Tools / Showreel Bench · Round 1 · Oct 2026

Can an LLM be a motion designer?

I gave Opus 5.5 and Fable 5.1 the same narrated scripts and asked each to write the animation code, scene by scene, in one shot. No peeking at its own renders. Here are all 24 videos.

videos, 2 models × 3 efforts × 4 scripts

120

scenes of generated animation code

scenes that failed the sandbox check

24/24

runs added sound design unprompted

TL;DR

Opus 5.5 came out ahead. As good or better than Fable at every effort level, at about 2.5× lower list price per token.
Max effort is the only clear quality jump. Medium and xhigh came out about even.
Medium is the value pick: about 85% of max's quality for 1/8 of the time and tokens.
Fable had the best single ideas on abstract scripts, but stumbled on concrete how-tos.

Same script, two models

Both at max effort, with sound. Names stay hidden until you pick one.

Video A

Video B

Scores

1–5 per video, averaged over the four scripts, plus time, output tokens and list-price cost per video. Round 1 was scored by an AI judge (Claude Opus 5.5) from stills, so read it as a first pass. Reader votes will replace it.

Model · effort	Avg score	Min / video	Output tokens	List $
Opus 5.5 · max	4.25	32	205k	$6.25
Fable 5.1 · max	3.90	24	106k	$7.97
Opus 5.5 · medium	3.75	4	24k	$0.83
Opus 5.5 · xhigh	3.75	17	104k	$3.19
Fable 5.1 · xhigh	3.75	15	70k	$5.29
Fable 5.1 · medium	3.25	5	24k	$2.03

What stood out

Opus at max kept one consistent world per video: the same kettle and mug, a stopwatch that becomes a play button, then a checkmark.
Fable's best concepts were the most original: a tiled photo for memory, start-early vs start-late curves for compounding.
Fable struggled with concrete objects: grey coffee, odd proportions, a mug that changes shape.
Every run added music or sound effects when the brief allowed it.