The short version
- Every run followed the script the same way: seven scenes, all in 3D, with 51–59 sound cues and 132–159 sounds placed on the timeline.
- Cost varied a lot. Sonnet 5.5 on medium effort made a first draft in 15 minutes for $3.05. Opus 5.5 on max took 172 minutes and cost $44.30.
- Higher effort meant more checking. The xhigh and max runs checked every scene change and had no hard cuts. The medium and high runs had 11 hard cuts between them.
- The films look most different in depth, lighting, layout and motion.
- What we're changing: Animro will run these checks on every draft, at every effort level.
The brief
The brief is a full script for a 60-second Animro launch film called “Type it. Watch it move.” A lime text cursor leaves the Animro chat box as a line of light. It draws a storyboard and builds a 3D product shot for a made-up app called Aster 2.0. Then it turns into a case study, a social ad and a short lesson, pauses for a feedback note, renders, and finally draws the Animro logo.
We made it hard on purpose. Every scene is in 3D, and the camera has to move through the whole film in one continuous shot, with no hard cuts. There is no music or voiceover. The soundtrack is only sound effects: about 50 cues and around 140 sounds, timed to the picture. The script is 7,885 characters long, just under the 8,000-character limit of the Animro chat.
Read the full brief
This is the exact text every run got, along with two logo files. Aster 2.0 and its numbers are made up.
ANIMRO LAUNCH FILM: "Type it. Watch it move." 60 s · 16:9 · no voiceover, no music: synced sound effects are the whole soundtrack. This is a finished script. Keep its scenes, timings, on-screen text and sound cues; put your craft into how it looks, moves and sounds. Animro (animro.com) is an AI motion studio: brief it in chat, approve the storyboard and styleframes, pin notes on any frame, export 1080p MP4 in 16:9, 9:16 and 1:1. Make it the most ambitious motion design you can render. Every scene is 3D (three.js): this concept needs far more than one hero 3D moment. Cinematic camera moves, depth of field, motion blur, a wide glow on lime, light sweeps across glass, particles, kinetic type. One continuous camera journey with no hard cuts: each scene hands over through a camera move, a morph or a match cut. BRAND Charcoal #101110 is the world; lime #D0ED80 the line, buttons and one word per headline; paper #F3F3EB type; ink #182016 type on lime. Sample films: launch lavender #C8BDF2 / plum #302447; case study sage #D7DED3 / forest #263E34; social coral #F28E65 / brown #442217 / sun #FFB54A; education lime / ink / paper. Type: Inter 600, tight tracking; labels in JetBrains Mono caps. Logo (attached SVGs): one continuous lime stroke drawing an "A", and the wordmark "animro" in paper with a lime full stop. The mark is stroke-only: in 3D, build it as a tube along its path. Look: premium, matte, precise. Frosted glass, clay, soft studio light, light grain. No glitch, lens flares, rainbow gradients, chrome, robots, stock "AI" imagery, full-frame flashes or strobes. THE HERO The lime caret leaves the chat box as a glowing 3D ribbon of light that draws every scene, carries the camera and finally draws the logo. 1 · BRIEF (0:00-0:05) Macro shot of the Animro chat box: a frosted-glass slab on charcoal, a rim light gliding along its edge, shallow focus. The lime caret blinks twice. 0:00.8-0:03.2 "A launch film for Aster 2.0. Lots of 3D." types in, each letter popping up in 3D. 0:03.4 A file chip pops in: aster-logo.svg. 0:04.0 The cursor presses the lime send button. The text is sucked into it, the box tilts away, and the caret shoots upward as the line. SFX: room tone from 0:00 · a key click on every letter · chip pop 0:03.4 · send thock 0:04.0 · suction whoosh 0:04.1 · line zip 0:04.5 2 · STORYBOARD (0:05-0:12) The camera cranes up after the line into dark space over a faint lime grid floor fading into fog. From 0:05.25 the line draws six storyboard cards in a curved arc, one stroke every 0.25 s; each snaps solid as a pencil sketch when its stroke closes: 01 Hook, 02 The app, 03 Features, 04 Hero shot, 05 Launch day, 06 End card. 0:07.5 Slow dolly along the arc, focus racking card to card. Super: Approve the storyboard. 0:08.6 The cursor clicks "Create styleframes". A light sweep crosses the cards, turning each sketch into a full-colour still. 0:10.6 The cursor clicks "Animate". The camera pushes through card 04 into its world. SFX: quiet zip at each stroke start · card snaps 0:05.50 to 0:06.75 every 0.25 s · long camera whoosh 0:07.5 · click 0:08.6 · glass shimmer sweep 0:08.8 · click + bright chime 0:10.6 · push-through whoosh 0:11.0 into a sub thump 0:12.0 3 · DRAFT (0:12-0:22) Sample: Aster 2.0, a fictional planning app, in lavender and plum. Top-left label: DRAFT · v1 0:12.0 Lime wireframe: a phone, three UI cards and a six-petal asterisk logo, with Bezier construction handles. 0:12.6 A scan plane sweeps down, rendering the materials: lavender-white clay phone, glass screen with the Aster app, extruded plum logo. 0:14.0 Hero turn: the phone rotates 40 degrees on a spring as the camera arcs; a specular sweep crosses the glass. Bokeh, dust in the light. 0:16.0 Three UI layers peel off the screen towards the camera at three depths: "Launch v2.0 ✓", a 100% ring, "Shipped". 0:17.5 The extruded asterisk swings out from behind the phone, spinning with motion blur, and locks beside it at 0:18.0. 0:18.3 Title slams in: Aster 2.0 / Available today. Slow drift to 0:22. SFX: power-up hum 0:12.0 · scan sweep ending in a lock click 0:12.6-0:13.8 · heavy rotation whoosh 0:14.0 · glass glint 0:14.6 · soft lift pops 0:16.0 / 0:16.3 / 0:16.6 · spin whirr 0:17.5 into a deep metallic lock 0:18.0 · punchy hit on the title 0:18.3 4 · ANY STORY (0:22-0:35) 0:22.0 Case study, sage and forest: the line shoots out of the asterisk and draws a rising curve; thirteen glass bars rise from a reflective floor under a low crane move. A counter rolls to +212% (online orders · SAMPLE DATA). 0:26.5 Social, coral, brown and sun: the tallest bar morphs into a 9:16 frame. "Stop the scroll." slams in word by word. The ad splits into three glass panes, 16:9, 9:16 and 1:1, fanning out in depth, each re-laid out. 0:31.0 Education, lime and ink: the 1:1 pane flips into a lesson, "Why do we have seasons?". The camera tilts from top-down to 20 degrees over a 3D Earth with a tilted axis orbiting a paper sun; the day/night line sweeps. Top-left label: CASE STUDY → SOCIAL CAMPAIGN → EDUCATION. SFX: line zip 0:22.0 · 13 glassy ticks, one per bar, 0:22.6-0:24.2 · counter rattle 0:24.4-0:25.6 · punchy hit on +212% 0:25.6 · morph whoosh 0:26.5 · hits on the three words 0:27.0 / 0:27.4 / 0:27.8 · glass-pane slides 0:29.0 / 0:29.3 / 0:29.6 · flip 0:31.0 · airy shimmer 0:31.5 · soft orbit whoosh 0:33.0 5 · REFINE (0:35-0:43) 0:35.0 Time freezes mid-motion; the frame desaturates slightly and the Animro player fades in (scrubber, frame counter). 0:35.8 A lime pin drops on the lesson title, a pen circles it, and a note types "Bigger title." 0:38.0 Send. A version chip flips v1 to v2, time resumes, and the title springs larger. 0:40.0 Pull back: v1, v2 and v3 hang as glass layers in depth. Super: Change anything. Just ask. SFX: tape-stop 0:35.0, then near silence · pin tick 0:35.8 · marker squeak 0:36.2 · a key click per letter 0:36.8-0:37.6 · send thock 0:38.0 · flip 0:38.2 · tape-start 0:38.3 · elastic spring 0:38.5 · long whoosh 0:40.0 · glass clinks 0:40.2 / 0:40.5 / 0:40.8 6 · FINAL (0:43-0:50) 0:43.0 The layers align into one frame; the cursor presses "Render final". A lime progress ring fills around it, accelerating. 0:45.5 Done: the ring releases a lime glow pulse, the stamp 1080p MP4 lands, and three format cards fan out: 16:9, 9:16, 1:1. Super: Export in every format. 0:48.0 The camera rockets backwards through a tunnel of this film's own frames, motion-blurred, brightness steady. SFX: send thock 0:43.0 · clock ticks accelerating 0:43.2-0:45.5 · bright ding + punchy hit 0:45.5 · card snaps 0:45.9 / 0:46.1 / 0:46.3 · huge rushing doppler whoosh 0:47.8-0:50.0 7 · ANIMRO (0:50-1:00) 0:50.0 The tunnel collapses into one lime point. 0:50.5 The attached animated lockup, recreated in code at its own timing: the stroke draws with its small rotate-and-settle as a lime 3D tube that settles into the exact flat logo; the wordmark rises from 0:51.2; all settled by 0:52.3. 0:53.5 Tagline: Turn your brief into an animated video. 0:55.5 CTA: Start creating at animro.com 0:57.0 Total stillness to the end. SFX: near silence 0:50.0-0:50.4 · pen-stroke zip rising with the draw 0:50.6-0:52.0 · sonic logo as the stroke completes at 0:52.0: a warm deep bloom with a long glass-shimmer tail · tiny pop on the lime full stop 0:52.2 · soft click on the CTA 0:55.5 · room tone out by 0:59.5 SOUND RULES Dense by design: about 50 cues and some 140 placed sounds, far more than usual. No music, voice, melodies or drum loops. Premium tactile palette: crisp clicks, glassy chimes, airy whooshes, deep clean impacts; no cartoon boings, laser zaps, trailer braams or glitch noise. Land each transient on its frame (within 1 frame), not the file start. One generated sound per kind, reused for repeats. Loudest: Aster title, +212%, render done, sonic logo; whooshes medium; UI ticks quiet. Keep the silence before the logo.
How we ran the test
- We tested two models, Claude Opus 5.5 and Claude Sonnet 5.5, each at four effort levels: medium, high, xhigh and max. Higher effort gives the model more room to think and work on each step.
- Each run started in a new Animro project with the same brief and the same two logo files. Sound effects were on. Music and voiceover were off.
- A test script approved the storyboard and the styleframes as soon as they were ready. Nobody gave any other feedback. Each run ended at its first draft.
- We removed the usual time limit, so no run was stopped early. Sonnet 5.5 on max spent 138 minutes on the animation step alone.
- Animro shows drafts as quick 960×540 previews. For this post we rendered every draft again at 1080p and 60 fps on a cloud GPU, using Animro's final-quality setting, which adds motion blur.
Watch all eight films
Five films use the final-quality render. Opus 5.5 on medium, high and max couldn't finish it: their scenes used more GPU memory with every frame until the GPU ran out. Those three are rendered at 1080p without motion blur.
Claude Opus 5.5
1080p, no motion blur ·
1080p, no motion blur ·
Final quality, with motion blur ·
1080p, no motion blur ·
Claude Sonnet 5.5
Final quality, with motion blur ·
Final quality, with motion blur ·
Final quality, with motion blur ·
Final quality, with motion blur ·
Cost and time
| Model · effort | Claude cost | Time | Model calls | Output tokens | Hard cuts |
|---|---|---|---|---|---|
| Opus 5.5 · medium | $19.16 | 77 min | 154 | 246k | 3 of 6 |
| Opus 5.5 · high | $17.92 | 64 min | 128 | 231k | 2 of 6 |
| Opus 5.5 · xhigh | $29.18 | 102 min | 163 | 374k | 0 of 6 |
| Opus 5.5 · max | $44.30 | 172 min | 173 | 571k | 0 of 6 |
| Sonnet 5.5 · medium | $3.05 | 15 min | 42 | 84k | 1 of 6 |
| Sonnet 5.5 · high | $7.35 | 32 min | 81 | 168k | 5 of 6 |
| Sonnet 5.5 · xhigh | $29.01 | 126 min | 178 | 478k | 0 of 6 |
| Sonnet 5.5 · max | $41.54 | 240 min | 282 | 801k | 0 of 6 |
Cost is what the Claude tokens for each run would cost at list price, from the brief to the first draft. Time is how long the agent worked over the same steps. Higher effort mostly meant the model wrote more. Time went up with output tokens, and Sonnet 5.5 on max wrote almost ten times as many tokens as on medium.
Effort didn't change the plan. Every run kept the script's seven scenes, made all of them in 3D and placed well over a hundred sounds. We ran each setup once, so some differences are down to chance. For example, Opus 5.5 cost more on medium than on high.
One moment, eight versions
The plan was the same in every run, but the films look very different. Each image below shows the same moment from all eight films. Opus 5.5 is on the top row and Sonnet 5.5 on the bottom, from medium effort on the left to max on the right. Click an image to see it larger.
At 6.5 seconds, the lime line is drawing six storyboard cards in a curve. In some films the line is thick and glowing. In others you can barely see it, and the cards stand out instead.
At 29.8 seconds the script says: “The ad splits into three glass panes, 16:9, 9:16 and 1:1, fanning out in depth.” Some runs spread the panes over a shiny floor. One hadn't filled the third pane yet at this point, and another placed them so they overlap.
At 32.5 seconds the camera tilts over a 3D Earth that circles a sun, for a lesson that asks “Why do we have seasons?” One run filled the whole frame with lime. The others kept the scene dark, with a glowing sun in the middle.
What higher effort changed
The brief asked for one continuous camera move through the whole film. Each scene should lead into the next with a camera move, a morph or a match cut. At 12 seconds, the camera should push through a storyboard card into the product shot. Opus 5.5 on medium jumps straight from frame 719 to a different shot at frame 720. On max, the camera move is smooth.
The xhigh and max films have 24 scene changes between them and 0 hard cuts. Those runs checked every scene change. The medium and high films have 11 hard cuts, and two of those runs said there were “no hard cuts” after looking only at still frames. The model and the tools were the same. The extra effort went into checking the work, and the idea for the film stayed the same.
What we're changing in Animro
- 01
Automatic checks
Animro will check every scene change and every scripted moment on every draft, at every effort level.
- 02
Better sound timing
Each run wrote its own code to time more than a hundred sounds. Animro will measure each sound once and place it on the right frame.
- 03
Built-in 3D tools
Depth of field, glow, glass and 3D text will be built in, so the agent can spend its time on the film itself.
- 04
Early test renders
Three drafts looked fine as previews but couldn't render with full motion blur. A short test render at final quality will catch this before a draft is delivered.
Notes on the method
- We ran each setup once, all on October 3, 2026. Small differences may be down to chance.
- Costs are Claude list prices for the tokens used. They are not Animro prices. They don't include sound generation (about $0.13 per run) or the 1080p renders.
- One run lost 27 minutes to a network outage. It picked up where it left off, and we removed those minutes from its time.
- A hard cut is a sudden change of shot from one frame to the next, without a camera move, morph or match cut.
- Aster 2.0, its app screens and the numbers in the case study are made up.
- The films end with “Start creating at animro.com” because the brief was written for launch. Animro is now open, and you can start for free.