Asset Library
1,738 video parts you can take today, and 557 more to learn from. Find one you like, then open it for the details.
Showing 121 of 2,295Show everything
No preview yet · 121
locate.mjsTool · Apache-2.0Finds a point in an image by drawing numbered grids over it with ffmpeg. Its auto mode posts the image to Google if GEMINI_API_KEY is set.verify-motion.mjsTool · Apache-2.0Compares the motion in a render against a reference export (Figma's own MP4) and fails under 15 dB. Described in the figma SKILL.md; the…remotion-studioSkill · Link onlyLaunching Remotion Studio to preview a video: when to let it open the browser itself and when to print the address for an in-app browser…creative/video-understand-usageSkill · Link onlyUsing the local vision model on frames and clips: modes, model choice, which frames to pick.meta/checkpoint-protocolSkill · Link onlyWhen and how to save a stage's state and when to stop and wait for the person. A stage marked as needing approval cannot be recorded as…meta/reviewerSkill · Link onlyThe self-review an agent runs after every stage: check the file against its schema, against the stage's review points and the style rules…character-animation-qaSkill · Link onlyReviewing local character animation: schema checks, browser previews, frame samples, final output checks.video-understandSkill · Link onlyUnderstanding a video locally with FFmpeg frame samples and Whisper transcription.web-design-guidelinesSkill · Link onlyReviewing interface code against web interface guidelines.audio_probeTool · Link onlyReads the length and stream details of an audio or video file with ffprobe. Runs on the machine; needs ffprobe. Works with no keys.composition_validatorTool · Link onlyChecks a planned composition before rendering: catches audio and picture lengths that do not match and similar plan errors. Runs on the…frame_samplerTool · Link onlyPulls still frames out of a video with FFmpeg, at set times or intervals. Runs on the machine; needs ffmpeg. Works with no keys.video_analyzerTool · Link onlyTakes a reference video apart: transcript, scenes, sampled frames and structure, written up as a brief the agent plans from. Runs on the…video_understandTool · Link onlyDescribes frames or clips with a local vision-language model. Runs on your own GPU with model weights you download; needs transformers…visual_qaTool · Link onlyAutomatic checks on a rendered video: black frames, frozen stretches and similar faults, from sampled frames. Runs on the machine; needs…character_animation_reviewerTool · Link onlyChecks a character animation: rig integrity, motion and render readiness, as a report. Runs on the machine. Works with no keys.BacklotTool · Link onlyA local web board that watches the projects folder and shows a production as it runs: stages, script, scene plan, generated assets as a…/backlotSkill · Link onlySlash command that opens the Backlot board for the current or named project. Copies for other agents sit in .cursor/commands…Slideshow risk scorerTool · Link onlyScores a scene plan from 0 to 5 on six things that make a video feel like a slide deck: repetition, decorative visuals, motion without…Scene variation checkerTool · Link onlyChecks a scene plan for sameness before any asset is made: shot sizes, camera movement, lighting, a flagged high point, texture words…Delivery promiseTool · Link onlyClassifies what a video has promised to be (for example led by motion or led by stills) at proposal time and locks it, so the render cannot…Checkpoint storeTool · Link onlyWrites and reads a saved state after each stage so a run can resume, and refuses to record a gated stage as done without recorded approval.Source media reviewTool · Link onlyInspects footage a user supplies (probe, sampled frames, scenes, transcript) and writes one normalized review file, so planning starts from…Scene pacing checkTool · Link onlyChecks that the steps of a fake terminal scene land in time with the narration cues, and fails loudly when they drift.Tool event logTool · Link onlyAn append-only log of tool events per project, written whenever a tool runs and read by the Backlot board.atelier_snapshots.pyTool · Link onlyRenders one review still per scene of a hand-written composition so the board can show them.backlot_simulate_run.pyTool · Link onlyFakes a pipeline run on disk so the Backlot board can be watched updating; useful as a demo of the board.backlot_screenshot_stage.pyTool · Link onlyBuilds fictional projects and captures the board screenshots used in the README, with Playwright.backlot_visual_eval.pyTool · Link onlyA screenshot test for the board: captures fixed views and compares them with saved ones.backlot_watch_captures.pyTool · Link onlyCaptures a board screenshot each time a watched project changes.kling_official_animated_explainer_e2e.pyTool · Link onlyAn end-to-end check of the Kling provider path; a free dry run by default, paid modes on request.atlas_media_smoke.pyTool · Link onlyAn opt-in paid test run of the Atlas Cloud media tools.character_qa_report schemaStarter · Link onlyA review of rig integrity, motion and render readiness. A JSON Schema; required fields: version, status, checks.decision_log schemaStarter · Link onlyThe audit trail: every meaningful choice with the options considered and the reason, appended and never edited. A JSON Schema; required…final_review schemaStarter · Link onlyThe self-review of a finished render: a technical probe, frames looked at from four points, an audio check, whether the promise was kept…review schemaStarter · Link onlyA stage review: findings each graded critical, suggestion or nitpick, with the agent's response and whether it was accepted. A JSON Schema…source_media_review schemaStarter · Link onlyThe review of footage a user supplied, with what it means for planning. A JSON Schema; required fields: version, files, summary…checkpoint schemaStarter · Link onlyThe resumable snapshot written after each stage. A JSON Schema; required fields: version, project_id, pipeline_type, stage, status…gsap-performanceGSAP · MITShort rules for smooth playback in a browser: animate transforms and opacity rather than width, height, top or left; will-change only on…aesthetic-rules (review rules)Skill · Apache-2.029 rules written as past rulings, in five groups: rhythm (R1-R4), look, camera and composition (Q1-Q11), sound (S1-S5), copy (C1-C3)…final-review (independent last check)Skill · Apache-2.0The brief for a fresh reviewer who checks the finished film against the plan, the chosen cards, the demo source and the aesthetic rules…Gallery (static preview site)Tool · Apache-2.0A static site that lists the 124 cards and 150 moves with search, filters and copyable names. The data is gallery/api/library.json. 54…smoke-render-demos.py (first-frame test of every demo)Tool · Apache-2.0Registers every demo that exports a duration into a generated root file and renders its first frame, to catch demos that compile but crash.shot-polish (the redesign review rig)Tool · Apache-2.0The author's tooling and notes from the October 2026 pass that re-made all 216 demos and kept 150: per-shot notes, a keep/cut decisions…judgment.md (the eight judgments)Skill · Apache-2.0The taste layer in full: eight judgments each with when, prefer, because, evidence and where it stops being true; twelve default reaches…slop.md (the review file)Skill · Apache-2.0The checklist run on a plan and on a render: the never-ship list, 25 ranked default patterns (N1-N25), a table of AI default against…taste/verdicts.md and taste/edits.mdSkill · Apache-2.0The author's own record of what one person said about specific films, and 18-odd before/after fixes with the reason for each. It is one…studies (two calibration strips)Skill · Apache-2.0Two JPEG strips of invented brands, each pair showing a default beside a decision, drawn in different styles on purpose so they are not…examples/selftest (runtime regression page)HTML · Apache-2.0A test page for the runtime itself: visibility tracks, per-word options, multi-line typing, origin claims, keyframe steps. It must print…Motion.selfTestGSAP · Apache-2.0Proves a film is seek-safe: plays every frame in order, then jumps to random frames and compares every element's style, transform and text…look.pyTool · Apache-2.0Inspects a rendered film: contact sheet, cut list, frames just before each cut, an audio audit, and gates G0 (render exists and moves), G2…lint.mjsTool · Apache-2.0The page-chrome lint, gate G5: loads the film's HTML and fails corner labels, tracked-caps labels, counters, timecodes, header or footer…blind_review.pyTool · Apache-2.0Builds a local page for choosing between two or more versions blind: random left/right, synced play, frame step, a pick and notes, with the…showcase comparisons (Cluely, Spotify, Wispr Flow)Skill · Link onlyThree before/after films, GIFs and posters that put other companies' brands on screen to show the difference the skill makes. The NOTICE…Safe mode (BLENDER_MCP_SAFE_MODE)Tool · MITAn optional checker, off by default, that reads every script the model sends before it reaches Blender and rejects it if it imports…Context log (BLENDER_MCP_CONTEXT_LOG)Tool · MITAn optional local log, off by default, of how many characters and estimated tokens each tool definition and tool result costs the…lookTool · MITReturns one picture of the scene so the model can check its work: the user's viewport, the view through the scene camera, an auto-framed…get_scene_infoTool · MITDescribes the scene in text: one header line, then one line per object with the fields asked for (location, size, whether it sits on the…get_addon_statusTool · MITReports the Blender version, whether the add-on matches the server, which libraries and generators are switched on, and whether data…disable_telemetryTool · MITTurns the opt-in collection of prompts, code, screenshots and scene data off from inside a chat. It can only turn it off, and it does not…record_trajectory_feedbackTool · MITAsks the model to file the user's reaction to a result (accept, reject, undo, or a correction in their words) into the author's training…open_viewportTool · MITShows the latest viewport picture in a panel beside the conversation.viewport_latestTool · MITReturns the stored viewport picture if it is newer than the one the panel has; does not touch Blender.viewport_captureTool · MITTakes a fresh viewport picture for the panel, on Refresh or by itself after the scene changes.viewport_pickTool · MITFinds the object under a click on the viewport picture, so a person can point at a thing and attach it to their next message.search_mentionsTool · MITLists scene objects, materials and collections matching a few typed letters, for @-mentions in the message box.blender:// resources (object, material, collection)Tool · MITThree MCP resource addresses that return one scene item's details as JSON, used by the @-mentions.Viewport app (apps/viewport.html)Tool · MITA single HTML page the server hands to clients that support MCP Apps: a live viewport picture beside the chat with refresh and…asset-optimizationBlender · MITThe gate before export: an audit checklist for geometry, UVs, materials and naming, fixes for common overruns, LOD steps at 50, 25 and 10…lookdevBlender · MITThe loop that gets shading right: gray clay pass, base materials, a neutral three-point rig, then beauty light, a screenshot, a list of…qa-reviewBlender · MITA ship or no-ship review backed by screenshots: five required views, four defect levels, and a verdict template. It tells the agent never…reference-image-match (shared reference)Blender · MITThe pack's best single page, though it is a shared reference and not one of the 94 skills: a written protocol for making a Blender scene…dream-loopSkill · MITThe skill itself, 2,900 characters: first get a target picture (the user's, or one made with an image generator and prompted as a real…Pro workflowSkill · MITThe loop with a judge: build a first pass, take a screenshot, give the screenshot and the target to a fresh sub-agent that scores…fal-batch.test.mjsTool · MITNine tests for the Fal helper that run against a mocked network: concurrency, resuming without paying twice, uncertain submissions never…preview-server.pyTool · MITA 46-line Python web server for the loop's screenshots: it serves the working folder on this machine only and accepts a PNG posted to…threejs-debug-profilerthree.js · MITHow to find why a three.js page is blank, broken or slow: reproduce first, then an ordered checklist for blank canvases (canvas size, WebGL…threejs-qa-releasethree.js · MITHow to prove a three.js game works the way a player meets it: build, open in a real browser, check console errors and non-blank canvas…inspect-threejs-canvas.mjsTool · MITOpens a running page in headless Chromium, checks the canvas is not blank, and measures the picture: color entropy, share of the dominant…check_evidence.pyTool · MITChecks that the screenshots and reports a finished game claims really exist on disk, belong to the current run, are not blank and recorded…Test hooks and diagnostics contractTool · MITTwo objects a page puts on window so a script can take a repeatable picture of it: diagnostics (frame count, renderer draw calls and…visual.spec.tsTool · MITA Playwright smoke test (one test, run on a desktop and a phone profile): the canvas is big enough and its pixels vary, so a blank or…visual-regression.template.tsTool · MITA Playwright screenshot-baseline template that seeds, sets a named state, freezes, hides debug UI and then compares against a stored…bot-playtest.template.tsTool · MITA Playwright bot that holds keys on a script and measures whether the game really plays: frames advanced, distance moved, score gained, and…Visual scorecardSkill · MITA ten-category rubric for judging a real-time 3D picture from 0 to 3 (art direction, hero, obstacles, rewards, world, materials, lighting…Render budgets and material kitSkill · MITStarting budgets for a browser 3D scene on desktop and phone (draw calls 300 and 150, triangles 750k and 300k, textures 60 and 40, shadow…Evidence manifest formatSkill · MITA small JSON format that declares which viewport and state pairs one verification pass must capture, with a run id, so old screenshots…audit.mjsTool · MITProves a figure's geometry instead of trusting the eye: records each part's true shape as you draw it, ray-casts every overlapping pair to…turn-audit.mjs (3D orbit audit, beta)Tool · MITChecks a 3D figure at every turn angle and in every motion clip: no ordering cycles, no loose groups. Slow by the repo's own account: 15 to…capture.mjsTool · MITScreenshots a figure with headless Chrome through the browser's own debugging protocol, with no npm packages: whole card, phone width, a…drive.mjsTool · MITRuns a scripted session in headless Chrome: move, click, key, wait-until, run JavaScript in the page, slow motion (for example 0.05x) and…lines.mjsTool · MITThe line-end check: on a built page, every seam, rule, rib, tick and pointer must end on a stroke, a dot or an outline. It rings the…turn-check.mjsTool · MITBrowser checks for 3D figures: fidelity against the kit's own builders, a z-buffer order check, pixel pops between tiny angle steps, line…turn-bench.mjsTool · MITTimes the 3D engine in Node: CPU time per kind of part.evals.jsonSkill · MITFive test prompts with what a good answer contains: a token bucket rate limiter, a connection pool for Next.js docs, a CDN edge cache as a…Blueprint proof harnessTool · MITA small offline page that loads one kit, lights it with one sun on neutral gray with no post-processing, and stands at four distances…validation-report.md templateSkill · MITA fill-in report of what was run and seen; every row needs evidence on disk, and an unobserved frame is never written up as passing.Visual review promptsSkill · MITA list of judgment prompts for any rendered 3D view, each with its cheapest check, written as conditions and not as a score: a blank…Frame and interaction inspectionSkill · MITWhich captures a scene owes (overview, close-ups, each important interaction) and what to check on each one, looking at the rendered result…capture.pyTool · MITServes a built scene from a local port, opens it in headless Chromium, visits its named views, clicks controls you name and writes PNGs and…capture_view_quality.pyTool · MITMeasures whether a captured frame is usable at all: near-black, filled by a wall, or flat. Used by capture.py's --expect-usable.manifest_contract.pyTool · MITChecks the catalog manifest is one the helpers support before any record path is read.direction.schema.jsonSkill · MITJSON Schema for one kind of catalog file (direction): required fields id, summary. Structural only; the repo's validate.py adds the checks…evidence-event.schema.jsonSkill · MITJSON Schema for one kind of catalog file (evidence event): required fields schema_version, event_id, change_id, timestamp, event_type…knowledge.schema.jsonSkill · MITJSON Schema for one kind of catalog file (reusable knowledge): required fields schema_version, id, revision, status, record_kind, title…recipe.schema.jsonSkill · MITJSON Schema for one kind of catalog file (recipe): required fields schema_version, id, revision, status, record_kind, title, aliases…source.schema.jsonSkill · MITJSON Schema for one kind of catalog file (source): required fields id, url, title, accessed_at, locator, scope. Structural only; the repo's…blueprint_validation.pyTool · MITChecks that a blueprint record, its module and its proof captures agree. Maintainer tooling for this repo; not part of the installed skill.build-index.pyTool · MITRegenerates the human-readable catalog pages and the coverage summary from the records; --check fails if they are stale. Maintainer tooling…catalog_validation.pyTool · MITThe catalog contract checks used by validate.py. Maintainer tooling for this repo; not part of the installed skill.check-docs.pyTool · MITChecks links and JSON examples in the repo's Markdown. Maintainer tooling for this repo; not part of the installed skill.check-retrieval.pyTool · MITRuns 249 authored search cases against search.py and reports which known intents find their record. Maintainer tooling for this repo; not…evidence-log.pyTool · MITAppends a contract-shaped evidence event and a record snapshot to a change log. Maintainer tooling for this repo; not part of the installed…evidence_log_events.pyTool · MITBuilds the event objects and snapshot paths for evidence-log.py. Maintainer tooling for this repo; not part of the installed skill.evidence_validation.pyTool · MITChecks evidence links and observations for validate.py without certifying that they are true. Maintainer tooling for this repo; not part of…kit-proof.pyTool · MITTurns one blueprint record into four real captures, a proof log with a hash per picture, and an evidence event. Maintainer tooling for this…kit_proof_harness.pyTool · MITAssembles the offline proof page for kit-proof.py with a vendored three: no bundler, no network, no install. Maintainer tooling for this…migrate-catalog.pyTool · MITPreviews or applies the one-file-per-record layout while keeping content and search results the same. Maintainer tooling for this repo; not…package.pyTool · MITBuilds repeatable ZIPs of the skill and the Codex plugin; never installs or publishes. Maintainer tooling for this repo; not part of the…showcase.pyTool · MITRenders the 21 showcase stills, one browser at a time, resumable, inside a time budget. Maintainer tooling for this repo; not part of the…