First snapshot round showed a ghost click-ring floating at the left edge of the opening logo frame: every click ripple was a fromTo tween, and GSAP's immediateRender painted the last click's starting state at t=0. Fixed with immediateRender:false on the ripple tweens — the logo frame is clean in the re-snapshot.
The pin state first pointed left, not down: the sharp corner (bottom-left radius) rotated +45° lands on the west point. Rotation changed to −45°, and the re-snapshot shows a proper downward map pin with the travelling accent dot as its hole.
The 'is this you?' click originally landed between the two chips; the cursor and ripple now target the Yes chip. The globe content also started small inside the circular state (300px in a 483px disc) and was enlarged to 384px so the users fill the globe.
Check passed with 0 errors. The 28px type floor at 1080p flagged 13 elements on the first pass — all reading text was raised to 28px and the two decorative labels (wordmark tagline, bottom caption) were marked data-layout-ignore. Three line_unread craft warnings remain and are inherent to the format: a 12-state loop holds each state ~0.8s by design, and the closing logo is the loop's first frame, so in playback it never actually leaves.
Original prompt (shared by creator)TestedCommunity
@verbove's one-shape UI morph loop — community prompt (tested)
Shared prompt: collected in the MIT-licensed awesome-opus5-5-videos repository, which invites anyone to copy a prompt and run it with their own agent; the prompt itself is credited to its author, @verbove. Our video is an independent test run of that prompt — the creator does not endorse Motion Prompt Lab.
Banned list —Beat grid —Build method —Self-review —
<inputs>
Ask me for:
• my product Kindred + URL
• 8–12 UI states that tell its story
• the real data shown in each state
• brand colors + fonts + accent color
• required formats (1:1, 16:9, 9:16)
</inputs>
<rules>
One HTML file. One canvas
One draw(t) function
No CSS transitions
No timers
No state carried between frames
One shape, never cut
Every state is the same element changing size, radius and color while the content swaps
A cursor drives the sequence with real clicks, typing and one drag
Real UI. Real data. No placeholders.
</rules>
<structure>
120 BPM grid
Something happens on every bea
logo → button → handle field → “is this you?” → loader → pin → globe filling with users → matches → tabs → RSVP → search → logo
</structure>
<motion>
Closed-form springs everywhere with only a tiny overshoot.
If a value changes target multiple times, sum one spring per change
Content enters after its container starts morphing and leaves before the next morph so text never overlaps
Use a short blur on transitions
Never fade black directly into the accent color. Move an accent element between states instead
Make tab indicators stretch by putting each edge on a different spring
Zoom the camera so every state fills the frame
Make the last frame equal the first so the whole thing loops
</motion>
<export>
First render one frame per beat as a contact sheet
Fix anything cramped or broken
Then render every frame in headless Chrome at 60fps, averaging 6 subframes for motion blur
Pipe into ffmpeg:
H.264 + yuv420p
Export all aspect ratios in parallel
</export>
motion design is becoming a prompt
Use this prompt
Paste into: Your AI coding agent (Claude Code, Cursor, …)
Heads up: this prompt writes code. It renders inside a coding agent — pasting it into Veo, Google Flow or Kling will not work.
Copy the prompt into your AI coding agent and let it build the project.
Replace the {{placeholders}} with your own words, numbers and brand.
Render the video — or skip the prompt and open the downloaded project files in HyperFrames directly.
The prompt is a director's spec: it asks for the product's own states and data, so our run supplies a fictional social app, Kindred ({{app}}), and walks the prompt's exact state chain — logo → button → handle field → “is this you?” → loader → pin → globe filling with users → matches → tabs → RSVP → search → logo — at a steady beat, last frame equal to the first. Two disclosed deviations: the original specifies one canvas with a single draw(t) function; our rendition is a HyperFrames DOM morph (one shape element, transform + radius morphs, GSAP back-ease standing in for the closed-form springs) following the same rules — one shape, never cut, content swaps only while the container morphs, the accent element travels between states instead of fading, and a cursor drives the sequence with clicks, typing and one drag. The only edit to the prompt text is {{app}} in the inputs line; the rest is verbatim, including the author's truncated 'every bea' line as collected in the repo.