Buzz Off: turning a two-year-old script into a Pixar-style short

A film in 20 shots, built shot by shot

One mood board to rule the film

Everything starts from a single generated board: title treatment, both characters, the set, the palette, and the opening quote — then the characters and room get cropped out and re-used as reference images for every single keyframe. Words drift; pictures don't.

The whole film on one canvas. Maya's turnaround, Buzz's two poses, the environment and five paint swatches — generated in one shot, then cropped into reference sheets.

A studio pipeline in four moves

Adapted from Episode 03's keyframe-pair workflow, rebuilt for character animation: the video model never invents a composition — it only animates between frames that were approved as cheap images first.

Keeping one room one room

AI models happily redecorate between shots. The fix is a three-layer continuity system — and it's the heart of this whole production.

• Layer 1 — a floor plan in words. Every prompt carries the same fixed geometry: window and sofa frame-left, desk and mirror behind right, one low table, exactly ONE mug in the whole room — and where the camera physically stands for this shot.
• Layer 2 — the master wide. The approved establishing shot is attached to every other generation as an image reference. Furniture positions survive in pictures far better than in prose.
• Layer 3 — chained frames. A shot's end frame is generated image-to-image from its approved start frame, so nothing drifts mid-shot.

The canonical set. Shot 02 doubles as the establishing shot of the film and the ground truth every other camera angle must obey.

Shot 01 — opening blur Shot 02 — focus reveal

The zero-cost shot. The film's opening frame isn't generated at all — it's the master wide put through an ffmpeg gaussian blur, so the cut into shot 02 plays as a focus-reveal. Free, and pixel-perfect continuity.

Three ways this went wrong first

Every case study here shows the misses. This episode had three recurring ones — each now a standing rule in the production bible.

Failure 01

Photorealism creep

Two shot types kept sliding into photography: frames with no character face, and extreme macros on a face — pores and all.

Every risky prompt now opens with "a cartoon animated world, NOT a photograph" — plus "NO pores" on face macros.

Failure 02

The scale-creeping mosquito

Every model wanted the mosquito to be a co-star: dragonfly-sized, in focus, grinning. It took three rounds of notes to shrink him.

New hard rule: true real-world scale — a speck the viewer has to look for, no wider than an eyebrow even in macro.

Failure 03

The fourth-wall trap

Over-the-shoulder angles kept "turning" Maya toward whatever wall the model put behind her — because the wall she faces is never seen on camera.

Over-the-shoulder shots are banned on this set; reverse angles became front three-quarters, and prompts now state her facing direction.

Hiding the star until shot 16

In the script, the mosquito's adorable design — the headband, the huge eyes — is only revealed when he lands in Maya's palm and unrolls a tiny yoga mat. So the prompts enforce it: before shot 16 he may only exist as a dark motion-blurred silhouette.

The design the audience hasn't earned yet. Buzz's reference sheet — reserved for the third act.

What the film shows instead. Shot 07: a speck-scale silhouette descending like a lunar lander.

Two clips, two tempos

Both animated with Kling 3.0 from a single approved keyframe each — one slow-burn macro, one full slapstick. The film's story clips will use locked-off cameras; these two deliberately break that rule for the showreel.

The keyframe behind the hero clip. Shot 12: tai chi meets cartoon frenzy, every prop still exactly where the master wide says it lives — and the mosquito a speck by the mirror.

Part II — the film itself