See Then Build

unleash your imagination

1 of 5 free pages opened. Sign in for unlimited access

Learn

A FIELD GUIDE TO WHAT YOU CAN BUILD

The tech behind the demo

Three.js. TypeScript. MCP. These names describe different ingredients. Learn what each ingredient makes possible—and when it belongs in your request.

Start with the result. “Let customers rotate a product and reveal its parts” gives your AI a useful target. Add “use Three.js” when that capability or an existing project calls for it. A longer stack is not a better prompt.

THE BUILDERYour AI + your direction

Plans, writes and revises the work.

THE INGREDIENTSCode, tools + connections

Draw the scene, respond to clicks, save data.

THE RESULTSomething you can use

A lesson, a product explorer, a working app.

Find a technology 49 entries · six practical uses
What we found in 124 posts—and what the evidence means

Reviewed 100 of 124 unique public archive posts across 19 nonempty editions, through 2026-09-20. Newer posts have not yet received this technology review. Checked original captions, cataloged tool attributions and retained creator-resource reviews. The strongest recurring technical names were Blender (28 mentions, including one comparison) and Three.js (15). TypeScript appeared in one reviewed creator prompt. Private saves, unarchived X replies and unpublished collection candidates are outside this review.

“Named in the archive” means the source names the technology; it does not mean we reproduced the creator’s workflow. Illustrations and suggested prompts are our explanations. Counts are distinct linked posts, not popularity or quality rankings. AI models and general creative services remain in the Tool Library; this guide focuses on implementation ingredients, media workflows and file handoffs.

CSS, JavaScript, SVG and WebGL are labeled background concepts, because their names were not established in the archived creator evidence. MCP mentions can describe use, a suggested option, or a choice to avoid it. POV-Ray appeared as a visual-style reference, not a verified implementation.

Read the complete post-by-post mapping

Show a thing in space

Make objects, explore their parts and render a scene.

01Three.jsPut an explorable 3D scene on a webpage

Browser 3D library · Named in the archive

A JavaScript library for building 3D scenes with objects, cameras, lights and materials inside a webpage.

Think of it as a miniature film studio: place the objects, light them, then give the viewer the camera.

What you could make

An onboarding guide where a new employee rotates a machine and separates its parts to see how they fit.

How to ask for it

Build a Three.js product explorer with orbit controls, labeled parts and a button to separate and reassemble the layers.

Know the boundary. A rotating engine can be an illustration without simulating forces. The library does not establish that the model is accurate.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Put an explorable 3D scene on a webpage

Interactive diagram of an exploded view; not a Three.js renderer.

ObjectsCamera + lightInteractive view
Separating layers can reveal relationships that a flat picture hides. The anatomy and temple demos apply this idea.

FROM THE ARCHIVE

An anatomy atlas that comes apart

Retained tool attribution

…The creator shares Human Atlas’s React/Three.js application and demo; the interface and repository credit BodyParts3D.…
Attribution source ↗
15 linked posts · view evidence
02BlenderMake and refine a reusable 3D asset

3D creation application · Named in the archive

Blender is a workspace for modeling objects, assigning materials, rigging characters, animating and rendering scenes.

Think of it as a digital sculpture studio with an animation stage attached.

What you could make

Make one product model, then reuse it for a brochure image, a rotating view and an assembly lesson.

How to ask for it

Build a simple Blender model of this object. Keep its parts separately named and show front, side and three-quarter views before adding detail.

Know the boundary. A polished render does not establish correct dimensions, a usable rig or who performed each modeling step.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Make and refine a reusable 3D asset

  1. 01Build the shape
  2. 02Add materials + controls
  3. 03Render or export
The character workflow in the archive describes assembly, rigging and repeated visual corrections.

FROM THE ARCHIVE

A beginner refines a character by showing the AI what to fix

Creator caption

…話題の「GPT-6 Astra × Blender」、実際に触ってみました。 これ、かなり衝撃です。 Blender初心者でもできる。ほぼ視点移動（ビュー）の操作だけでここまで作れました。 モデルはTripo製、ステージはBlender上で作っ…
Attribution source ↗
28 linked posts · view evidence
03Python / Blender scriptingRepeat scene-building steps reliably

Programming language and application API · Named in the archive

Python can automate a sequence of operations. Blender exposes an API so Python can create objects, change scenes and render views.

Think of it as a written workshop recipe that can be repeated without rebuilding every part by hand.

What you could make

Create 12 product color variants and render each from the same camera angle for a comparison sheet.

How to ask for it

Write a Blender Python script that applies these colors to the same model and renders each with identical camera and light settings.

Know the boundary. Python runs the instructions you give it. It does not decide whether the object or the comparison is meaningful.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Repeat scene-building steps reliably

  1. 01One scene + 12 colors
  2. 02Repeat the same operations
  3. 0312 comparable renders
The forest-house guide describes a headless Blender/Python workflow; the finished video does not show its script executing.

FROM THE ARCHIVE

A furnished forest-house walkthrough

Retained tool attribution

…creator’s build thread describes Blender, headless Python, lighting, Unreal Engine 5 export and iterative scene refinement.…
Attribution source ↗
1 linked post · view evidence
  • A furnished forest-house walkthrough

    Retained tool attribution: creator’s build thread describes Blender, headless Python, lighting, Unreal Engine 5 export and iterative scene refinement.

04CyclesCalculate how light makes a scene look

Rendering engine · Named in the archive

Cycles is Blender’s renderer for making images by tracing light through a modeled scene and its materials.

Think of it as a virtual photography session where the computer calculates the light bouncing around the room.

What you could make

Compare morning and evening lighting for a proposed training space using the same furniture layout.

How to ask for it

Render the same Blender room in Cycles with two clearly labeled lighting setups. Keep camera and materials fixed for comparison.

Know the boundary. Believable lighting does not prove the materials or room are physically accurate.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Calculate how light makes a scene look

  1. 01Geometry + materials
  2. 02Lights + camera
  3. 03Rendered image
Change the lighting setup while holding the room constant to make a useful comparison.

FROM THE ARCHIVE

A furnished forest-house walkthrough

Retained review of creator resource

…plan, headless Blender/Python workflow, Cycles lighting, UE5 export, iteration and storyboard comparison. Reply text verified in Xpoz data; attach…
Attribution source ↗
1 linked post · view evidence
  • A furnished forest-house walkthrough

    Retained review of creator resource: plan, headless Blender/Python workflow, Cycles lighting, UE5 export, iteration and storyboard comparison. Reply text verified in Xpoz data; attach

05Unreal Engine 5Let someone walk through the scene

Real-time 3D engine · Named in the archive

An engine for building interactive 3D experiences, including scenes that respond as a person moves through them.

Think of it as turning an architectural model into a place you can enter.

What you could make

Walk a client through a proposed exhibition space to assess sightlines before building it.

How to ask for it

Turn this approved room model into an Unreal walkthrough with clear movement controls and collision boundaries.

Know the boundary. A walkthrough does not establish construction accuracy. One archive post reports a Blender-to-Unreal workflow; other mentions concern borrowed assets or past projects.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Let someone walk through the scene

  1. 01Import a room
  2. 02Set controls + boundaries
  3. 03Walk through it
The forest-house recording shows a furnished walkthrough; engine use is creator-attributed.

FROM THE ARCHIVE

A furnished forest-house walkthrough

Creator caption

…launch blog post. From a Blender scene to a Unreal Engine 5 walkable experience. 🧵 https://t.co/xGqe5iXoRv…
Attribution source ↗
3 linked posts · view evidence
06ShadersGive a surface a changing visual treatment

Graphics programs · Named in the archive

Small graphics programs calculate aspects of how a scene is drawn, such as how a surface’s pixels look.

Think of it as a recipe for painting every tiny part of a surface, rather than a single saved picture.

What you could make

Give a product hero an iridescent surface that changes as the viewer rotates it.

How to ask for it

Make the surface shimmer with a shader. Keep the labels stable and readable, and offer a plain-material view.

Know the boundary. A water-like shader is not necessarily fluid physics. “Web shaders” is named in the archive; a specific shader language is not established.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Give a surface a changing visual treatment

  1. 01Surface position
  2. 02Color / lighting calculation
  3. 03Shimmering result
The racing demo explicitly names web shaders and neon shader lighting.

FROM THE ARCHIVE

Voidrunner anti-gravity racing prototype

Creator caption

…you can possibly build on ThreeJS + Web shaders bro! The most important things are blistering sense of speed, tight drifting, boost mechanics, trac…
Attribution source ↗
2 linked posts · view evidence
  • Voidrunner anti-gravity racing prototype

    Creator caption: you can possibly build on ThreeJS + Web shaders bro! The most important things are blistering sense of speed, tight drifting, boost mechanics, trac

  • A gallery of coded 3D interfaces

    Creator caption: and fixing mistakes. Liquid Glass, shaders, motion design, procedural three.js models, and full environments. All from a few prompts. Some pr

07WebGLSend graphics work to the GPU

Background · browser graphics API · Background concept

A browser interface for drawing accelerated graphics on a canvas using the device’s graphics processor.

Think of it as the connection from a stage designer’s plans to the machinery that draws the stage.

What you could make

Display an interactive product model in a browser without asking the visitor to install a 3D desktop app.

How to ask for it

Build a small browser 3D scene and include a helpful fallback if graphics support is unavailable.

Know the boundary. Three.js is a higher-level scene library; WebGL is a lower-level graphics API. WebGL was not established by name in the archived posts.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Send graphics work to the GPU

  1. 01Three.js scene
  2. 02WebGL drawing commands
  3. 03GPU draws the image
These are different layers. Other rendering backends exist; this diagram explains one possible arrangement.
08BodyParts3DStart with existing geometry, then explain it

Anatomical asset dataset · Named in the archive

A dataset of 3D anatomical structures that can supply the objects used in an anatomy visualization.

Think of it as an atlas of parts that the application arranges and labels.

What you could make

Create a lesson that isolates a few muscles and shows their position relative to nearby bones.

How to ask for it

Use attributed anatomical assets with suitable licensing. Preserve source labels and have the teaching content reviewed for accuracy.

Know the boundary. The anatomy demo credits BodyParts3D. Do not infer that an AI generated all its geometry or that the lesson is medically validated.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Start with existing geometry, then explain it

  1. 01Existing anatomical parts
  2. 02Select and label structures
  3. 03Explore their relationships
Assets supply content; React and Three.js can supply controls and display.

FROM THE ARCHIVE

An anatomy atlas that comes apart

Retained tool attribution

…not imply Astra modeled all 2,234 assets: BodyParts3D is explicitly credited in the visible UI.…
Attribution source ↗
1 linked post · view evidence
09VoxelsBuild a world out of tiny blocks

3D representation · Named in the archive

A voxel represents a value in a volume; in block-style scenes, occupied cells become visible building blocks.

Think of it as pixels with depth: small cubes can assemble a wall or a whole landscape.

What you could make

Make a simplified factory layout whose rooms can be removed to show what is behind them.

How to ask for it

Create a small voxel-style layout with removable sections and a legend for what each color represents.

Know the boundary. Block detail and large creator-reported counts do not establish a useful simulation or accurate scale.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Build a world out of tiny blocks

  1. 01Small volume cells
  2. 02Assemble rooms and structures
  3. 03Remove cells to reveal inside
The kingdom post explicitly describes voxels and destruction; its numerical counts were not independently validated.

FROM THE ARCHIVE

A voxel kingdom under siege

Creator caption

…2.4 million voxels. 1,500 soldiers. 160 working NPCs. one prompt. Fable 5.1 one shotted the largest playable medieva…
Attribution source ↗
1 linked post · view evidence
  • A voxel kingdom under siege

    Creator caption: 2.4 million voxels. 1,500 soldiers. 160 working NPCs. one prompt. Fable 5.1 one shotted the largest playable medieva

10Meshes, UV maps and rigsGive a model a skin and a way to move

3D asset techniques · Named in the archive

A mesh defines a surface; UV coordinates map flat artwork onto it; a rig provides controls for posing it. Retopology reorganizes the mesh, and shape keys store alternate shapes.

Think of it as a puppet: the shell is the mesh, the painted covering is the texture, and the controls are the rig.

What you could make

Create a reusable presenter character with a wave, a pointing pose and a smile.

How to ask for it

Deliver the mesh, UV layout, textures and rig separately. Test the shoulder bend, hand pose and facial expressions before final rendering.

Know the boundary. A static good-looking character may bend badly. Borrowed motion and generated geometry are separate contributions.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Give a model a skin and a way to move

  1. 01Mesh: shape
  2. 02UV + texture: surface
  3. 03Rig + shape keys: pose
The archive’s character workflows mention these stages. Each stage needs its own check.

FROM THE ARCHIVE

A beginner refines a character by showing the AI what to fix

Retained tool attribution

…and Astra with Blender MCP for assembly, rigging and corrections. Motion is explicitly borrowed. The main post gives a 12-step workflow. The charact…
Attribution source ↗
12 linked posts · view evidence
11Procedural geometry / OrbitControlsMake a repeatable shape, then inspect it

Geometry method and camera control · Named in the archive

Procedural geometry builds shapes from rules and parameters. Three.js OrbitControls lets a viewer orbit, zoom and pan around a target.

Think of it as a recipe for a tiered cake, plus a turntable for inspecting it.

What you could make

Change the number of floors in a schematic building, then rotate it to inspect its massing.

How to ask for it

Create procedural tiers with a floor-count parameter. Add OrbitControls and an exploded view so I can inspect the structure.

Know the boundary. Rules can create consistent repetition without proving architectural accuracy. OrbitControls moves the view; it is not a geometry generator.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Make a repeatable shape, then inspect it

  1. 01Choose tier count
  2. 02Rules generate the repeated shape
  3. 03Orbit to inspect
The temple’s reviewed prompt names procedural geometry, OrbitControls and separation/reassembly.

FROM THE ARCHIVE

Temple of Heaven architectural explorer

Retained review of creator resource

…prompt specifying TypeScript/Three.js, procedural geometry and textures, architectural elements, lighting, OrbitControls, exploded/reassembled animation, and…
Attribution source ↗
5 linked posts · view evidence

Build an interactive page

Structure, behavior and the tools that prepare a web app.

12TypeScriptCatch some mistakes before the app runs

Programming language · Named in the archive

JavaScript with a type system that checks what kinds of values your code expects. It helps catch mismatches while you build.

Think of it as labels on workshop drawers: bolts go in the bolt drawer, not the paint drawer.

What you could make

A quote calculator whose code expects a numeric hourly rate, so passing the text “forty” is flagged during development.

How to ask for it

Use TypeScript for this calculator. Define the input types, validate real user inputs at runtime, and test a worked example.

Know the boundary. Types disappear when code becomes JavaScript. TypeScript has no distinctive visual style and does not automatically check incoming data at runtime. The temple video cannot demonstrate type safety.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Catch some mistakes before the app runs

price: number
✓ 40 A number× "forty" Text: type mismatch

Catch this mismatch before running the app.

This is an explanatory type-checking example, not a compiler running on this page.

FROM THE ARCHIVE

Temple of Heaven architectural explorer

Retained review of creator resource

…a detailed Chinese prompt specifying TypeScript/Three.js, procedural geometry and textures, architectural elements, lighting, OrbitControls, explod…
Attribution source ↗
1 linked post · view evidence
  • Temple of Heaven architectural explorer

    Retained review of creator resource: a detailed Chinese prompt specifying TypeScript/Three.js, procedural geometry and textures, architectural elements, lighting, OrbitControls, explod

13ReactKeep controls and the display in sync

Interface library · Named in the archive

A JavaScript library that builds interfaces from reusable components and updates them when their state changes.

Think of it as a switchboard: change one switch and the corresponding part of the display updates.

What you could make

An anatomy lesson where selecting “muscles” updates the button state, explanation and visible body layers together.

How to ask for it

Build reusable React controls for selecting a system. Keep the selected label and the visible diagram synchronized.

Know the boundary. React manages the interface. It does not provide a database, payments or a 3D model by itself.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Keep controls and the display in sync

  1. 01Click “muscles”
  2. 02Selection state changes
  3. 03Controls + scene update
The source atlas combines React interface controls with a Three.js scene and existing anatomy assets.

FROM THE ARCHIVE

An anatomy atlas that comes apart

Retained tool attribution

…workflow. The creator shares Human Atlas’s React/Three.js application and demo; the interface and repository credit BodyParts3D.…
Attribution source ↗
1 linked post · view evidence
  • An anatomy atlas that comes apart

    Retained tool attribution: workflow. The creator shares Human Atlas’s React/Three.js application and demo; the interface and repository credit BodyParts3D.

14HTMLGive the page meaningful parts

Page structure · Named in the archive

HTML marks the structure of a webpage: headings, paragraphs, links, forms and buttons.

Think of it as a labeled outline before a document is designed.

What you could make

A workshop invitation with a real heading, date, agenda and registration button that can be read by assistive technology.

How to ask for it

Create one HTML page with a meaningful heading structure, a labeled form and a clearly named registration button.

Know the boundary. A single HTML file can also contain styles and scripts. “Single file” does not mean no JavaScript or no external dependencies.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Give the page meaningful parts

  1. 01<h1> Workshop title
  2. 02<p> Date and agenda
  3. 03<button> Register
HTML identifies the parts; CSS can style them and JavaScript can add custom behavior.

FROM THE ARCHIVE

A gallery of coded 3D interfaces

Creator caption

…prompts I used: "Recreate this in a single HTML file with three.js. Keep comparing it against the reference and fix any differences." "Add Liquid…
Attribution source ↗
1 linked post · view evidence
  • A gallery of coded 3D interfaces

    Creator caption: prompts I used: "Recreate this in a single HTML file with three.js. Keep comparing it against the reference and fix any differences." "Add Liquid

15CSSMake the same content fit different screens

Background · styling language · Background concept

CSS describes layout and appearance: type, color, spacing, responsive arrangement and visual transitions.

Think of it as the room arrangement and wardrobe for the structure HTML supplies.

What you could make

Show three workshop cards in a row on a laptop, then one per row on a phone without changing their content.

How to ask for it

Keep the content the same. Use CSS to make three columns on wide screens and one readable column on narrow screens.

Know the boundary. Styling does not create your data or business rules. CSS was not established by name in this archive’s creator evidence.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Make the same content fit different screens

  1. 01Same three workshop cards
  2. 02Wide screen: three columns
  3. 03Phone: one column
A layout rule changes arrangement while preserving the information.
16JavaScriptMake a page respond to an action

Background · programming language · Background concept

A programming language that can calculate results, react to clicks and change what a webpage shows.

Think of it as the moving mechanism behind a control panel.

What you could make

A workshop calculator where 20 seats × $40 immediately displays $800 of potential revenue.

How to ask for it

Make a calculator that updates revenue when I change seats or price. Label assumptions and handle empty or invalid numbers.

Know the boundary. A calculation in a webpage does not automatically save across devices or charge a customer. The term was not established in archived creator evidence.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Make a page respond to an action

  1. 0120 seats × $40
  2. 02Calculate on input
  3. 03Display $800
Illustrative calculation. Revenue is not profit; expenses would need their own inputs.
17ViteSee code changes while you build

Development and build tool · Named in the archive

Vite serves a web project during development and prepares its code and assets for release.

Think of it as the workshop where you preview a prototype before packing it for delivery.

What you could make

While building a 3D playroom, change a wall color and inspect the result in a local preview.

How to ask for it

Use the project’s existing setup. If starting a Vite project, give me a working preview and a production build I can check.

Know the boundary. Vite is neither the 3D engine nor the public host. It has no signature visual effect in the finished room.

Official reference
EXPLAINED VISUALLY See Then Build illustration

See code changes while you build

  1. 01Edit the room code
  2. 02Preview the change
  3. 03Build files for release
The playroom’s retained creator prompt names Vite as setup infrastructure.

FROM THE ARCHIVE

A miniature playroom you can explore

Retained tool attribution

…workflow. Astra, a detailed Three.js/Vite prompt, a modeled playroom, characters, toys and an orbitable camera.…
Attribution source ↗
1 linked post · view evidence

Explain with moving images

Assemble scenes, animate changes and edit a finished video.

18RemotionReuse one video structure with new content

Video from code · Named in the archive

Remotion creates video using React components that describe what appears at each frame.

Think of it as a video assembly line: the structure stays the same while the title, figures and images change.

What you could make

Produce the same 30-second monthly update with this month’s three metrics and a new voiceover.

How to ask for it

Use Remotion for a three-scene update with editable text and data. Render a short sample and check text timing before the full export.

Know the boundary. It assembles specified scenes and assets; it does not inherently generate footage or verify the narration.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Reuse one video structure with new content

01Opening: question
02Middle: labeled explanation
03Ending: recall prompt
The T-cell video creator names Remotion, Imagegen and HeyGen as different parts of the workflow.

FROM THE ARCHIVE

A staged educational video about T cells

Creator caption

…studying! I also had Astra to use the Remotion plugin for the video generation. Astra also used Imagegen to create the visuals, it also suggested…
Attribution source ↗
1 linked post · view evidence
  • A staged educational video about T cells

    Creator caption: studying! I also had Astra to use the Remotion plugin for the video generation. Astra also used Imagegen to create the visuals, it also suggested

19After EffectsAnimate editable layers and combine footage

Motion design and compositing application · Named in the archive

Adobe After Effects combines layers and controls how their appearance changes over time. It is used for motion graphics and visual effects.

Think of it as a stack of transparent sheets on an animation stand, each with its own timing.

What you could make

Animate a product explanation while keeping its titles, arrows and diagram pieces editable for later revisions.

How to ask for it

Create separate editable layers for labels, arrows and shapes. Use named controls so I can change wording and timing.

Know the boundary. A flattened video does not prove the source layers remain editable. Check the project as well as the export.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Animate editable layers and combine footage

01Text + diagram layers
02Keyframes or expressions
03Editable animated explanation
The whiteboard demo names layers, text, cursor and camera animation.

FROM THE ARCHIVE

An editable whiteboard animation for a product walkthrough

Creator caption

…matters. Getting this result manually in After Effects would’ve taken me 100 years. With Higgsfield AI Motion Designer, I give GPT-6 Astra a reference an…
Attribution source ↗
3 linked posts · view evidence
20Cavalry / effectorsChange a rule and move many elements

Procedural motion design · Named in the archive

Cavalry is a motion design application. Effectors can apply coordinated changes to many elements using rules.

Think of it as a conductor moving a whole section of an orchestra together.

What you could make

Make 40 words gather into a title, then change the timing from one control instead of editing each word.

How to ask for it

Use an effector to stagger the words. Keep one control for spacing and one for timing, and show the change on a short preview.

Know the boundary. The source creator describes a shift from individual keyframes to effectors; that does not establish every project is easier this way.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Change a rule and move many elements

01A set of words
02One timing / movement rule
03Coordinated motion
The archive’s typography example names the application and an effector-based approach.

FROM THE ARCHIVE

Astra-assisted typography animation in Cavalry

Retained tool attribution

…Chinese caption explicitly attributes the Cavalry recreation to GPT6 Astra and credits @fafa818969 for the reference animation. The caption reports t…
Attribution source ↗
1 linked post · view evidence
21Premiere, DaVinci Resolve and CapCutChoose the shots and shape the final story

Video editing applications · Named in the archive

Premiere is an editing workspace used here with a custom archive panel. DaVinci Resolve combines editing, color and audio work. CapCut offers clip editing, captions and effects. Each assembles media on a timeline.

Think of it as an editing table: choose what stays, cut what drags and make the pieces tell one story.

What you could make

Turn a folder of customer interviews into a 90-second case study with readable captions and a clear before/after.

How to ask for it

Build a timeline from these clips. Keep source references, editable captions and separate audio tracks, then export a review copy.

Know the boundary. An editor is not the same thing as a video-generation model. The fantasy-film creator reports selecting 127 clips from 847 takes; those figures are attributed.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Choose the shots and shape the final story

01Premiere: select takes
02Resolve: edit + grade
03CapCut: assemble + caption
These are three alternative editing workspaces, not three mandatory stages. Source posts name each separately.

FROM THE ARCHIVE

A 22-minute fantasy film assembled from generated takes

Creator caption

…writes the prompts to the house rules. A Premiere Pro (@AdobeVideo) panel that browses the archive, drops selects on the timeline, versions the proje…
Attribution source ↗
5 linked posts · view evidence
22Marvelous DesignerTurn flat pattern pieces into clothing

Garment design and cloth simulation · Named in the archive

A garment application that joins sewing-pattern pieces and simulates how cloth drapes around a body.

Think of it as a virtual sewing room where flat fabric pieces become a dressed form.

What you could make

Preview a uniform’s silhouette and movement before commissioning a polished promotional render.

How to ask for it

Create the garment from separate pattern pieces, simulate a simple pose, inspect intersections, then export it for the final scene.

Know the boundary. A simulated garment is not proof of manufacturability, sizing or real fabric behavior.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Turn flat pattern pieces into clothing

  1. 01Flat sewing patterns
  2. 02Join seams + simulate cloth
  3. 03Garment on a 3D body
The dress post describes pattern construction, cloth simulation and a Blender export.

FROM THE ARCHIVE

Turn a garment concept into a draped 3D preview

Creator caption

…sewing patterns and build the dress in Marvelous Designer &gt; Simulate the cloth You get an animated 3D garment exported to Blender. https://t.co/fkJKMjg5I…
Attribution source ↗
1 linked post · view evidence
23Camera solving / compositingMake a 3D object belong in real footage

Visual-effects techniques · Named in the archive

Camera solving estimates the camera motion from a shot. Compositing combines the rendered object and original footage into one image.

Think of it as matching a new actor’s stage position and lighting before inserting them into an existing scene.

What you could make

Place a proposed kiosk in a moving street video to communicate its size and location.

How to ask for it

Match the camera, lighting, grain and blur. Check the object’s contact with the ground throughout the shot.

Know the boundary. A convincing composite is constructed imagery, not evidence the object was physically present.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Make a 3D object belong in real footage

  1. 01Solve camera movement
  2. 02Render the added object
  3. 03Combine light, grain and blur
The street-robot post explicitly describes camera solving and compositing.

FROM THE ARCHIVE

Place a 3D robot into real street footage

Creator caption

…pipeline it handled through Higgsfield: >Solve the camera. >Light and render the robot. >Match the CG to the footage, including blur and grain. >Finish th…
Attribution source ↗
1 linked post · view evidence
  • Place a 3D robot into real street footage

    Creator caption: pipeline it handled through Higgsfield: >Solve the camera. >Light and render the robot. >Match the CG to the footage, including blur and grain. >Finish th

Connect the working parts

Let software use tools, save information and run online.

24Node.jsRun the working logic behind an app

JavaScript runtime · Named in the archive

Node.js executes JavaScript outside a web browser, for servers and automation scripts.

Think of it as the back office behind the public-facing counter.

What you could make

Run a portal that stores the creative brief, references, prompts and selected takes for a film project.

How to ask for it

Create a Node service for this project archive with validated inputs and clear records of each take.

Know the boundary. Node is a runtime, not another language or a hosting company. The source film does not show its portal backend working.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Run the working logic behind an app

  1. 01Browser sends a request
  2. 02Node runs application logic
  3. 03Return the requested records
The film creator names a Node portal holding the story bible, references, prompts and takes.

FROM THE ARCHIVE

A 22-minute fantasy film assembled from generated takes

Creator caption

…Pixar film. The tools I built: a portal (Node on @flydotio) that holds the bible, the refs, every prompt and every take, with a Claude agent (@claud…
Attribution source ↗
1 linked post · view evidence
25Fly.ioKeep your application available online

Application hosting · Named in the archive

A hosting platform that runs application processes in virtual machines called Fly Machines.

Think of it as renting a powered workshop where your software can keep running.

What you could make

Host the small project portal your remote team needs to reach from their own computers.

How to ask for it

Deploy this existing application with the required configuration and verify its public route and a read-only request.

Know the boundary. Hosting supplies a place to run code. It does not create the app, guarantee its reliability or make its records secure by itself.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Keep your application available online

  1. 01Application code
  2. 02Running Fly Machine
  3. 03Reachable online service
The creator attributes the film portal to Node on Fly.io; this is infrastructure, not video generation.

FROM THE ARCHIVE

A 22-minute fantasy film assembled from generated takes

Creator caption

…tools I built: a portal (Node on @flydotio) that holds the bible, the refs, every prompt and every take, with a Claude agent (@claudeai) that…
Attribution source ↗
1 linked post · view evidence
26SupabaseRemember information across visits and devices

Backend platform · Named in the archive

A platform offering a Postgres database, authentication, storage and related backend services.

Think of it as the filing cabinet and access desk behind an app.

What you could make

Remember a learner’s chapter progress so they can resume on a different device.

How to ask for it

Store reading progress per account and test that one learner cannot read or overwrite another learner’s records.

Know the boundary. Tables and access policies still need careful design. The source video does not independently verify private data persistence.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Remember information across visits and devices

  1. 01Learner saves progress
  2. 02Account-scoped database row
  3. 03Resume on another device
StoryComet’s creator names Supabase and accounts as part of the build.

FROM THE ARCHIVE

A multilingual library of pop-up stories

Creator caption

…and souvenirs - Day 3: polish, domain name, supabase, accounts, Stripe and optimization The first three books are 100% free so you can try the whole ex…
Attribution source ↗
1 linked post · view evidence
  • A multilingual library of pop-up stories

    Creator caption: and souvenirs - Day 3: polish, domain name, supabase, accounts, Stripe and optimization The first three books are 100% free so you can try the whole ex

27StripeConnect membership access to a payment

Payments platform · Named in the archive

Stripe supplies payment interfaces and APIs for accepting payments and managing related billing operations.

Think of it as the checkout counter attached to a membership library.

What you could make

Sell access to a course while keeping a free sample lesson available to everyone.

How to ask for it

Build a test-mode checkout and connect confirmed payment events to access. Show success, cancellation and failure paths.

Know the boundary. A payment-looking screen is not evidence that charging, refunds or access changes work. The source names Stripe but checkout was not tested.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Connect membership access to a payment

  1. 01Choose membership
  2. 02Payment checkout
  3. 03Confirmed result updates access
The diagram explains an integration flow; it does not make a payment.

FROM THE ARCHIVE

A multilingual library of pop-up stories

Creator caption

…3: polish, domain name, supabase, accounts, Stripe and optimization The first three books are 100% free so you can try the whole experience. Members…
Attribution source ↗
1 linked post · view evidence
  • A multilingual library of pop-up stories

    Creator caption: 3: polish, domain name, supabase, accounts, Stripe and optimization The first three books are 100% free so you can try the whole experience. Members

28MCPGive an AI a defined connection to a tool

Model Context Protocol · Named in the archive

A shared protocol for AI applications to discover and use tools, resources and prompts supplied by a compatible server.

Think of it as a standard socket: the adapter makes the connection understandable, while permissions determine what can pass through.

What you could make

Let an assistant ask a video editor to inspect a timeline and perform supported edits.

How to ask for it

Use the connected editor’s MCP tools to inspect the project, propose an edit and verify the resulting timeline.

Know the boundary. MCP does not grant universal access or replace permissions. Nine archive mentions include six affirmative workflow attributions; the other three describe nonuse, advice to avoid MCP, or an alternative access method.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Give an AI a defined connection to a tool

  1. 01AI application
  2. 02MCP server + allowed tools
  3. 03Editor performs a supported action
An API is a software interface; an MCP server may present that API to an AI in a shared tool format.

FROM THE ARCHIVE

A long-delayed train trip becomes a nineteen-minute travel film

Creator caption

…年。今天终于剪辑成了一支近 20 分钟的 4K 旅行电影。GPT-6 Astra 通过 MCP 操作达芬奇，从整理素材一直做到成片导出，只花了大概半小时。 它做了这些事： - 从 479 段、约 90 GB、超过 4 小时的原始视频中整理时间线、挑选镜头，剪出有故事的旅程。 - 清理了我口头…
Attribution source ↗
9 linked posts · view evidence
29CLI / computer useChoose commands or visible controls

Two ways to operate software · Named in the archive

A CLI accepts text commands. Computer use operates an application through its visible interface, such as clicking menus or reading the screen.

Think of it as giving a precise written order versus using the same controls a person sees.

What you could make

Batch-export 20 assets through commands, or operate a design app whose needed action is available only in its interface.

How to ask for it

Use a documented command if available; otherwise use visible controls. Check the exported files rather than assuming a click succeeded.

Know the boundary. Neither method guarantees success. The orchestration interface in the archive was a demo, and the game recording does not prove uninterrupted autonomy.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Choose commands or visible controls

  1. 01CLI: command → operation
  2. 02Computer use: screen → click
  3. 03Both: inspect the result
The ad-workflow post names a CLI; the Minecraft and Blender examples describe computer use.

FROM THE ARCHIVE

A visual plan for coordinating a long-form ad workflow

Creator caption

…to sit under the whole thing - one CLI controls what generates, in what order, and with which model - delivery comes out with cinematic mo…
Attribution source ↗
4 linked posts · view evidence
  • A visual plan for coordinating a long-form ad workflow

    Creator caption: to sit under the whole thing - one CLI controls what generates, in what order, and with which model - delivery comes out with cinematic mo

  • A bat modeled and rendered in Blender

    Retained review of creator replies: prior games and was not added. Short CLI recommendations are not a replication guide. Creator reply 2096255851266748702 states computer use

  • A Minecraft mining run ending with a diamond

    Creator caption: Astra to mine a diamond in Minecraft using computer use, then went to sleep. Woke up to a diamond in its inventory 🤯🤯🤯 Setup in the replies. https://t.co

  • How a Tripo character is prepared for Blender

    Creator caption: MCPをなるべく使わせない。9割9分Computer useでやらせる。←マジのガチで一番大事 ・絶対にTripo使う。Astraだけでやらせるとネコアルクが量産される。いかなる場合もTripoでモデル生成→Astraでセットアップの流れ。Astraにゼロか

30ComfyUI / 3D Gen StudioConnect a media pipeline as visible steps

Visual workflow software · Named in the archive

ComfyUI connects operations and models as nodes. The archive’s 3D Gen Studio example is a separate local 3D workflow application whose tutorial includes ComfyUI setup.

Think of it as a recipe laid out as connected stations, with each station taking an input and passing on a result.

What you could make

Start with a reference image, generate a rough 3D head, then open the mesh for inspection and cleanup.

How to ask for it

Show each pipeline stage and save its output. Stop for review after the mesh is produced before attempting rigging.

Know the boundary. The main demo shows a mesh workflow; its auto-rigging stage only reached a starting state. Do not describe that stage as a completed result.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Connect a media pipeline as visible steps

  1. 01Reference image
  2. 02Model node produces geometry
  3. 03Inspect and edit the mesh
Workflow software coordinates models; it is not itself a single generative model.

FROM THE ARCHIVE

Move from a reference image to an inspectable 3D asset

Creator caption

…3D Gen Studio is the best free open-source desktop application for local 3D AI workflows. You can generate, ret…
Attribution source ↗
1 linked post · view evidence
31App Intents / model delegationExpose app actions to system experiences

Apple integration framework; experimental source claim · Named in the archive

App Intents lets an app describe actions and data for experiences such as Siri and Shortcuts. The archived post also reports an experimental model-delegation integration.

Think of it as publishing a menu of actions the system is allowed to request from your app.

What you could make

Make a supported “start my timer” action available through a system shortcut. The source explores asking another model to produce a file.

How to ask for it

For a normal app, use documented App Intents actions. Treat the source’s model-delegation experiment as a research example.

Know the boundary. The creator says model delegation requires a private entitlement. Public documentation for general access was not established; this is not a reproducible setup guide.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Expose app actions to system experiences

  1. 01App declares an action
  2. 02System requests a supported action
  3. 03App returns its result
Public App Intents and the creator-reported private model-delegation mechanism have different availability.

FROM THE ARCHIVE

Bring Claude into the interface where a task already starts

Creator caption

…running as a Siri Extension using the Model Delegation API in App Intents. Claude can appear in Siri’s “Ask…” menu and supports the same underlying flow a…
Attribution source ↗
1 linked post · view evidence
32Bend2 / MetalExpress a computation and run work in parallel

Programming language and GPU API · Named in the archive

Bend is a language exploring parallel execution and formal properties of code. Metal is Apple’s API for graphics and parallel computation on its GPUs.

Think of it as a work plan divided among many workers, plus the connection to the machinery they use.

What you could make

Build a small interactive simulation or graphics experiment and measure it on the intended device.

How to ask for it

Use the project’s supported language and GPU path. Profile a small example before claiming a performance advantage.

Know the boundary. The boss-battle creator reports Bend2 and Metal. On-screen FPS is not an independent benchmark, and formal checks cover only specified properties.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Express a computation and run work in parallel

  1. 01Bend2: express the work
  2. 02Metal: access Apple GPU
  3. 03Measure the actual output
These are distinct layers in the creator-attributed implementation. Bend’s public availability descriptions were inconsistent at review.

FROM THE ARCHIVE

A custom 3D boss battle

Creator caption

…Astra built this from scratch on Bend2 and tbh I'm kinda speechless? for a context, Bend2 has no game engine, no 2d graphics, let alone 3…
Attribution source ↗
1 linked post · view evidence
  • A custom 3D boss battle

    Creator caption: Astra built this from scratch on Bend2 and tbh I'm kinda speechless? for a context, Bend2 has no game engine, no 2d graphics, let alone 3

  • A custom 3D boss battle

    Creator-reply attribution; supports Metal, not an independent benchmark: typed functions running on Metal

33Machine code / reverse engineeringUnderstand an existing program from its behavior and instructions

Executable instructions and inspection · Named in the archive

Machine code is the low-level instruction representation a processor executes. Reverse engineering studies an existing program to infer how it works.

Think of it as examining a running mechanism when its original design notes are missing.

What you could make

Inspect a permitted old program to understand a file format or document how a behavior works.

How to ask for it

Explain the observed behavior and show evidence for each inferred mechanism. Keep experiments within software I am authorized to inspect.

Know the boundary. The game demo shows a running result, not every claimed modification. A visible behavior does not establish the exact underlying patch.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Understand an existing program from its behavior and instructions

  1. 01Observe behavior
  2. 02Inspect executable instructions
  3. 03Test a bounded explanation
The SuperAstra caption explicitly describes raw machine code and reverse engineering classic games.

FROM THE ARCHIVE

Prompt-driven edits inside classic SNES games

Creator caption

…favorite ways to test new models has been reverse engineering classic games. This is hard because it's raw machine code. Astra is way better than anything I've s…
Attribution source ↗
1 linked post · view evidence

Read and model the world

Find objects, track motion and estimate spatial relationships.

34MuJoCoRehearse a mechanism in a specified model

Physics simulation engine · Named in the archive

A physics engine that models bodies, joints, contacts and forces, often used for robotics.

Think of it as a rehearsal room with rules: the rehearsal is only as faithful as the model and parameters.

What you could make

Compare how a modeled gripper holds an object under two friction assumptions before planning a real experiment.

How to ask for it

State the modeled forces and simplifications. Compare the simulation with observed motion and label illustrative components.

Know the boundary. The source explicitly uses simplified mechanics and illustrative cable deformation. It does not validate full tendon forces or real robot behavior.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Rehearse a mechanism in a specified model

  1. 01Bodies + joints + parameters
  2. 02Physics engine computes motion
  3. 03Compare with a real reference
A close visual match is different from a validated physical model.

FROM THE ARCHIVE

A robotic-hand reference becomes a simplified motion reconstruction

Creator caption

…from scratch and recreated the motion in MuJoCo, with simplified mechanics and illustrative cable deformation next step would be modeling the full…
Attribution source ↗
1 linked post · view evidence
35RF-DETRFind objects within an image

Object-detection model · Named in the archive

A family of vision models that can identify and locate object categories within images.

Think of it as an inventory assistant placing a labeled box around each thing it recognizes.

What you could make

Find products on a shelf or players in a sports frame, then inspect misses and false detections.

How to ask for it

Detect the requested objects and return labeled bounding boxes with confidence values. Review a representative sample manually.

Know the boundary. Detection locates objects in a frame. It does not by itself track an identity or establish perfect accuracy.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Find objects within an image

  1. 01Video frame
  2. 02Detect player and ball regions
  3. 03Boxes + labels + confidence
The basketball workflow references RF-DETR. This diagram is explanatory, not model output.

FROM THE ARCHIVE

Basketball footage becomes a tracked court map

Retained tool attribution

…and synchronizing those IDs with RF-DETR/BoT-SORT tracks. Percentages, accuracy and savings remain creator-reported.…
Attribution source ↗
1 linked post · view evidence
36BoT-SORTKeep the same identity across video frames

Multi-object tracking · Named in the archive

A tracking method that combines motion and appearance information to associate detections over time.

Think of it as a numbered bib that should stay attached to the same person as they move.

What you could make

Track how one worker moves between stations in an authorized process-study video.

How to ask for it

Track detections with stable IDs and flag gaps, occlusions and possible identity switches for review.

Know the boundary. Similar appearances and occlusion can cause an ID to jump to the wrong person. Detection and tracking need separate checks.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Keep the same identity across video frames

  1. 01Frame 1: person 07
  2. 02Frame 2: match movement + appearance
  3. 03Frame 3: keep ID 07
The basketball post describes RF-DETR detections and BoT-SORT tracks as separate ingredients.

FROM THE ARCHIVE

Basketball footage becomes a tracked court map

Retained tool attribution

…and synchronizing those IDs with RF-DETR/BoT-SORT tracks. Percentages, accuracy and savings remain creator-reported.…
Attribution source ↗
1 linked post · view evidence
37OCRTurn visible writing into usable text

Optical character recognition · Named in the archive

OCR reads characters from an image and returns text that software can search, compare or store.

Think of it as a typist reading from a photograph.

What you could make

Extract a jersey number, receipt total or printed label for a human to verify.

How to ask for it

Extract the visible numbers with their image locations. Preserve uncertain readings instead of silently guessing.

Know the boundary. Blur and stylized lettering can produce errors. Reading a number does not automatically identify the correct person.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Turn visible writing into usable text

  1. 01Image of jersey “23”
  2. 02Recognize the digits
  3. 03Text value: 23
The basketball creator specifically describes OCR for player numbers. “23” here is an authored example.

FROM THE ARCHIVE

Basketball footage becomes a tracked court map

Creator caption

…re-identify players across plays - OCR player numbers - recognize player in possession - detect court keypoints - map player positions and…
Attribution source ↗
1 linked post · view evidence
38HomographyMap one flat view to another

Geometric transformation · Named in the archive

A transformation maps points on a plane between views—for example, from a camera image of a court to an overhead court diagram.

Think of it as stretching a photographed sheet until its corners line up with a flat template.

What you could make

Turn locations on a filmed trade-show floor into a simplified overhead movement map.

How to ask for it

Calibrate known points on the floor plane and map contact positions to the plan. Treat off-plane objects separately.

Know the boundary. One homography does not recover arbitrary 3D structure. An airborne ball can be placed incorrectly on the court map.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Map one flat view to another

  1. 01Camera view: known floor points
  2. 02Compute plane-to-plane transform
  3. 03Overhead plan: matching points
The basketball tutorial names court keypoints and homography; the creator acknowledges placement errors.

FROM THE ARCHIVE

Basketball footage becomes a tracked court map

Retained review of creator resource

…while explaining court keypoints and homography. Observed t.co/qPaLCbkHLD resolves with HTTP200 to the Colab URL; the public source notebook exists…
Attribution source ↗
1 linked post · view evidence
  • Basketball footage becomes a tracked court map

    Retained review of creator resource: while explaining court keypoints and homography. Observed t.co/qPaLCbkHLD resolves with HTTP200 to the Colab URL; the public source notebook exists

39COLMAPWork out camera positions and scene structure

3D reconstruction software · Named in the archive

Software that estimates camera positions and reconstructs 3D structure from overlapping photographs.

Think of it as a surveyor comparing photos taken from different viewpoints.

What you could make

Make a rough spatial reference for a room from a suitable set of overlapping images.

How to ask for it

Use overlapping views with visible texture. Show recovered camera positions and flag missing or unreliable regions.

Know the boundary. Reflective surfaces, low texture and changing scenes can cause trouble. COLMAP is not Gaussian splatting and does not guarantee navigable geometry.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Work out camera positions and scene structure

  1. 01Overlapping photos
  2. 02Estimate cameras + shared points
  3. 03Reconstructed scene structure
The office-drone source reports COLMAP plus DA 3. The flying video does not validate the map.

FROM THE ARCHIVE

An office drone demo exposes the reliability gap

Retained tool attribution

…GPT-6 Astra. Creator replies describe a COLMAP plus DA 3 office-reconstruction pipeline and 2. 8% estimated overall success. The linked primary re…
Attribution source ↗
1 linked post · view evidence
40DA 3 / depth estimationEstimate how near or far surfaces are

Learned spatial estimation · Named in the archive

Depth Anything 3 is a family of models that estimates depth and spatial geometry from images. The retained source attribution abbreviates the ingredient as “DA 3.”

Think of it as adding a near-to-far layer to a photograph.

What you could make

Use a depth estimate to separate a foreground subject from the background for a rough spatial explanation.

How to ask for it

Show the depth estimate beside the original image. State whether scale is relative or metric and inspect obvious edge errors.

Know the boundary. Predicted depth is not a measured distance from a sensor. Outputs and scale depend on the chosen model; the archive does not specify its variant.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Estimate how near or far surfaces are

  1. 01Ordinary image
  2. 02Estimate relative spatial depth
  3. 03Near / middle / far layers
The primary project documents the model family; the archive only retains the shorter DA 3 attribution.

FROM THE ARCHIVE

An office drone demo exposes the reliability gap

Retained tool attribution

…Creator replies describe a COLMAP plus DA 3 office-reconstruction pipeline and 2. 8% estimated overall success. The linked primary report evalu…
Attribution source ↗
1 linked post · view evidence
41CADDescribe an object with dimensions and parts

Computer-aided design · Named in the archive

CAD is a category of software for creating and modifying designs, often with explicit geometric constraints and dimensions.

Think of it as a digital drafting table that remembers measurements.

What you could make

Lay out a proposed controller’s knobs, buttons and enclosure before commissioning detailed engineering.

How to ask for it

Create a dimensioned concept with separately named parts. Label assumptions and identify what needs engineering review.

Know the boundary. The source calls its result a CAD model without establishing a specific package, manufacturing tolerances or production readiness.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Describe an object with dimensions and parts

  1. 01Dimensions + constraints
  2. 02Construct named parts
  3. 03Inspect the design
The DJ-controller post demonstrates a design representation, not a manufactured product.

FROM THE ARCHIVE

A DJ-controller concept explains its own assembly

Creator caption

…parts, read Chinese datasheets, built a CAD model, ordered everything, then made a Blender animation showing how to assemble it.…
Attribution source ↗
1 linked post · view evidence
42Doppler / sonar sensingInfer movement from changes in a sound signal

Sensing principles · Named in the archive

Sonar uses sound to probe a space. Doppler-based methods look at frequency changes related to relative motion.

Think of it as listening for how a returning signal changes as something moves.

What you could make

Prototype a hands-free gesture control in a small, controlled setup, then compare detected motion with a visible reference.

How to ask for it

Explain the proposed sensing path and test it against deliberate motion and no-motion controls.

Know the boundary. The hand-scrolling video does not establish that the claimed mechanism caused the scrolling or that it works across devices.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Infer movement from changes in a sound signal

  1. 01Emit a sound signal
  2. 02Motion changes the received signal
  3. 03Estimate movement
This illustrates the claimed sensing idea, not a verified reconstruction of the demonstration.

FROM THE ARCHIVE

Hand gestures accompany hands-free laptop scrolling

Creator caption

…Astra made me a sonar app that emits undetectable audio to scroll up/down on your computer. It uses the doppler effect t…
Attribution source ↗
1 linked post · view evidence

Hand off an editable result

Choose files another person or application can actually use.

43JSON / Excalidraw filesKeep the parts editable instead of flattening them

Structured data and application format · Named in the archive

JSON stores structured values as text. An Excalidraw file uses a specific JSON structure to describe editable drawing elements.

Think of it as a packing list with instructions about each part, rather than a photograph of the packed box.

What you could make

Deliver twelve icons as separate editable shapes that a colleague can resize, recolor or rearrange.

How to ask for it

Return a valid Excalidraw file with separately editable elements. Open it in Excalidraw and verify the shapes before calling it finished.

Know the boundary. Generic valid JSON is not automatically an editable drawing. The application’s schema gives its fields meaning.

Official reference Tool Library
EXPLAINED VISUALLY See Then Build illustration

Keep the parts editable instead of flattening them

  1. 01Named fields + lists
  2. 02Excalidraw’s drawing schema
  3. 03Editable icons and shapes
The icon post names Excalidraw’s open JSON format; another creator shares prompts as JSON.

FROM THE ARCHIVE

A prompt becomes a set of twelve custom icons

Retained tool attribution

…explains that Excalidraw uses an open JSON format. The finished grid and download confirmation are visible; the exported file was not reopened…
Attribution source ↗
2 linked posts · view evidence
44CSVMove rows and columns between tools

Tabular text format · Named in the archive

A plain-text format for tabular records, with fields separated by delimiters and rows separated by line breaks.

Think of it as a spreadsheet reduced to its rows of values.

What you could make

Export a reading list with title, author and source URL so it can be opened in a spreadsheet.

How to ask for it

Return a CSV with these exact columns. Quote fields correctly and check that commas inside a title do not create extra columns.

Know the boundary. CSV does not preserve rich spreadsheet styling, multiple sheets or a typed data model. The source’s export is a creator claim.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Move rows and columns between tools

  1. 01title,author
  2. 02"Sample, with comma",Alex
  3. 03Open as two columns
The quoted comma belongs inside one field. The Siri post discusses producing CSV through a delegated model.

FROM THE ARCHIVE

Bring Claude into the interface where a task already starts

Creator caption

…Siri AI. For example, Siri can’t create a CSV, but Claude can easily return one when asked. When system interaction is required, such as settin…
Attribution source ↗
1 linked post · view evidence
45VRMCarry an avatar into compatible applications

Humanoid-avatar format · Named in the archive

A file format for exchanging humanoid 3D avatars and associated standardized information between compatible tools.

Think of it as a labeled travel case for a digital character.

What you could make

Move a company mascot into a supported presentation or avatar app while retaining its model and supported expressions.

How to ask for it

Export VRM and test it in the intended destination. Check pose, materials and supported expressions after import.

Know the boundary. The format does not supply speech, intelligence or motion capture. The archived rough-avatar result did not establish successful interoperable export.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Carry an avatar into compatible applications

  1. 01Avatar model
  2. 02VRM export
  3. 03Check in the destination app
The creator names VRM and expresses disappointment with the rough result; it is a useful boundary example.

FROM THE ARCHIVE

A rough VRM avatar prototype

Retained tool attribution

…working expressions, animation quality or VRM interoperability. Avoid reproducing incidental local paths from the screen.…
Attribution source ↗
1 linked post · view evidence
  • A rough VRM avatar prototype

    Retained tool attribution: working expressions, animation quality or VRM interoperability. Avoid reproducing incidental local paths from the screen.

46A .cube LUTReuse a color treatment across clips

Color lookup table · Named in the archive

A .cube LUT describes a mapping from input colors to output colors for a compatible image or video editor.

Think of it as a color translation table: this blue becomes that blue, consistently.

What you could make

Apply an approved color treatment to a series of interview clips, then adjust exposure for each shot.

How to ask for it

Create a .cube LUT for this reference treatment and show before/after samples without clipping faces or losing shadow detail.

Know the boundary. A LUT does not contain footage or perform every kind of image adjustment. A grading claim still needs visual inspection.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Reuse a color treatment across clips

  1. 01Original clip colors
  2. 02Apply color lookup table
  3. 03Consistent color treatment
The CapCut video shows paired room treatments and a LUT import panel; its generated file was not inspected.

FROM THE ARCHIVE

Describe a color treatment, then import it into your editor

Creator caption

…PC: describe a look, generate a custom .cube LUT or animated .srt subtitles, then fine-tune inside CapCut. Astra creates, CapCut PC is where the…
Attribution source ↗
1 linked post · view evidence
47SRT subtitlesKeep the words aligned to the video

Timed text format · Named in the archive

An SRT file pairs subtitle text with start and end times so a player or editor can display it with a video.

Think of it as cue cards with a clock printed on each one.

What you could make

Create a caption file a colleague can correct without re-editing the underlying footage.

How to ask for it

Deliver a separate SRT. Check the transcript, timestamps, line lengths and reading time against the spoken audio.

Know the boundary. Basic SRT stores timed text, not animated motion. The creator mentions “animated .srt subtitles”; any animation would come from the editor’s treatment.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Keep the words aligned to the video

  1. 0100:00:01,000 → 00:00:03,000
  2. 02“Welcome to the workshop.”
  3. 03Editor decides its appearance
The source’s subtitle claim is separate from the LUT workflow visible in the sampled video.

FROM THE ARCHIVE

Describe a color treatment, then import it into your editor

Creator caption

…generate a custom .cube LUT or animated .srt subtitles, then fine-tune inside CapCut. Astra creates, CapCut PC is where the edit comes together.…
Attribution source ↗
1 linked post · view evidence
48AIFFHand over a recording without confusing it with a model

Audio file format · Named in the archive

An audio container commonly used for sampled, uncompressed sound. It stores a recording, not an interpretation of the scene.

Think of it as an envelope carrying the sound recording.

What you could make

Send a clean voice recording to an editor for a narrated explainer while preserving the original file.

How to ask for it

Inspect the audio format, sample rate and channel count. Keep the original and make a separate edited copy.

Know the boundary. AIFF does not mean AI-generated. A sound recording alone does not guarantee an accurate 3D reconstruction of its source.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Hand over a recording without confusing it with a model

  1. 01Recorded sound
  2. 02AIFF audio file
  3. 03Playback or further processing
The barking-room post uses AIFF as an input and explicitly says the result is not a 3D map.

FROM THE ARCHIVE

An acoustic room hypothesis from a barking recording

Creator caption

…gave GPT-6 Astra .aiff audio file containing a dog barking and asked it to reconstruct the space - well, as best it could…
Attribution source ↗
1 linked post · view evidence
49SVGKeep a diagram sharp and editable

Background · vector graphic format · Background concept

SVG describes two-dimensional graphics using shapes, paths and text rather than only a fixed grid of pixels.

Think of it as a drawing recipe that can be redrawn at a new size.

What you could make

Deliver a process diagram whose labels and connecting lines remain sharp when enlarged for a slide.

How to ask for it

Create an SVG with editable text and simple shapes. Check label wrapping and readability at the intended presentation size.

Know the boundary. Embedding a photograph inside SVG does not turn it into vector artwork. SVG was not established by name in archived creator evidence.

Official reference
EXPLAINED VISUALLY See Then Build illustration

Keep a diagram sharp and editable

  1. 01Shapes + paths + text
  2. 02Scale the vector instructions
  3. 03Sharp diagram at a new size
Explanatory background: a useful output choice for many diagrams, not a claimed stack ingredient in these posts.