artwork ready, split it into movable parts, assemble it in Unity, watch it, fix it. Depending on the round, I needed art preparation, animation, programming, or testing.

The creative direction came from me. I wanted the letters to stay crooked after the first tug. I wanted the rabbit and the dad's reaction. I also had to decide whether a movement was funny, whether the set had any atmosphere, and when to say: this is shit, let's try again.

The machine offered solutions and ideas too, and I could choose from them. Producing something quickly didn't mean it was accepted. After “it works” came the question of whether I actually liked watching it. The creativity comes from me, and so does the nagging. This arrangement needed a surprising amount of the latter.

Why Unity and why this fake 3D

The first ABS version was still 2D. Separate sprites, separate layers, a supposedly controllable stacking order. We couldn't get our Unity implementation to handle it properly: the arm bled into the shirt, outlines disappeared or popped up in the wrong place. Even after fixing the z-index, another frame would turn up with broken overlaps.

That's where the idea of using planes in 3D space came from. Keep the original drawing, but give the body, arm, and head actual positions in depth. I liked the paper-cutout look and cardboard edges too. That became the “fake 3D”: drawings arranged in space, with actual 3D wooden letters and physics in Keocko. We didn't need a fully modelled object you could walk around for every element.

There was a more personal reason I leaned towards Unity. I've never used a timeline-based video editor. I've learned so much shit already; somehow I never had the patience for that. But I could explain what I wanted to see: when he should lift the mug, how long to wait before the second tug, which movement a sound should follow. I wanted to find out whether prompts could get me to a working Unity animation.

After this, it certainly doesn't look impossible. For each fix, I described what looked wrong, asked for another video, and watched it back. Codex did the technical work. I still had to think through the timing and movement; learning to operate the editor was the part I could get around.

We're also working on several Unity games, and these intros will be integrated natively. The game will run the scene, objects, animation controller, and sounds, rather than play a pre-rendered MP4. We can configure what happens next and how skipping works, adapt the aspect ratio, and keep editing the movements later.

The video is for demonstrating the result and watching revisions back. For the game, I needed the importable Unity assets.

From 2D struggles to a cardboard character

The basic After Bedtime Studios scene already worked as a still image. A scruffy, pot-bellied, barefoot dad with coffee and a keyboard. The studio name beside him, then a stuffed rabbit flies in. The rabbit throwing is real: my son Olivér keeps chucking his bedtime rabbit at me. The reference material came from home.

The original After Bedtime Studios visual reference with the dad, keyboard, and studio name

The starting reference. This had to become an animatable character.

In the early 2D rounds, the elbow bent in the wrong place, the hand and mug didn't fit together, and the shirt outline broke. We combined parts that belonged together into single drawings, but even an agreed merge could disappear in the next revision.

After moving into 3D space, the character's structure finally became easier to understand. The body, head, and arms sat on separate depth planes; their positions determined what covered what. We kept the drawing, but could now treat it as a set we could take apart and edit.

The visual improvements grew from that. The lettering gained thickness and a restrained shadow, while the character got a consistent ink outline, a paper border, and a darker cardboard edge. It came much closer to the cut-out, handmade world I wanted.

I didn't like the 3D wobble and surface shading we tried on the character, so we dropped those. The depth on the letters worked, though. That gave us the final direction: keep the character drawn, tie it together with the cardboard edge, and give the lettering some weight. We preserved the original movements and the timing of the joke throughout.

A frame from the current landscape After Bedtime Studios intro

The later version with depth: a more consistent character, cardboard edges, and thicker lettering, while keeping the personality of the original drawing.

The rabbit's ears and a loud HEY

“Make the rabbit's ears flap” first produced movement where the ears still pointed up. Even on the floor. I had to spell out that they should trail behind during flight, then be pulled down by gravity as the rabbit fell. The motion stayed manually timed so the hit would land at the same point on every playback.

The sound took some searching too. My verdict on an early synthetic “boom-boom-boom, slurp” attempt was simply “the sound is HORRIBLE”. I sent my own sample, then generated better sounds with ElevenLabs. Its free allowance was more than enough for this job. The three lines of text needed different “pik-pak-puk” sounds, followed by a HEY after the hit, which we shifted back and forth quite a few times.

Keocko's trumpet music, on the other hand, gave me a hit straight away. ElevenLabs initially returned four alternatives: three were completely irrelevant, and the fourth was pretty much exactly what I wanted to hear. When it went under the fumbling puppet theatre, I laughed until I cried. Three misses and one bullseye from the same request. Luckily, one was enough.

The built player had another surprise: after Unity's own logo, I was already watching the rabbit fly. The animation clock had been running underneath the splash screen. We needed a separate check that the story only started afterwards, and that skipping stopped the sound properly too.

The wooden letters fell too neatly

Keocko started with my own logo in seven separate pieces. Preserving the original shape mattered: the final geometry comes from the CDR file, and PRESENTS uses the supplied Readex Pro font.

We only built the fall at first. Even that needed work. The letters lay too neatly on the floor, and the fall felt like lunar gravity. I wanted more chaos: pieces tipping over, landing on each other, wood that felt heavy.

Later, we noticed the piece that stayed out of place barely moved on the first tug. The joke was supposed to be that it tried to get into position and failed. We left more letters crooked and added half a second before the second tug, enough time to notice the mistake.

The letters rising didn't always look like a consequence of the strings being pulled, either. Eventually I described it like this:

• they fall, the string is slack (this is good, actually perfect)
• he yanks it, it goes taut, and that makes every letter jump up
• the string goes a little slack again before the next tug
• it goes taut again, and everything falls into place

The fade screwed us over in both projects

The separate ABS body parts faded differently, pulling the character apart as it darkened. In Keocko, the background disappeared behind the letters too early.

A screenshot of Keocko's transition problem, with the background disappearing behind the letters too early

The question that came with this screenshot: “why did the background disappear behind my letter before the fade?”

In both cases, the completed picture needed to fade as a whole. In Keocko, the physical theatre, strings, and wooden letters form one unit; the clean logo and PRESENTS are separate overlays. Then we still had to adjust the overlap, because the logo briefly seemed to fade back out and pulse.

The portrait version nearly became a new project

“Let's make a portrait mobile version” brought another set of rounds. In ABS, the character went above the lettering. Keocko's set was harder to fit into the new aspect ratio. The suspension changed, the backcloth lost its separate layer, and the lights became a lovely regular string of bulbs.

I wanted a cable someone had hung up carelessly. Lower on the right, shifted a little off centre. Producing that carefully considered lack of care took quite a few messages.

An intermediate portrait Keocko concept with curtains and an empty stage opening

An intermediate portrait concept. The final mobile version lost the side curtains but kept the rear backcloth.

At one point, we were heading towards a complete 3D rebuild because the floor and sense of depth looked awful. Eventually we returned to the layered approach. The camera looks straight across at stage level, with a wooden frame and boards in front and a separate backcloth behind. We kept the chosen look and the working structure.

For mobile, we adjusted the strings, fall, and sound to the higher starting point, while keeping the tug rhythm shared with landscape. A new aspect ratio didn't mean a new puppeteer.

Here are both intros back to back in portrait, at 9:16. This has its own layout rather than being a crop of the landscape video: the set and lettering fit the narrower frame, while both jokes keep the same timing.

The team built it but I still had to sign it off

The long prompts set out the story. Then I had to sit down, watch it, and explain what wasn't right. A cropped screenshot of a problem often did more than another paragraph.

The most annoying moments were when an earlier decision got lost: the merged hand fell apart again, or the mobile version reinvented parts that already worked. Then I had to steer the work back to the original idea.

The delivery includes native Unity scenes and assets, checked by importing them into a clean project and testing full playback and skipping. The web demonstration runs Unity splash, Keocko, then After Bedtime Studios. The two intros can also be integrated into games separately.

Keocko runs for 8.25 seconds; ABS for six. When I watch them now, I see my own wooden letters and the rabbit. I still check the elbow, though.