Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →gooseworks-ai/render-model-comparison-grid
Render a 'model comparison grid' video from a config — a fal-style "same prompt, N contenders" showcase — a dark real-DOM stage where per beat a monospace prompt fades in centered, docks to a small top strip, then a labeled 2-4 panel grid (static images OR muted video clips, mixable per cell) staggers in and holds for comparison, plus a minimal end card — frame-stepped via Playwright (video cells are frame-seeked deterministically) and encoded with FFmpeg. Deterministic assembly, FREE (cell medi
Passed 21 of 22 test cases with the skill, and 8 of 22 without it.
and 10 more in the full report
13 fixed8 passed either way1 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
One of 13 test cases the skill fixed. Same model, same prompt, with and without the skill.
What is the standard holding duration for the minimal end card in the baseline comparison grid timing layout?
| Checked | Without skill | With skill |
|---|---|---|
| specifies end card duration of 2.5 seconds | Missed | Passed |
The instructions your model reads when the skill loads.
Render the 'model comparison grid' format from a config. The signature of this format is a
"Same prompt. N models." gauntlet: a dark stage where, per beat, a PROMPT eyebrow +
the (condensed) prompt fades in centered in monospace and holds readable ~0.8s, then
docks to a small top strip while a grid of 2-4 labeled panels staggers in (0.15s apart)
and holds for side-by-side comparison. A persistent model/variant label sits under each
panel; column order is identical on every beat. Ends on a minimal end card (headline +
column names only — no meta-stats line).
The grid is media-agnostic per cell: any cell is a static image or a muted video clip (i2v outputs, screen recordings), mixable within one beat. Video cells loop during the hold and are frame-seeked deterministically (the renderer awaits each seek), so the render never depends on wall-clock playback timing.
The renderer itself is FREE/deterministic (Playwright frame-step + FFmpeg). The paid inputs
are separate capabilities: the cell images come from create-image-fal, the cell
clips from create-video-fal, and the music bed from create-music-elevenlabs.
Prompt text and labels are real DOM — never AI-rendered.
Default shape: 5 beats × 4.5s + 2.5s end card = 25.0s @ 1280×720/30fps, all configurable
from one config.json.
build_composition.py --config config.json --output hyperframe.html ; render_seekable_hyperframe.py hyperframe.html master-silent.mp4 --fps 30 --width 1280 --height 720 — dark stage, staggered grid, deterministic, $0. The config schema is documented at the top of scripts/build_composition.py; scripts/config.example.json IS the shipped worked example (re-point the cell paths at your own media; its contenders, beat axes, prompts, end-card copy and music prompt are the demo's — the recipe's user choices supply yours, never copy them as defaults).
build_composition.py validates every cell path and the column count (2-4), infers each
cell's media type from its extension (.png/.jpg/.jpeg/.webp → image; .mp4/.mov/.webm/.m4v
→ muted video), and emits a self-contained HTML that exposes window.mediaReady() +
window.renderAt(t). render_seekable_hyperframe.py awaits both, so <video> cells seek
to the right frame before each screenshot — never a frozen first frame.
renderAt(t) — never CSS animation-delay/transitions
(Playwright scrubbing traps delayed animations in pre-state)..movtest cases passed, vs 8 / 22 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.