← All skills

Give It a Clock

Turns five answers into a ready-to-paste Seedance prompt for a 5-second logo sting — three timestamped segments, a locked camera, and a last second that belongs to the mark alone.

identitymotionvideoseedanceminutes
What it hands backFive answers in · one Seedance prompt out
Vibary
Harbour Row

Two runs of the same template. This run failed the hold law, and is shipped because the measurement is the point. Frame-to-frame change never reaches the noise floor: it is still 0.57 at the cut, around three times the 0.2 floor the other run settles to, and the brass droplets visibly change position and count between 4.0s and 4.9s. What failed was the stillness, not the resolve — the wordmark itself is correct and fully formed by 4.0s. This is the run that produced the clearing rule: it is not enough to say the logo holds, the third segment has to say that the remaining material has left the frame, and `--camerafixed true` has to be on.

Measured rather than eyeballed, because the whole law is about timing. Frame-to-frame change peaks at 12.4 during the gather, falls to 0.56 at 4.0s and to the codec noise floor of about 0.2 by 4.2s, holding there to the cut — so the hold is real, and it starts about two tenths of a second later than the field asked for. VIBARY is correctly spelled and fully closed, and nothing crosses it once it has resolved. The eight reference templates this was washed from all specified “about 5 seconds” and then, between them, carried not one timestamp — while Seedance has read `0s-3s` segments natively all along.

Generated for this library

Input
  Mark      a wordmark reading exactly VIBARY, geometric
            sans, all caps
  Material  clear glass, slow-drifting blobs that refract
            and merge
  Beats     0.0-1.8 arrival · 1.8-4.0 gathering, readable by
            about 3.2s · 4.0-5.0 the hold
  Ground    plain near-black, one soft key upper left
  Palette   cool white and pale cyan against near-black

Returned  the prompt below, ready to paste

  A glass wordmark reading "VIBARY" forms out of drifting
  liquid blobs on a plain near-black ground, resolves, and
  holds still.

  0.0s-1.8s: Clear glass blobs drift in from the left and
  right edges and float slowly. Nothing resembles letters
  yet.
  1.8s-4.0s: The blobs pull toward the centre and settle
  onto the letterforms of "VIBARY". The word becomes fully
  readable by about 3.2s and only ever gets clearer.
  4.0s-5.0s: The wordmark "VIBARY" is complete and
  perfectly still. All remaining glass has left the frame.
  Nothing crosses the letters. The camera does not move.

  Plain near-black background, no environment. One soft key
  light from the upper left, constant throughout. Cool white
  and pale cyan glass. The text reads exactly "VIBARY" —
  six Latin letters, correctly spelled, no Chinese characters. No
  distortion of the
  letterforms, no shattering, no debris, no extra text.

  --ratio 16:9 --resolution 1080p --duration 5 --camerafixed true
SKILL.mdthe whole thing, as you would paste it© vibary.ai

You fill in 5 fields — the braced lines at the top. Everything below them is fixed.

  1. 1The mark, and what it says
  2. 2The material it arrives out of
  3. 3The three beats, with seconds
  4. 4The ground and the light
  5. 5The palette

Give It a Clock

Take the five fields below and write a Seedance prompt from them. The prompt you produce is the deliverable — not this document, and not a description of what to shoot. Follow every line.

First — do you have the fields?

If any field below still has braces around it, stop and ask. Put all the questions in one short message, and show the example under each one so they can see the shape of a useful answer. Wait for the replies, then write the prompt. Never fill a brace yourself: a value you invented is a value nobody asked for, and it will look exactly as considered as the ones they did give you.

Ask in whatever language they are writing to you in. Translate the questions and the examples; do not make someone read English to tell you what they want.

The examples are only examples. If someone answers a question, use their answer; never fall back to the example because you prefer it.

If the fields are already filled in, ignore this and carry on.

Fields — replace these five lines only

  • The mark, and what it says: {a wordmark or symbol, and its exact text}
  • The material it arrives out of: {what the mark is made from, and how it moves}
  • The three beats, with seconds: {0.0–x, x–y, y–5.0, each named}
  • The ground and the light: {a background, and one light direction}
  • The palette: {two or three colours}

What you produce

One Seedance prompt, in this order and nothing else. Write it in the language the fields were given in.

``` <one sentence: the mark, the material, the ground, and that it resolves and holds>

0.0s-<a>s: <beat one — the material arrives, nothing resembles the mark yet> <a>s-<b>s: <beat two — the material finds the mark's structure; say when it becomes readable> <b>s-5.0s: <beat three — the mark complete, static, nothing crossing it, camera stopped>

<closing constraints: ground, light direction, palette, the exact spelling and script of the mark, legibility, and what must not happen>

--ratio 16:9 --resolution 1080p --duration 5 --camerafixed true ```

The timestamps are the point. Seedance reads 0s-3s segments natively, so the beats go in as segments, not as prose about timing. Three segments, no more, and they tile the whole five seconds with no gap and no overlap.

Punctuate them the way the prompt's own language does — a plain colon after the range in Latin scripts, a full-width one if you are writing in Chinese or Japanese. The model reads the timestamps, not the punctuation; a stray full-width colon in a Spanish prompt is just litter.

Spell the mark inside the prompt, in quotes, exactly as the field gives it — and say it again in the closing constraints. Video models get letters wrong, and saying it twice is the cheapest insurance there is.

The mark's script is the field's, not the prompt's. A prompt written in Chinese for a Latin wordmark must say so outright — *the text is Latin letters only, no Chinese characters* — because a model reading a Chinese prompt will otherwise render Chinese in the picture. The same holds in reverse: an English prompt for a Chinese mark states that the characters are Chinese. Name the script every time; the language you are writing in has no vote on it.

### The parameters

  • --ratio 16:9 unless the fields ask otherwise. 9:16, 1:1, 4:3, 3:4 and 21:9 are the other options.
  • --resolution 1080p. Drop to 720p only if the model refuses.
  • --duration 5. The only other value is 10, and a sting is five.
  • --camerafixed true by default. The one thing this format must deliver is a still, legible mark at the end, and a locked camera removes the commonest way that fails. Set it to false only if the fields ask for a move — and then the move belongs in beats one and two, never in the hold.

The clock law

This is a five-second film whose entire job is to leave a mark on screen, legible, held. Templates in this genre describe the journey in detail — particles gather, glass fuses, metal pours, paper folds — and never say when anything happens. Eight reference templates for this format were checked: every one specified "about 5 seconds" and then used only staging words — first, gradually, finally. Not one carried a single timestamp.

Then, at the end, each listed among its don'ts: *don't let the refraction be so strong the logo can't be read*, *don't let particles block the subject*. That is naming the failure and offering no route away from it. Timing is the route.

The five seconds are divided into three beats, and every beat has a number. The fields give them:

  1. Arrival — the material enters and moves. Nothing resembles the mark yet.
  2. Gathering — the material finds the mark's structure. It becomes recognisable partway through this beat, not at the end of it.
  3. The hold — the mark is complete, still, and clean.

The hold law

The last full second belongs to the mark alone.

From the start of beat three to 5.0 seconds:

  • the mark is fully resolved — every letter closed, every edge clean, nothing still forming
  • it is static. No drift, no breathing, no residual ripple, no parallax
  • nothing crosses it. No particles, no highlight sweep, no lens flare, no leftover material passing in front
  • the camera has stopped

A brand intro that is still resolving at 4.8 seconds has not delivered its one deliverable. If your effect needs longer, start it earlier — never borrow from the hold.

Say the clearing out loud in beat three. The commonest failure is not the mark forming late; it is stray material left in frame, still moving, after the mark is done. Write *all remaining material has left the frame* into the third segment, not just *the logo holds*.

Legibility

The mark is readable from the moment it is recognisable, and never becomes less so. Legibility only increases across the five seconds; it does not dip.

  • Refraction, reflection and particle density are tuned so the mark reads *through* them, not despite them.
  • If an effect at full strength would obscure the mark, run it at half strength. The effect serves the mark.
  • Letterforms are never distorted to fit the material. The material conforms to the letterforms.
  • Every character is correct. A mark with a wrong or malformed letter is a failed take, however good the motion.

Ground

  • Plain and dark, or plain and near-white, as the fields say.
  • No environment, no room, no horizon, no set.
  • One light direction, consistent for all five seconds.
  • Whatever depth-of-field exists is on the ground, never on the mark.

Never

Put these in the closing constraints of the prompt you write:

  • ✕ a mark still forming, drifting or settling in the last second
  • ✕ any material, particle or highlight crossing the mark during the hold
  • ✕ camera movement during the hold
  • ✕ legibility that decreases at any point
  • ✕ letterforms distorted, thickened or thinned by the effect
  • ✕ a wrong, malformed or missing character
  • ✕ the mark rendered in the prompt's script rather than the field's
  • ✕ shattering, explosion, spatter or debris
  • ✕ a background with a scene in it
  • ✕ text besides the mark itself

The result

A prompt you can paste straight into Seedance: three timestamped segments that tile five seconds, a mark spelled twice, a locked camera, and a last second in which nothing happens except the thing you made the film for.

The whole template, in one piece. The steps below are the entire workflow.

How to use it

  1. Copy the whole template with the button above and give it to any agent.
  2. Answer the five questions it asks. It writes the Seedance prompt for you — you do not fill the braces yourself.
  3. Paste the prompt it returns straight into Seedance. The `--ratio`, `--resolution`, `--duration` and `--camerafixed` flags are already on the end of it.
  4. Scrub to the start of your hold. The mark should be complete and motionless from there to the cut; if it is still settling, the effect was too long, not the rule too strict.

What you swap

The mark, and what it says
a wordmark reading exactly VIBARY, set in a clean geometric sans, all caps
The material it arrives out of
clear glass, arriving as slow-drifting liquid blobs that refract the background and gradually merge
The three beats, with seconds
0.0–1.8 arrival: blobs drift in from the edges · 1.8–4.0 gathering: they settle onto the letterforms, readable by about 3.2s · 4.0–5.0 the hold: resolved, static, all remaining glass gone from frame
The ground and the light
a plain near-black ground, with one soft key from the upper left
The palette
cool white and pale cyan in the glass, against near-black

Before it runs

  • A mark you can spell out exactly. Video models get letters wrong, and a sting with a malformed character is a failed take however good the motion.
  • An effect that can finish early. If your material needs four seconds to settle, it has taken the hold — start it sooner or use less of it.
  • A willingness to check the clock. The whole skill is a promise about timing, and the only way to know it was kept is to scrub to the hold and look.

When to reach for it

Logo stings, title cards, launch films, the three seconds before a keynote. Templates in this genre describe the journey in loving detail and never say when anything happens, then list “don’t let the effect obscure the logo” among the don’ts — naming the failure without offering a route away from it.

What changes

  • A finished Seedance prompt, not a brief to translate — paste it and shoot.
  • Three `0.0s-1.8s` style segments that tile the five seconds with no gap and no overlap, which is Seedance’s own native syntax for timing.
  • The last full second reserved for the mark alone — complete, still, with every remaining particle explicitly cleared from frame.
  • The wordmark spelled twice inside the prompt, because video models get letters wrong.
  • `--camerafixed true` by default, which removes the commonest way the hold fails.

Pairs with