Skip to main content

Roadmap

Twelve phases, built in order on one scene file, so nothing made early has to be redone later. Phases 1 to 8 are largely built and the Studio is in alpha at kinetic-scenes.unysonplus.com: the editor, export, the Unyson+ element, AI assist, 3D, sound and voice. Phases 9 and 10 are in progress (fluid motion, product tours and interactive scenes), phase 11 makes the AI a real motion director, and phase 12 adds a Video mode.

48 of 95 tasks completeStudio v0.0.23

Phase 1 · FoundationsDone

5/5

The parts everything else stands on: the project, the scene file, the player that plays it, the responsive stage and the route to the live site. At the end of this phase a hand-written scene file plays correctly on the subdomain.

  • Project scaffold — Done

    A GitHub repository with a React + TypeScript editor built by Vite, and a separate plain-JavaScript player package. Both build to static files.

  • Scene file, version 1 — Done

    The JSON format with tracks, clips, preset settings, keyframes, assets and breakpoints, plus a schema that validates it and a version number so later formats can upgrade older files.

  • Player runtime — Done

    Reads a scene file and builds one GSAP timeline. Seekable and repeatable: no unseeded randomness and no real-time physics, so any frame can be shown exactly.

  • Responsive stage — Done

    An aspect ratio per breakpoint, the three height modes (ratio, ratio capped to the window, fill the window), sizes in CSS container units, and a safe-area guide.

  • Deploy to the subdomain — Done

    A GitHub Action builds the editor and publishes it to GitHub Pages at kinetic-scenes.unysonplus.com (HTTPS) on every push to main.

Phase 2 · EditorDone

10/10

The working studio: a canvas, a timeline and a first preset library, enough to build a hero animation by hand from start to finish.

  • Canvas and layers — Done

    Text, image and shape layers on the stage; click to select, drag to move with snapping to the centre and the safe area, resize with handles, and double-click to edit text in place.

  • Multiple selection, Align and Distribute — Done

    Select several layers with Shift or Ctrl + click, a dragged box or Ctrl+A; move them together; align their edges or centres to each other, the stage or the safe area; distribute them by edge, centre or equal gaps; and lock layers so they stay put.

  • Groups — Done

    Group layers (Ctrl+G) to manage big scenes: a group has a timeline of its own, opens with a double-click (breadcrumb, everything else dimmed, Escape steps back), moves in time as one by dragging its bar, and takes presets as a whole. Ungrouping keeps every layer where and when it was.

  • Timeline — Done

    Tracks and clips; drag, trim and move clips; a playhead with frame-accurate scrubbing; zoom; play, pause and loop.

  • Inspector — Done

    Edit the selected clip's settings (text, image, colours, duration, easing), with per-breakpoint overrides for position, size and type size.

  • Preset library, first set — Done

    22 presets in a drag-and-drop panel, each with a live preview on hover: text (rise, slam, stretch, scramble, typewriter, blur, flip, tracking), reveal and exit (fade, wipe, iris, slide, grow), move and scale (pop, zoom, spin, keyframes) and emphasis (wave, pulse, shake, float, recolour).

  • Measured eases and better defaults — Done

    Twelve named eases measured from professional motion pieces (Soft land, Snap and settle, Pop over, Leave…), each with what it is for. Exits default to an accelerating curve, staggered text gets a natural uneven rhythm, and the preset defaults follow measured timing.

  • Check scene — Done

    A Check tab that reviews the scene as you work: text that leaves before it can be read, text too small on a screen size, text off the stage or on other text, clips fighting over a layer, long stretches with nothing moving. One-click fixes, Apply all, and Fix with AI.

  • Undo, redo and saving — Done

    Full undo and redo (a drag is one step), the working scene saved in the browser automatically, and project files (.scene.json) to download and reopen.

  • Playback modes — Done

    Play on load, play when scrolled into view, and loop or play once.

Phase 3 · ExportDone

5/5

Get a finished scene out of the Studio and onto a website.

  • Standalone HTML export — Done

    One HTML file that contains the scene and the player and runs anywhere.

  • Embed snippet — Done

    Copy-and-paste embed code from the Export window: the scene inline plus the player, loaded from its permanent versioned address. Works in any page that accepts HTML, such as a Custom HTML block in WordPress.

  • Images and fonts in exports — Done

    Uploaded images travel inside the scene. Web fonts can be embedded in the downloaded page (Latin character sets packed in, others left linked), so it works offline.

  • Player versioning — Done

    Downloaded pages carry their own copy of the player; embeds load a published player version that never changes. The build refuses to publish a changed player under an existing version number.

  • Accessibility — Done

    Respect reduced-motion settings, keep text as real text, and give decorative layers the right hidden roles.

Phase 4 · Unyson+ integrationIn progress

2/3

Put Studio scenes into Unyson+ pages and share one preset library with the Animation Engine.

  • Page-builder element — Done

    The Kinetic Scene element (Animation Engine → Kinetic Scene) plays a scene inside a Unyson+ page with the bundled player: paste the scene file or the embed code, and override the height mode, header height, start and loop for that placement.

  • Open in Studio — Done

    Open in Kinetic Scenes opens the Studio with the element's scene; Save to page sends the edited scene straight back into the element. Each side only accepts messages from the other.

  • Shared preset library — Planned

    Presets usable both in the Studio and in the [Animation Engine](/animation-engine), so an effect is built once.

Phase 5 · AIDone

6/6

Describe an animation in words and get an editable scene, using the AI Dev Kit on the user's own computer.

  • Studio-to-kit connection — Done

    The Studio reaches the AI Dev Kit on the user's computer (localhost:8787). The kit only answers the Studio's own pages. Browsers ask once before a website may reach apps on the device; the Studio explains the prompt and what to do if it was refused.

  • Prompt to scene — Done

    The AI tab turns a request into a scene file built only from the preset catalogue (sent with each request, so the kit never goes stale). The result is validated; an invalid answer gets one automatic repair round. It lands on the timeline as ordinary clips, one undo step.

  • Edit by prompt — Done

    "Change this scene" or "Change the selected layers": the AI returns the whole scene with only the asked-for change, keeping every other layer and id, and re-checks the layout at every screen size so bigger text never overlaps or leaves the safe area.

  • AI assist panel — Done

    A floating ✦ AI assist button over the stage opens a chat with the AI Dev Kit: who is answering, scope chips (this scene, the selected layers, a new scene), starter ideas, and for every answer one undo step, "Go back to before this" and a Check scene count. The Studio sends its motion rules with each request.

  • Custom code clip — Done

    An escape hatch for effects no preset covers: a clip that runs your own GSAP code. It runs only where the page allows it (the Studio, your own exports, and WordPress pages by authors trusted with unfiltered HTML), a failing clip never stops the scene, and the AI never writes one.

  • AI Dev Kit hub — Done

    The kit's dashboard opens on a Home page with a card per job (Site Converter, AI Assistant, Kinetic Scenes): whether each is ready, when it last called the kit and from which site, plus the AI in use. Retitled Unyson+ AI Dev Kit.

Phase 6 · 3D and finer controlDone

4/4

3D objects on the timeline, and hand-keyframed motion for anything the presets don't cover.

  • 3D clips — Done

    3D layers drawn by three.js: extruded 3D text, shapes, extruded SVG logos and glTF/GLB models, with matte to chrome finishes and studio lighting. Spin 3D, Tumble in, Orbit and Camera push presets; every frame drawn from the timeline, so 3D seeks, loops and scroll-links. three.js loads only for scenes with a 3D layer.

  • Keyframe and curve editor — Done

    Keyframes clips with a track per property (move, scale, rotation, opacity, blur), an ease per keyframe and custom curves drawn with two handles (overshoot and anticipation allowed). Add keyframes at the playhead, drag diamonds on the timeline, or drag the layer on the stage to set its position keyframe.

  • Scroll-linked playback — Done

    Starts: As the page scrolls. The scroll position is the playhead, forwards and back. Pin while playing holds the stage below the header for a set number of screens; smoothing glides the playhead. Built without a scroll plugin, in the Studio, the exports and the Kinetic Scene element (with a Scroll length option), plus a Try scrolling test page.

  • More presets — Done

    35 presets in six groups, including Count up, Soft wipe, Blinds, Turn, Sheen, Drift, Gradient flow, Particles and the 3D set (Tumble in, Spin 3D, Orbit, Camera push), plus transitions between groups (Crossfade, Push, Zoom through, Soft wipe, Blinds, Iris, 3D turn) applied from the group Inspector.

Phase 7 · LaterIn progress

3/5

Useful additions once the core works, none of which change the scene file.

  • Video export — Done

    Export → Video renders MP4 (H.264) or WebM (VP9) at Full HD, HD, 4K, vertical or square, 24 to 60 fps. The AI Dev Kit (1.12.34) draws the scene frame by frame in headless Chrome and encodes it with ffmpeg, so every frame is exact, 3D included.

  • Share links — Done

    Export → Share link puts the whole scene, compressed, inside the link: anyone who opens it gets their own copy in the Studio. No account, no server and no storage; custom code clips are removed from shared scenes.

  • Save to Google Drive — Planned

    Projects on any computer without our own server: people sign in with Google and their scenes are saved to their own Drive. Planned if people ask for cross-device projects.

  • Accounts and cloud projects (only if needed) — Planned

    If real accounts are ever needed (a project library, short links, embeds that update in place), they run on a free cloud tier, separate from the website hosting, with uploads on object storage and embeds still served from GitHub Pages.

  • Click-to-advance presentations — Done

    Timeline markers and Playback → Advance: On click. The scene stops at each marker until a click, a tap or the arrow keys, with a pulsing hint; full screen is a presenter mode; without motion it jumps slide to slide.

Phase 8 · Sound and voiceIn progress

5/9

Voice-overs, music and sound effects, and scenes made from a voice-over with AI.

  • Audio track — Done

    Voice-overs, music and sound effects on the timeline with waveforms, trimming, volume and fades, played in step with the animation. Scenes with sound start muted on pages (a Tap for sound button), and video exports mix the sound in.

  • Built-in sound effects — Done

    25 sounds made by the player itself (whoosh, swish, pop, click, typing, impact, thud, riser, chime, sparkle, shimmer and more): nothing to download, identical in the Studio, on pages and in video exports. Give any clip a sound and it moves with the clip; Add suggested sounds places matching ones on a scene's key moments.

  • Sound library in the AI Dev Kit — Planned

    A one-click download of about a thousand free public-domain (CC0) sound effects, searchable from the Studio, plus an optional search of Freesound's public-domain sounds with your own free key.

  • Generate a sound — Planned

    Describe a sound ("a soft paper swoosh, 0.6 s") and the AI Dev Kit makes it: with your own sound-effects API key, or with an open model it downloads and runs on your computer.

  • Styles (scene presets) — Done

    Eight ready looks (Night, Clean, Kinetic, Pastel, Neon, Editorial, Playful, Luxe), each a palette, a font pairing, a motion personality, a pace and a sound palette. Apply one to a scene (colours and fonts, motion and sounds, each optional), start a new scene in it, or give it to AI assist as art direction. Save your own scene as a style.

  • Brand kit — Planned

    Your logo, colours and fonts as a style of your own, edited directly, plus styles exported and imported as files to share with a team.

  • Transcription in the AI Dev Kit — Done

    Select a voice-over and click Transcribe: whisper.cpp runs on your own computer through the AI Dev Kit (downloaded once, then offline; Fast 148 MB or Accurate 488 MB). Every word gets its time and an emphasis score from how loudly, how high and how long it is said, and the delivery is summarised (pace, energy). Words show on the timeline and a click jumps to one.

  • Make a scene from a voice-over — Done

    Transcribe a voice-over, then click Make a scene from this voice-over: the words are grouped into beats (phrases between pauses and sentence ends) and AI assist builds kinetic type timed to them, the key words of each beat entering as they are said, the strongest words landing as hits with a sound, transitions at pauses, and motion matched to the delivery, in the style picked in AI assist. Check scene treats text the voice is saying as read by ear.

  • Studio tools for AI apps (MCP) — Planned

    The AI Dev Kit offers the Studio as tools to AI apps such as Claude Desktop, so a scene can be made or changed from a conversation.

Phase 9 · Fluid motionIn progress

4/17

Scenes that flow like hand-made motion graphics: one element turning into the next, with transitions, a camera, drawn paths, glow and impact.

  • Scene transitions — Done

    Six transitions made for whole groups: Slash (a glowing line sweeps across and reveals the next scene), Flash, Whip pan, Glitch, Dip to colour and Zoom through, each with its sound, alongside Crossfade, Push, Soft wipe, Blinds, Iris and 3D turn. Transition to the next group places the matched pair; the leaving group's own exits give way to it.

  • Whip and light-streak transitions — Planned

    The cut hand-made reels use most: a velocity-matched whip (the old scene accelerates out with heavy directional blur, the new one arrives from the same direction at the same speed and settles), and a light streak fired exactly at the cut to hide the join.

  • Camera — Done

    One camera over every layer that pans, zooms, rolls and tilts in perspective, keyframed on its own Camera row with a curve per move, ready-made moves (push in, pull out, pans, tilt reveal, slow drift, Dutch roll) and a Camera view switch for editing the flat layout. Transitions stay exact under a moving camera.

  • Exact frames when jumping — Done

    Jumping to a moment (scrubbing, seeking, video export) shows exactly what playing to it shows. The player's self-test now checks every frame where any of the 41 presets or a camera key starts or ends, plus the frames either side (231 frames, all identical), and video export takes each frame at an exact frame time, so a cut never lands a frame early or late.

  • Safer video export — Done

    Before the first frame, video export waits for fonts, every image (loaded and decoded), images used as backgrounds and every 3D model, and tells you in the Studio if anything failed to load or ran out of time, instead of silently rendering blank frames.

  • Paths and drawn lines — Planned

    Lines, curves and connectors that draw themselves, layers that travel along a path (a comet that writes a word), and fading trails behind moving things.

  • Word-reveal carrier — Planned

    A text preset where the first word slides in big and fast and each following word travels less (for example 360, 120, 60, 25, 12 px), so a line arrives with momentum instead of in a uniform stagger.

  • Image layers (AI) — Planned

    Split into layers: an image becomes a group of stacked transparent layers (background, subject, objects, text) in exactly the same place, so a picture can be revealed piece by piece instead of all at once, each part animated on its own. With a depth per layer, a camera move turns a flat photo into a 2.5D parallax shot. Runs through the AI Dev Kit: Qwen-Image-Layered through the user's own fal.ai key (or on a local GPU that can hold it), with a lighter local fallback (cutting out the subject and filling in behind it) that runs on an ordinary computer.

  • Glow and motion blur — Planned

    Neon glow on text and shapes, and blur along the direction of fast moves, so quick motion reads as speed rather than a jump.

  • Impact bursts — Planned

    One-off bursts of sparks, shards or rings on a hit, landing with the sound and the word they punctuate.

  • Captions with the spoken word highlighted — Planned

    Captions built from a transcribed voice-over that light up each word as it is said, in the style's colours, alongside or instead of the big kinetic type.

  • Picture leads the voice on punchlines — Planned

    In scenes made from a voice-over, a reveal finishes on its word instead of starting on it, so the picture lands as the word is heard (the viewer sees, then hears). Punchlines lead slightly; transcripts reach the AI as compact phrase lines, leaving more of its budget for the scene.

  • Music beat sync — Planned

    Drop music on the timeline and the Studio finds its tempo, beat grid and strongest moments, right in the browser (no Python): each beat scored by its onset strength, local contrast, loudness and bass. The beats show on the audio row; a big reveal can snap up to 0.15 s onto a strong moment and small entrances up to 0.10 s onto a beat, only a few are locked per scene, and text that must be read lands on every other beat at fast tempos. AI assist gets the same cue list.

  • Music-reactive glow — Planned

    A loudness and bass curve computed once from the music lets a few things breathe with it (a glow, a background's warmth, a card's presence). Precomputed, so it seeks and exports exactly; never equaliser bars or strobing.

  • Sound that sounds designed — Planned

    Fewer effects, each tied to something visible; each sound started early by its own attack so its hit lands on the frame; a softness rating on every built-in sound so repeated ones stay gentle; and video export mixed to a standard loudness (-14 LUFS, peaks under -1 dB) with music lowered under a voice-over.

  • Interface building blocks — Planned

    Ready-made pieces for product videos: a prompt box that types, highlighted tags, a node diagram with connecting lines, timeline and progress bars, stacking list cards.

  • Benchmark against hand-made videos — Planned

    Rebuild reference motion-graphics videos from their voice-overs with the new tools and compare them frame by frame, the way the Site Converter is measured against its sources, until the Studio's version holds up next to them.

Phase 10 · Product tours and interactive scenesIn progress

4/13

Show Unyson+ (or any product) working, from screenshots instead of screen recordings: a cursor that moves, hovers and clicks, zooms into the part that matters, and scenes visitors can click, follow and explore. Links and the looping ending came first (a hero's call to action needs them now); hover effects next, then paths and drawn lines (Phase 9), then the tour tools.

  • Links on layers — Done

    Any text, shape, image or group can be a link (a web address, a page on the same site, an anchor, email or phone; optionally a new tab). On the stage it is a click area; the real link sits in the scene's text for search engines and screen readers, so keyboard users reach it with Tab and the layer shows a focus ring. Other kinds of address are left out.

  • What happens at the end — Done

    Stop on the last frame, loop everything, or loop the last part: after one full play, the scene repeats only from a moment you choose, so a closing call to action keeps breathing (and stays clickable) while the rest holds. Idle presets made for it (an endless glow, a slow breathe) are still to come.

  • Scroll a screenshot — Done

    The Scroll preset scrolls a tall screenshot inside its frame (or pans a wide one across), from one point to another, so a real settings page or site can be shown moving inside a browser window, tilting with the camera.

  • Pin to the stage — Done

    A layer or group can be pinned to the stage instead of the scene's frame: its position becomes a percentage of the stage, so badges, logos and corner calls to action sit in the real corner of any box the scene fills, whatever its shape. Sizes keep the frame's units; pinned layers stay put when the camera moves.

  • Layouts by shape — Planned

    Besides the layouts by width (desktop, tablet, phone), an extra layout for stages much wider than the design (an ultrawide hero), so the content can spread out instead of showing the page at the sides.

  • Hover effects — Planned

    Presets for layers people can point at (lift, grow, glow, brighten, a shine sweeping across, an underline drawing in, a colour change), and custom CSS for anything else where the page's author is trusted. Hovers never fight the animation, and on phones a tap just follows the link.

  • Cursors — Planned

    Built-in cursors drawn big enough to read in a video (about 7% of the frame wide in a full-frame scene), which enter from off-screen and leave the same way, never fade in on the spot, and change from an arrow to a pointing hand over buttons. The cursor carries the viewer's eye from one thing to the next, and its click visibly causes what happens next.

  • Move to a target — Planned

    Pick a layer (the Save button) and the cursor travels to it along a natural curve with real pointer easing, so a tour is built by pointing, not by keyframing.

  • Click effect — Planned

    The cursor presses, a ring ripples out, the button dips and a click sounds, all landing on the same frame.

  • Focus and spotlight — Planned

    Frame a region and the camera zooms to it; a spotlight dims everything else; a callout label points at the thing being explained.

  • Cut a piece out of a screenshot — Planned

    Drag a box over part of a screenshot (a button, a panel, a row) and it becomes its own layer in exactly the same place, ready to press, slide or glow, with the gap behind it filled from its surroundings. Interface screenshots are mostly rectangles, so this needs no AI.

  • Before and after states — Planned

    Swap to a second screenshot of the same screen exactly on the click frame (the menu opened, the setting saved), or type into a field, so actions have visible results.

  • Hotspots — Planned

    Clickable areas that jump to a part of the scene, so a visitor can choose what to see next: an interactive demo instead of a single linear play.

Phase 11 · The AI directorPlanned

0/9

Make the AI Dev Kit the best motion designer of the tools out there, using only the AI subscription you already pay for (Claude, ChatGPT, Cursor): no API keys, no per-use cost. The difference is not a bigger prompt but a way of working: the AI plans with you, builds, checks its work, looks at rendered frames, fixes what it sees and hands you a scene you can still edit, for a website or a video.

  • The Kinetic Scenes skill — Planned

    Everything an AI needs to make a good scene, written for any agent: the workflow, the scene format, every preset and its settings, the motion rules, styles and tones, sound, and a checklist. Generated from the Studio's own code on every release, so it never goes out of date; published at kinetic-scenes.unysonplus.com/skill/ (any AI can read it from a link) and copied into the AI Dev Kit for agents working on your computer.

  • Tools the AI can use — Planned

    The kit gives the AI hands and eyes: inspect (capture a website, read a project, transcribe a voice-over), check (the real validator and Check scene), stills (render frames to look at), beats (analyse music) and render (video export). Offered as kit commands and as a local MCP server, so Claude Code, Codex and Cursor can call them while they work.

  • Plan first — Planned

    For a new scene, the AI first writes a storyboard you can adjust: the hook in the first two seconds, the highlights, the ending, each act's length, what appears one by one, the transitions and the sound. It builds only once you agree, so long scenes come out right and you steer before the work is done.

  • Look, criticise, fix — Planned

    After building, the kit renders the settled frame of every act and the middle of every transition; the AI looks at them with the Check scene report, criticises its own work like a reviewer (cramped layouts, low contrast, muddy transitions, unreadable text) and fixes it, at most a few rounds. Only then does it show you the result.

  • Quick scene from a website or a project — Planned

    Paste a URL (or point at a project folder) and get a scene: the kit's capture service renders the site and takes its real colours, fonts, copy and screenshots; the AI plans and builds from them, shows the product in use rather than describing it, and never invents claims. The result is an editable scene that works as a hero and exports as a video.

  • Any of your subscriptions — Planned

    Your Claude subscription through the claude command (already working), your ChatGPT subscription through OpenAI's codex command, and Cursor through cursor-agent. The kit picks what you have; nothing needs an API key.

  • An AI inside the open Studio — Planned

    The Studio registers its own tools with the browser (WebMCP): look at the scene, inspect a layer, render a frame at any moment, edit text, move layers, add and change clips. An AI in the browser (ChatGPT Desktop already supports it; Chrome is trialling it) can then work on the scene in front of you, with every change in the normal undo history.

  • Better rules for every step — Planned

    The craft the best tools use, built into the rules and checks: tone presets that set structure (number of acts, hold lengths, transition style, sound density); one new thing at a time; no two acts in a row with the same entrance and layout; a pause of 0.3 to 0.75 s before a big payoff; entrances under about 0.8 s; two or three transition styles per scene; no muddy crossfade between two busy layouts.

  • The best still frame, picked for you — Planned

    The Still frame (what visitors who turn off animation see, and a video's thumbnail) is suggested automatically: the most complete settled moment, with every word in and nothing mid-transition.

Phase 12 · Video modePlanned

0/9

A motion-graphics companion for video creators who edit elsewhere: make kinetic type, backgrounds, objects and effects to layer over Reels, TikToks, Shorts and YouTube videos. Web stays the main mode.

  • Web and Video modes — Planned

    Choose what a scene is for when you start it, and switch later. Web (the default) keeps responsive layouts, scroll and click playback, search-engine text and the WordPress element. Video sets one fixed format, plays once at an exact length, puts sound first and shows the tools and checks that suit video.

  • Video formats and platform safe zones — Planned

    9:16 (1080×1920), 1:1, 4:5 and 16:9 at a chosen frame rate, with guides for where TikTok, Reels and Shorts cover the picture with their buttons and captions.

  • Transparent overlay export — Planned

    Export the motion graphics alone, with a transparent background (WebM with alpha, or ProRes 4444), ready to drop on top of footage in CapCut, Premiere or DaVinci Resolve. MP4 for finished pieces, and a cover image.

  • Your footage as a reference or background — Planned

    Bring in a video or photo to design against: line up type and effects with what happens on camera, then export the graphics alone or the full composite.

  • Video checks and AI rules — Planned

    Check scene and AI assist tuned for video: a hook in the first seconds, scene pacing, caption ratios, platform length limits and loudness.

  • One project, several formats — Planned

    Lay a scene out once for 9:16, 1:1 and 16:9 and export them together.

  • Video export without the AI Dev Kit — Planned

    Investigate rendering video in the browser itself (WebCodecs), so video export works on the website with nothing installed.

  • Simple video tools (later) — Planned

    Light trimming and speed changes for footage layers. Editing stays the job of a video editor; the Studio is for what goes on top.

  • Videos that feel like one camera move — Planned

    For videos (not website loops): how a scene leaves sets how the next one arrives (same direction and speed, cutting mid-motion), one main direction for the whole film with other directions saved for meaning, causes and effects on the same frame, and no idle floating to fill time: every second something tells the story. A check measures every cut.

Order of work​

  1. Foundations. The scene file, the player and the responsive stage, because everything else reads or writes them.
  2. Editor. The canvas, the timeline, the inspector and the presets: enough to build a real hero by hand.
  3. Export. Standalone HTML and an embed snippet, each recording the player version it was made with.
  4. Unyson+ integration. The Kinetic Scene element with an Open in Kinetic Scenes round trip.
  5. AI. Prompts produce scene files through the AI Dev Kit, so every result can be edited on the timeline.
  6. 3D and finer control. three.js layers, a keyframe and curve editor, and scroll-linked playback.
  7. Later. Video export, click-to-advance presentations and share links.
  8. Sound and voice. Built-in sound effects, voice-overs with their words, styles, and scenes made from a voice-over.
  9. Fluid motion. Transitions, the camera, paths, beat sync, captions and the motion that hand-made reels use.
  10. Product tours and interactive scenes. Links, hover effects, the looping ending, pinning to the stage, cursors, focus and screenshots that come alive.
  11. The AI director. The AI plans with you, builds, checks, looks at rendered frames and fixes what it sees, with tools from the kit and only the subscription you already have; plus a quick scene from any website.
  12. Video mode. Formats, transparent overlays for video editors, and videos that feel like one camera move.

The plan and the reasons behind it are on the overview; what already exists elsewhere is on the research page.