Holds the object still without losing your Auto-rotate settings, so turning it off resumes exactly where you were.
10,000 rendered icons from thiings.co. Build ramp uses them like a character set — the first icon fills the darkest cells, the last the brightest. Place drops one onto the canvas instead, where it moves, resizes and rotates like any placed image.
Icons by thiings.co, licensed for commercial use.
Fills every character cell with this image instead of a glyph. For an image you can position and drag, use Place image on canvas; for one behind everything, use Background under Colour.
Soothing ambient soundscapes for focus and flow. Audio reactive.
A slow swell to pace your breathing against, on the canvas and in the sound. The music is optional: with the toggle off, a session is the breath alone.
Use your camera to shape the animation to your mouth movement. Works with audio reactive.
Drag to reorder. Top of the list is front of the canvas. Effects and Shape are locked because the mask is a knockout: anything drawn after it would tear its negative space, so they hold their place while everything else moves around them. A layer dragged below Effects gets effected and masked; the same layer above them does not.
Turn words into a moving, responsive layer. Write the copy, arrange it in space, give it a pulse.
Drops a mirror ball over the canvas and throws light around it. ⌘D toggles it from anywhere. It sits over the canvas rather than in it, so it is not in exports.
Mirrors your camera in the ball's tiles. Off, the tiles reflect the canvas instead, so there is always something real in them. Asks for camera permission the first time.
The light brightens and the ball speeds up with whatever is driving the animation, so it works with an MP3, the microphone or the generative music.
Puts a pen in your hand and a toolbar at the bottom of the canvas. The strokes are clean vector lines rather than a sketchy hand-drawn style. ⌘⇧D toggles it. The drawing IS in exports: it sits in the layer stack like any other layer, so you can drag it above or below the effects and the shape mask. The tools load the first time you turn this on.
On, turning the layer off puts the pen away but leaves the drawing on the canvas and in exports. Off, the drawing goes away with the toolbar.
Drops onto the canvas as a free-floating image you can drag to move and resize from the corners. Dragging a file straight onto the canvas does the same thing. This is not the same as Character Set → Image, which fills every character cell, or Background, which sits behind everything.
Post-processes the whole canvas, after the animation and every layer. Included in PNG, video and GIF exports; the SVG export stays vector and is unaffected.
Pay once. Own it forever. No subscription, ever.
Forever opens soon. Leave your email and the founding price is locked for you on day one.
What is working? What is missing? What should DOTCLOUD do next?
Global to the app, not saved with a preset or shared with a guest. Keys are held for this session only and cleared when you reload.
Pull a palette straight out of a Figma file's colour styles. The token is used once, for that request, and is never stored.
Used when you transcribe audio with Whisper, under Layers, Subtitles. Held for this tab only and cleared when you reload. Sent only to OpenAI, only when you start a transcription.
Captures, backgrounds and characters all spend from one meter. Prices are set on the server, so this is what they actually are.
Bring your own key for a provider and its actions stop touching the meter: they bill your account instead. The key travels only to that provider, never into a shared room, and like every key here it is held for this session and cleared when you reload.
Switching pillar reshuffles the panel, starts or stops the music and adds or removes the dancer. A band crossing the canvas hides that work: the switch happens at its midpoint. Turn it off and the switch is exactly what it was before. Full is glimm's own 1100ms and is the default: a sweep you do not notice is not hiding anything. Quick is 560ms, for switching constantly.
Reduced motion switches immediately with no band, which is the honest version of this rather than a shortened one: the transition IS the movement.
Keeps your own browsing out of the analytics, so testing does not move the numbers. The setting is stored in THIS browser for THIS domain, so set it once in each browser you use, and again if the address changes.
DOTCLOUD Explore is a kinetic studio for live visual experiments. Start with a template, move one control, then keep only what makes the piece better. Patterns are built from ASCII characters, emoji, gradients and shaders, and everything exports.
Select a kinetic template from the dropdown at the top of the panel. Each one previews as a live thumbnail rendered with your current colours, so the list shows what you would actually get.
Some templates draw their own picture rather than a character grid: the Dia gradients (Ocean, Sunset, Aurora, Candy, Ember, Mono, 3D Fold), the shader templates, Stop Motion and Add Media. Each has a Remove Characters switch. Leave it on for the picture itself; turn it off and the picture drives character density instead, so the glyphs take its shape.
The Dia gradients follow your cursor and the music. Moving over the canvas slides the tall part of the field and sets how far it has risen. Their own section, directly under Template, holds the rest: Bars, Blur, Peak height, Valley, Fold angle, and Audio drive. With audio running the bars take the spectrum, lows in the middle out to highs at the edges, and the blur tightens as it gets loud. Set Audio drive to 0 for cursor only.
The Shaders category holds eleven, and they split into three kinds. Heatmap, Liquid Metal, Gem Smoke, Metal, Chrome and Glass Object run inside a shape you choose under Logo Animation, so the picture takes the form of a circle, a star, an uploaded SVG or whatever you dropped into Add Media. Mesh Gradient, Neuro Noise, Warp and Grain Gradient fill the whole frame and have no shape. Lines is drawn rather than shaded: ridges across the frame that rise over the shape, and it reads the shape as a landscape rather than as a mask.
Metal, Chrome and Lines answer to your cursor. The highlight on Metal and Chrome sits where the pointer is and slides across the surface as you move; on Lines the pointer pushes a ripple through the ridges near it. All three settle back to the middle when the pointer leaves the canvas, so an untouched piece still looks lit. Every shader also has Audio drive: set it to 0 for cursor and time only.
Those three open on black, the same near-black the app itself is built on. Picking one sets the background colour and switches its gradient off, because a polished surface needs a dark surround to read as an object rather than as a hole cut in the canvas. It is a starting point, not a lock: change the background to anything you like and it stays changed.
Sections fold themselves away when they cannot reach the picture. Turn Remove Characters on and there is no character pass to colour, so Foreground closes itself. It is closed, not hidden, because it is one switch from mattering again: turn on a shader's Use Color Palettes and the ramp is taken from these very pickers. Background stays open throughout, because the background colour is painted on every frame whatever else is running. Open a folded section whenever you want; it will not close again under your hand.
This section only applies while ASCII is being drawn. On a source with Remove Characters switched on there is no character pass to style, so the whole section hides itself rather than offering controls that do nothing. Turn Remove Characters off and it comes back.
Choose from preset character sets or type your own:
.:-=+*#%@░▒▓█·•●◉◎Switch to emoji mode with 35+ curated presets organized by category:
You can also open the emoji picker to build your own custom set from any emoji.
The Kaomoji section under Material holds two things, and the difference between them matters. At the top is Characters: ten faces that were auditioned by ear, one click each to place. Below the search is the Kaomoji library: 10,327 glyphs, static. Either becomes an actor you can drag, resize and pin.
Only the ten characters speak, though everything on the canvas breathes and blinks. They are Dot, Bead, Ring, Caret, Blush, Gentle, Moon, Flat, Sheep and Beam, and each wears twenty-two mouth shapes: the same set text-to-speech and game engines use, cut from that same catalogue so every glyph is one the corpus really uses as a mouth. The ten differ in eyes, cheeks and brackets and share one mouth vocabulary, which is what makes them read as ten characters rather than ten fonts. A character's row lets you swap it for one of the other nine.
With a character selected, clicking another one in Characters recasts it rather than adding a second: same position, size and rotation, a different face. Hold Shift to add another instead, or click empty canvas to deselect first.
Select an actor and press Delete to remove it, or use the × on its row. Backspace does the same. Neither fires while you are typing, and both are undoable with ⌘ Z.
A selected actor wears a dashed box. The four corners resize it, and the round handle on the stalk above rotates it: hold Shift while turning to snap to 15°. The box turns with the actor, so the corners keep working however far round it has gone. Moving, resizing and rotating are each one undo step.
A new actor arrives at the size of the last one you placed. Actors normally share one cast size and resize together; type a pixel size into an actor's row and it is pinned on its own, and the next one you add is pinned to match. The ↺ beside the number puts an actor back with the cast.
A library kaomoji keeps the shape you found it in. Nothing can make it speak, and that is a limit of the glyph rather than a missing feature: a mouth has to be located inside a face before anything can swap it, and only the ten had that done in advance.
Mouths follow, under the cast, picks what drives them:
Mouth tuning is the pipeline's own instrument panel: the range curve, how fast the mouth opens and closes, how long it holds a shape, and how strongly it resists switching. The defaults are the tuned settings and most voices never need to move them; the dials are here for the voices that do. The line above them reports what your voice is actually producing, and human speech runs 4 to 8 mouth shapes a second, so dial until it sits in that band and looks right. Copy settings puts the numbers and the measurement on your clipboard.
Mouth sensitivity sets how far a given voice or a given jaw opens the mouth. It is only a scale: the mouth already has inertia, opening faster than it closes, and holds each shape long enough to read, so it will not flicker whatever you set here. Turn it up to reach the wide open vowels, down for something more restrained.
Under the picker, a line reports what the input is actually doing: whether audio or a face is arriving at all, a meter of the level it can see, and how far open that is pushing the mouth. If it says no audio the microphone has not been started; if it says nothing above the room it can hear something but nothing louder than the background yet.
The two live inputs drive one character, not all of them, and it is the one you have selected. Click a different character in the cast and the microphone or camera follows it; the panel says which one it is on. There is one mic and one camera in the room, so a whole cast mouthing your voice would read as several copies of you, where one of them doing it reads as that one talking. The music has no such problem, because it really is playing to all of them.
They breathe and they blink, on two clocks of their own. Breathing swells each actor a few pixels and settles it again over about four seconds; blinking shuts the eyes from the top and the bottom at once, the way a real lid closes, every few seconds. Both run at their own pace per actor, so a cast never pulses or blinks in time with itself, and neither is affected by the template's Speed: turning the pattern up does not make a face hyperventilate. Pausing the canvas holds them still.
Both have their own switch under the cast. Blinking is on by default; breathing is not. A blink reads as alive on sight, where breathing is a slow swell most people would not notice was missing, so it is there to be turned on rather than something every scene starts doing.
The two switches are global, David 2026-08-19. They reach everything on the canvas — the ten characters and every library glyph — rather than only the cast. Every glyph can breathe, because breathing is a swell in size and anything has a size. Blinking needs eyes, and a glyph has to be taken apart before anything knows where they are, so each glyph's row tells you what it found: Breathes · blinks when it could, Breathes when it could not. Around one library kaomoji in twenty parses cleanly enough, so expect most to breathe only. The ten characters were taken apart in advance, which is why they always do both. That row's switch is still there for one case: when what the parser found is punctuation rather than eyes and the blink looks wrong, turn it off for that glyph alone.
The Background section covers both meanings of the word, in two halves:
The gear beside Search settings opens Settings: things that belong to you rather than to the piece you are making. Nothing in here is saved with a preset or shared with a guest, and keys are held for the session only and cleared when you reload.
Figma colours pull a palette straight out of a Figma file:
Settings holds an OpenAI API key field, used only by Whisper transcription under Subtitles.
Make patterns respond to sound in real-time:
Bands, Beat Sync and Subtitles start collapsed. Click a heading to open it. Bands shows six named instrument ranges (Kick is 60-120Hz, Snare is 2-6kHz) so you can see which one is firing before deciding what it should drive.
Beat Sync is the half that CUTS rather than follows: on each locked beat something snaps to a new state and holds it. Cut: colour jumps your palette's hue by an interval of the colour wheel, Cut: shape swaps the shape overlay for another silhouette, and Cut: 3D object throws the model to a new angle. Cut amount sets how far a cut is allowed to go, and for shapes it is the chance a given cut moves at all.
Modulation targets (toggle on/off):
Party lives here too: the disco ball drops over the canvas and throws light around it, brightening and speeding up with whatever is driving the animation. ⌘D toggles it from anywhere. It sits over the canvas rather than in it, so it is not in exports.
Tip: If audio doesn't play after uploading, click the Play button — browsers require a user interaction to start audio.
Display synced lyrics or transcribed words over the canvas in real-time.
Draw specific regions that react to different audio frequency bands independently — a feature unique to DOTCLOUD.
Frequency bands per zone:
Presets: Center ellipse, Horizontal Bands, Vertical Bands, Radial rings, Quadrants — each auto-assigns frequency bands.
Invert zones: Toggle to make everything outside the zones reactive instead.
Paste straight from Figma. In Figma use Copy as → Copy as SVG or Copy as PNG, then press ⌘ V here: it lands on the canvas as a placed image you can drag, resize and export. The same works for a screenshot, or anything copied out of another browser tab or image editor. Repeated pastes step down and across so they do not stack on one spot.
Figma's plain ⌘ C puts its own private format on the clipboard rather than a picture, so use one of the two Copy as options. An SVG arrives as vector artwork and stays crisp at any size.
Three different things can put an image on the canvas. They are not interchangeable, and only one of them is draggable:
.json files⌘ \ hides and shows the control panel, and the button beside the search box does the same. With it hidden the artwork has the whole frame; a small button top left brings it back.
Undo covers every setting in the panel, including the ones that live behind a template's own section. Typing in a text field keeps your browser's normal text undo.
Several looks here started as somebody else's work, and it is named rather than absorbed.
Built by Cartoon Clouds