rlaope/oh-my-hermes / skills/omh-show-reel
omh-show-reel
Unverified[omh] Video to make from an idea or a product: an explainer, motion graphics, kinetic type, a logo intro, a 3D scene, a data-viz animation, a music visual, a short or a source-grounded product film, authored and rendered to a measured MP4 with locale/export QA and source delivery. Use when the user says: show-reel, show reel, showreel, product showcase video, animated product film, motion showcase, animated portfolio, launch film.
references/art-direction.md
View on GitHub# Independent artistic acceptance and the weakest-scene rebuild
How to set art direction per segment, the predictable screens to avoid, motion
principles, rhythm, typography, safe areas and transitions are in
`references/craft-design.md`. Before an expensive render, compare three distinct
narrative/visual concepts, not palette swaps. For each useful reference record exact source/locator, revision
or inspection date, what composition/motion/sound decision it informs, rights and
limits. A supplied pinned reference can be enough; do not invent a trend claim or
seek unrelated references just to fill a quota. Never confuse imitation with
brand coherence: retain untouched official masters, establish a restrained visual
language, and select material/effect according to the message.
For every timed segment decide medium, palette, type, texture, composition,
focal point, movement/weight, sound intent, entry/exit and cut motivation. Render
style frames at delivery size AND a short moving proof before full production.
Authentic product captures, material/photographic composition, depth, kinetic
phrases, seeded geometry and quiet holds are options, not a required effect list.
## Hero-frame acceptance before the full render
A film that passes every lint, layout and contact-sheet check can still be
rejected on sight. Judge the hero frame (usually the first, which is also the
thumbnail) as a launch frame before rendering the whole film. Ask whether a top
product team could open its launch film on that frame. Then name each element
that reads as template or cheap, with the change that fixes it:
- A product shown in 3D is a lit model with real materials (a GLTF device, for
example, with the product screen as a live texture), not stacked CSS layers
or flat cards. A stacked-layer object reads as cardboard however it is
animated.
- Materials are measured, not eyeballed: crop the product region from a
rendered frame and read its mean luminance and clipped share. A light-coloured
surface whose bulk sits near white with a radial hotspot reads as a glowing
panel, so raise roughness, lower clearcoat or dim the key light until its
tonal range opens.
- The title is set in the brand's own face and colour (see
`references/storyboard-and-evidence.md`), the background carries the brand's
own art or a deliberate material, and on-screen product UI reads as a
shipped app rather than bare boxes.
When the first frame is the thumbnail, a value it shows is already at its final
state there: a counter caught at zero or mid-roll puts the wrong number on the
film's preview still. A film whose concept opens another way on purpose
(`references/craft-workflow.md` has that intent written in `concept.md`) names
the frame that serves as the still instead and passes its time to
`finish.py --poster`.
## Frame checks before the first style frame
Run these on the storyboard, then again on the rendered style frame. Each is a
measurement with a stated number, not an impression.
- `frame_fill`: the area a product shot fills depends on the subtitle mode,
using the zones `references/craft-design.md` (Layout and safe areas)
defines. With no subtitles, a product-led shot is full-bleed: it uses the
full frame minus the 46 px edge inset, so a 1080p product shot is at least
about 988 px tall (1920x1080: at least 1828x988), and in 9:16 it still keeps
the top 250 px and bottom 420 px clear. With subtitles, it fills at least 90%
of the width and height of the body safe area (16:9: x 120-1800, y 120-900)
and the subtitle band y 940-1040 stays clear. A shot that is not
product-led, such as a title or a quote, follows the safe areas instead.
Text rows are at least 2% of the frame's shorter side (22 px at 1080). Choose terminal columns and
rows, or the viewport, from that area and row height; never keep a default
size and letterbox it. Record the area used, the measured fractions and the
row height.
- `clear_space`: a logo or identity mark (one the product renders, such as an
ASCII logo) keeps clear space of at least one of its own text rows (one
terminal row for an ASCII logo) from the frame edge and the window edge.
Flush against either is REVISE. An intentional crop, as `craft-design.md`
allows, is exempt; headings follow the type and safe-area rules instead.
- `decorative_text`: no text the product does not render and the brief does
not need. Window title bars, `cols×rows` dimension labels, invented paths
and prompts, and fake version strings are removed, not restyled.
- `palette_families`: for a composed frame (promo shot, poster, slide, end
card), list every hue family on screen with its hex values and where it
appears (background, window chrome, product palette, logo, captions,
accents). The captured product or brand palette is given and never
repainted. The surrounding composition adds at most one accent family beyond
it. Neutrals, semantic-state colors and data-series palettes are listed but
not counted. Check text contrast on its actual background: at least 4.5:1,
or 3:1 for large type. A composed frame that adds more than one accent
family beyond the product or brand palette, or misses contrast, is REVISE
before render.
In an authored 9:16 frame, run these checks on a layout composed for that
frame: content stacked down the height between the top and bottom zones that
`references/craft-design.md` (Layout and safe areas) keeps clear. With
`frame_fill`, record the y of the highest and the lowest content edge. A shrunk
landscape layout, the 16:9 arrangement scaled to the narrow width with unused
height under it that the storyboard does not name as intended space, is REVISE:
recompose it for the tall frame. Captured footage that cannot be recomposed
keeps its lane's framing (`references/game-capture.md`) or takes a virtual
camera that follows the subject.
## Anti-slide evidence, not keyword approval
Record `anti_slide_review` from actual segment and transition evidence. Compare
adjacent shots' hierarchy, focal geometry, media, pose and rhythm. Reject repeated
headline/card/rail grammar when every shot is a centered title plus the same
boxes/rail with only text/color swapped. Do not repair it by sprinkling particles
or changing a palette. Recompose the narrative around the evidence/semantic object,
change medium/focal scale/camera purpose, or remove an unnecessary card. Rhythm
includes the arc a scene runs: two adjacent scenes that both build one value and
then rest repeat one template even when their layouts differ, so give the second
another arc, such as opening on its result. A coherent
recurring motif is valid when it develops meaning; justify that from the output,
not a checklist keyword or numeric quality score.
Name strongest and weakest scenes, why each succeeds/fails, and keep/fix/rebuild
for every segment. Reserve actual iteration/time budget to rebuild the weakest.
When the weakness is the subject itself (the feature shown is thin, not its
staging), record `product_gap` with the capability the film needs and report
it to the user as a product change; do not spend the budget on particles,
glow, palette or camera moves to hide it.
`before_after_and_boundaries` requires same-time before/after samples, consecutive
moving strips and both entry/exit cut boundaries from real renders. Re-render and
review the final encoded revision; approval of the earlier draft is not approval
of the new file. Budget exhaustion with unresolved defects means REVISE/HOLD.
## The no-context judge
`references/craft-qa.md` (Critique) has a subagent with no context of the
making judge the sheets, the frame strips and its question table. Three
additions bind that pass to the record:
- The record names who judged. Each keep/fix/rebuild row carries `judged_by`,
`maker` or `no_context`, and `artistic_verdict` says which of the two it
rests on. Where the host can start no fresh session the value stays `maker`
and the delivery says the no-context pass was not observed.
- Ask one question before the table's own: what does each scene say, read from
the frames alone, and is anything misspelled, cut off or at odds with another
scene? The judge has no script, so the maker compares that reading with the
segment's `message`; a scene read differently from its message is fix or
rebuild.
- The judge runs two rounds by default, a number the brief's revision budget
may change. Each round is a new session, so the second judge also has only
the frames and the questions. A rebuild still open after the last round goes
to the person with the judge's rows and the frames: they choose another
round, cutting the scene, or accepting the film with the defect named in the
delivery. Until they do, `artistic_verdict` stays REVISE.
## Four independent verdicts
`technical_verdict` binds final hash to decode/frame/color/audio/size observations.
`artistic_verdict` binds that SAME revision to actual delivery-size style, story,
material/motion/cut critique, source fidelity, anti-slide decision and rebuilt
weakest scene. `playback_verdict` needs observed full-speed viewing;
`listening_verdict` needs observed hearing of final audio, or reasoned not_applicable
for a muted brief. Missing required observations remain HOLD/not_observed.
Automated motion/loudness numbers, source inspection and tool names cannot fill
those fields. Native-language and accessibility review remain separately scoped.
A technique/locale fixture is never the complete commissioned film or evidence
that real product actions shown as concepts actually ran.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |