agentleFS
LogicOberLogicOber/Worldview-Skills/skills/core-engine/film-prop-sheet/SKILL.md

film-prop-sheet

Skill2 starsChanged 50 days agoApache-2.0View on GitHub

Stage 4 of the AI film pipeline — generate and lock reference sheets for recurring props and wardrobe items (match ball, aged ball, homemade trophy, football kit on a ghost mannequin, notebook), do text/number/logo edits as a separate pass, and handle props that carry legible on-screen text. Use when a prop appears in more than one shot, must be worn consistently, carries text, or is a story clue.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Never ask one model to nail a design *and* its fixes in one pass.

    line 52
  2. Very specific:

    If glyphs warp, retry with gemini-3.1-flash-image-preview (stronger at legible text).

    line 62
  3. Distinctive:

    If the text is also a story clue, pre-render the written page as a prop sheet (notebook open, text already written) and pass it as a …

    line 70
  4. Distinctive:

    When there is no budget for a sheet, the substitute is a unified description sentence: one sentence, written once, pasted verbatim into …

    line 85

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Prop & Wardrobe Sheet (stage 4)

Why

A prop that recurs is an identity, like a face. A prop that carries text is a place where the model will write scribbles unless told the exact words. A prop that is a story clue (the brooch in an Odyssey, the hand-painted "SANTIAGO 30") must look identical every time or the clue is broken. All three earn a sheet, locked once, referenced everywhere.

Which props need a sheet — decision

PropSheet?Reasoning
appears in ≥2 blocks (match ball)yesrecurrence = identity
carries text/number/logo (kit, notebook)yes + text passtext is where slop shows first
held in close-up (notebook, trophy)yesfills the frame; must be real
story clue / distinctive item (a scarf crest, a brooch)yescontinuity-critical by definition
era marker (old ball, homemade trophy)yesprops sell the era fastest
set dressing (lamps, plants)no — lives in the platenobody tracks it
generic crowd items (flags, scarves en masse)no — text onlyno individual identity

Rule of thumb from the pipeline author: if you predict the next generation will show this object and a viewer would notice it changed, attach its image. If you predict it won't be in frame, don't.

Spec — single object

SUBJECT — one <prop>, <real size>, photoreal. Single object, single clean hero presentation. <material, surface, construction>. <graphic / print>. <wear level: showroom / lightly used / heavily aged: yellowed panels, cracked seams, frayed stitching, grass stains, under-inflated>.
VIEW — one object, 45° three-quarter front, whole object in frame, centred.
BACKGROUND — seamless neutral mid-grey studio sweep. No gradient banding, no horizon line, no cast props, no text overlays.
LIGHTING — flat even reference lighting: broad soft key at 45°, equal fill opposite, subtle top rim to lift the object off the grey. WB 5600K, soft contact shadow.
COLOR (60:30:10, locked at asset stage) — 60% <…> / 30% <…> / 10% <…>.
STYLE — 8K photoreal product/reference photography. Real materials, real weight, correct contact shadow.
CONSTRAINTS — grey bg only; one object; single view; even legible lighting; true-to-life scale; whole object sharp. 16:9 or 4:3.

Era props: describe the age physically (peeling panels, patchy gold paint over rust, wire and twine bindings, no welds). "Old" alone gives a clean ball with a sepia tint.

Spec — wardrobe / kit (ghost mannequin)

Prop sheet of one <team> football kit on an invisible ghost mannequin — no body, no head, no face. Two views side by side on one 16:9 frame: FRONT (left) and BACK (right). Jersey, shorts, socks, boots present and aligned as worn.
Colours: <exact>. Crest: <position, simple mark>. Number <NN> on the back in <white block numerals>. Real knit texture and drape.
Plain neutral grey seamless background, flat even studio light, 5600K, soft contact shadow under the boots. Photoreal; no text other than the number.

Vibrant is fine — the video model dims it to scene light.

Two-pass rule: render, then edit

Never ask one model to nail a design and its fixes in one pass; a word-level fix is a different job than a render.

  1. Render with gpt-image-2, batch of 4, pick.
  2. Edit the pick with the pick as reference and a one-line instruction:
cohub generate "Change the crest to a simple hexagonal monogram and swap the number 7 to 23. Keep everything else identical — colours, cut, layout, background." \
  --model gpt-image-2 --image assets/candidates/kit/pick.png --param size=2048x1152 --param quality=high --async --json

If glyphs warp, retry with gemini-3.1-flash-image-preview (stronger at legible text).

Props with legible text in the video

Text on a prop fails silently: the acting and camera can be perfect and the notebook is scribbles. Three things fix it, all of which go in the shot prompt, but the decision is made here:

  1. Quote the exact lines: "GUILT — missed penalty" / "Self-blame, unresolved" / "Roots: childhood / football".
  2. Hold the surface ≥2 s in a dedicated close-up at an angle that makes text legible (high angle looking down on the page).
  3. If the text is also a story clue, pre-render the written page as a prop sheet (notebook open, text already written) and pass it as a reference.

Generate / pick / lock

SK=<abs path of film-pipeline>/scripts
$SK/batch.sh --model gpt-image-2 --copies 4 --param size=2048x1152 --param quality=high \
  --out films/<slug>/assets/candidates/ball films/<slug>/prompts/prop-ball.txt
$SK/poll.sh films/<slug>/assets/candidates/ball

Pick test: one object, whole and sharp; neutral grey; correct scale cues (a ball is ~22 cm, not a beach ball); text/number legible and not mirrored; no render look (glow, emissive edges, perfect symmetry). Lock to assets/props/<name>.png, public URL, asset-map.json with ≤15-word desc, blocks, and continuity_critical (e.g. ["number 23","crest"]).

Text fallback when a sheet cannot be generated

When there is no budget for a sheet, the substitute is a unified description sentence: one sentence, written once, pasted verbatim into every prompt where the prop appears. The sentence names material, shape, one asymmetric detail and wear:

the ancient brass lamp on a round stone pedestal, four empty socket holders around its rim, worn runes engraved on the metal

Store it in asset-map.json under text_fallback. Never reword it between prompts — paraphrase is how the lamp gains a fifth socket. Upgrade to a real sheet the moment credits allow; the sentence then becomes the sheet's prompt.

Every prop has an origin

A prop that matters is introduced, not discovered. The script places a beat in which someone brings it from somewhere and a line or an action says where: "the department assignment, posted an hour ago — I printed it." A prop that is simply there in a character's hand reads as a continuity error even when it is drawn perfectly. Check this in the script before the sheet is made; the sheet cannot fix it.

Screens as props

A laptop or phone that must show a real web page is a prop whose "texture" is a screenshot. Do not describe the page; capture it — see film-screen-capture.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.