Claude motion graphics prompts that work with Opus 5.5

Why one-line prompts disappoint

A short prompt like "make an amazing 15 second motion graphics video" is a fine benchmark of what a model can do unguided, and the famous one-shot showreels on X are exactly that. But for a video you want to post, a one-liner leaves every decision to defaults: a generic gradient, centred text, fades between everything, no idea where the platform buttons sit. One creator put it bluntly on X: by default Claude's editing looks poor, and the accounts that do well are the ones directing it.

A good motion prompt is less a request than a production brief. The parts are always the same, which is why a template works.

The seven parts of a motion graphics prompt

• Format: frame size, frame rate, duration and platform. 1080 by 1920 at 30 fps for Shorts, Reels and TikTok.
• The contract: one HTML page, every frame a pure function of t, a seek(t) function, no Date.now or unseeded random, no network. Without this, Claude may write an animation that only plays live and cannot be captured.
• Look: background, a palette of three to five hex colours with a role for each, one or two typefaces, texture. Say what the accent colour is for.
• Motion: the easing, how things enter and leave (cuts, wipes, draws on), how long a typical move takes, whether the camera moves.
• Beats: the script or the story as a list of moments with times. The hook in the first two seconds. A new visual every two to three seconds.
• Prohibitions: what to never do. Banned moves are as useful as wanted ones: no lens flares, no centred paragraphs, no text in the bottom fifth.
• Review: ask Claude to render stills at four timecodes and critique them against the brief before the full render.

Keeping look and motion as separate parts is the habit that pays off most. It lets you say "same look, faster cuts" or "same motion, warmer palette" without the rest of the video shifting.

Template 1: kinetic type

Big words that land on the beat of the voiceover. Cheapest to render and the base layer for hooks and quotes.

Make a 30 second vertical video as one HTML page.

FORMAT
1080x1920, 30 fps, 30 s. Expose window.__film = { width, height, fps,
duration, ready, seek(t) }. Every frame is a pure function of t. No
Date.now, no unseeded Math.random, no network. Fonts are local files.

LOOK
Background #0d0d0f. Words in #f4f1ea. One accent, #ff3b2f, used only on
the single most important word of each beat. One heavy grotesque
typeface, weights 800 and 400. Fine film grain at 4% opacity.

MOTION
Words arrive on the beat with a 0.25 s ease-out from 40 px below. Cuts,
not crossfades, between beats. Numbers count up over 0.6 s. No camera
moves except a 2% push-in across each beat.

BEATS (times come from voice.json word timings)
0.0  Hook: [THE SURPRISING NUMBER]
2.0  [WHAT IT MEANS]
...  one beat per sentence of the script below
END  Payoff line, held for 1.5 s

SCRIPT
[paste the script]

NEVER
Centred paragraphs. More than 6 words on screen. Text outside
x 90..910, y 260..1420. Gradients, glows, lens flares.

REVIEW
Before the full render, capture stills at 0.5 s, 8 s, 16 s and 28 s,
show them as one contact sheet, and list anything that breaks the
brief. Fix those, then stop for my approval.

Template 2: ink stickman explainer

A single stick figure drawn on as the narrator talks. Reads as hand made, which suits life lessons and how-things-work topics.

Make a 35 second vertical explainer as one HTML page, same FORMAT
rules as before (1080x1920, 30 fps, window.__film, pure function of t).

LOOK
Warm paper #f5f1e8 with a faint fibre texture. Ink #1b1b1b, one red
#d6402e for emphasis only. Lines have slight pressure variation, drawn
with a seeded wobble so they look hand made but identical every run.
Hand-lettered labels, never more than 4 words.

MOTION
Every line draws on along its path (stroke dash offset driven by t),
0.4 to 0.8 s per stroke. The stick figure has three poses; it moves
between them with a 0.3 s ease-in-out, never teleports. One idea per
scene; scenes change with a page-turn wipe.

STORY
Hook as a question in the first 2 s. Then [3 to 5 steps of the
explanation]. End on one line that answers the hook.

NEVER
More than one figure. Clip art. Colour fills larger than a label.
Anything drawn outside the safe area.

Template 3: diagram explainer

Graphs and shapes that draw themselves on a dark grid. Best when the topic has a real number or a curve in it.

Make a 40 second vertical explainer as one HTML page, same FORMAT
rules as before.

LOOK
Background #0e1420 with a 60 px grid at 6% white. Primary line
#58c4dd, highlight #ffd166. Labels in a clean sans, numbers in
tabular figures.

MOTION
Axes draw first (0.5 s), then the data. Curves draw on left to right,
driven by t. When a value matters, a dot rides the curve and a label
counts to the value. Shapes morph into the next idea instead of
cutting.

DATA
[paste the real numbers and their source]. Plot them exactly; do not
smooth or invent points. Put the source in small type at the end.

NEVER
3D effects. Pie charts. Any number not in DATA.

The DATA block is the most important line in this template. A diagram explainer that plots an invented curve is worse than no video, and Claude will happily invent one if you do not give it the numbers.

Follow-up prompts that fix a render

After the first stills, these are the follow-ups we use most. Each names one problem and one fix, which works far better than "make it better".

• "The hook is not readable at phone size. Make the first beat at least 140 px and hold it for 0.4 s longer."
• "Beats 4 to 7 all use the same entrance. Vary them: one wipe, one draw-on, one scale from 0.9."
• "The accent colour appears on three words in beat 3. Keep it on the number only."
• "Captions sit under the platform buttons at 26 s. Move everything inside y 260 to 1420."
• "Nothing moves between 12 s and 15 s. Add a slow push-in or draw the next element earlier."
• "Render the four stills again and score each against the brief out of 10, with one sentence on what would make it a 10."

Craft rules worth putting in every prompt

• One hero element per frame. If two things move, one of them should be quiet.
• A new visual every 1.5 to 3 seconds on a Short; open a new question about every 10 seconds on anything longer.
• Write numbers twice in the script: "$440 million" for the caption, "four hundred and forty million dollars" for the voice, so text to speech reads them correctly.
• Lock the character before the story. If a figure appears, describe it once, precisely, and refer back to that description.
• Ban the phrases that make narration sound machine written: "in today's fast-paced world", "let's dive in", "it's important to note".
• Ask Claude to look at its own output. A contact sheet of stills is the cheapest review you can buy.

Public prompts worth reading

Several creators have published long, detailed prompts for specific looks: a flat space explainer with a full hex palette and mascot, a paper collage documentary with separate image and video prompts, a stickman survival series with its hook and HUD rules. They are worth reading to see how much detail a strong prompt carries. They are their authors' work, so read them at the source rather than pasting them into a product, and name your own versions generically rather than after the studio whose style inspired them.

Community indexes such as awesome-opus-5-5-video-prompts collect hundreds of examples with previews and the inputs each one needs.