Multimodal icons

63 icons in the multimodal group of AI & LLM: Aspect crop, Audio diarisation, Audio embedding and 60 more — each in outline and duotone at three weights, with the code for React, Vue, Svelte, Flutter and more.

63 icons · 1 of 10 groupsOpen in browser

The 63 multimodal icons in AI & LLM, described

Aspect cropaspect-crop: The photo frame with a wide box inside it — a picture cut to the shape a place needs.
Audio diarisationaudio-diarise: A speech bubble with a recording cut in two — where one voice stops and another starts.
Audio embeddingaudio-embed: A speech bubble with an arrow in it — a recording turned into a vector.
Audio encoderaudio-encoder: A speech bubble with a tag around a point — the part of the model that hears.
Audio generationaudio-gen: Audio generation — sound a model made, speech or music synthesised from scratch.
Audio modelaudio-model: An audio model — a model that works on sound, listening or transcribing speech.
Audio timestampsaudio-timestamp: A speech bubble with a clock in it — each word tied to the second it was said.
Audio to textaudio-to-text: A speech bubble with a wave that becomes a line — sound turned into words.
Audio transcriptionaudio-transcribe: A speech bubble with lines of text in it — what was said, written down.
Voice activity detectionaudio-vad: A speech bubble with silence, a burst, and silence — where in the recording someone is speaking.
Background removalbackground-remove: The photo frame with a hill and the sky crossed out — the subject kept, the rest dropped.
Chart understandingchart-understand: The photo frame with bars in it — a chart read for what it says.
Cross-modal searchcross-modal-search: A speech bubble with a lens in it — a search that reads what was said and shown alike.
Denoisedenoise: Denoise — take the noise out of a signal, smoothing and cleaning it.
Depth mapdepth-map: The photo frame with lines that widen as they come nearer — how far away each part is.
Document layoutdocument-layout: The photo frame with a title and two columns — where the blocks of a page are.
Face blurface-blur: The photo frame with a head over a blur — a face hidden before the picture is shared.
Frame extractionframe-extract: The camcorder with one frame pulled out — a single still taken from the stream.
Frame interpolationframe-interpolate: The camcorder with a point between two frames — a frame made up to sit between real ones.
Image editimage-edit: The photo frame with a pencil across it — a picture changed by instruction.
Image generationimage-gen: Image generation — a picture made by a model rather than taken by a camera.
Image generationimage-generation: Image generation — a picture that did not exist until a model made it.
Image maskimage-mask: The photo frame with a region and its centre marked — the part of a picture a step applies to.
Image segmentationimage-segment: The photo frame with a hill and a corner picked out — every pixel given to the thing it belongs to.
Image to textimage-to-text: The photo frame with a small hill over lines of text — a picture described in words.
Keypointskeypoint: The photo frame with points where the joints are — a body reduced to the places that move.
Lip synclip-sync-ai: The camcorder with lips in it — a mouth moved to match the words.
Mesh generationmesh-generate: The cube's outline with a vertex at its heart — a surface built from points and faces.
Modality mixmodality-mix: A machine frame with two lines joining into one — text, image and sound taken in together.
Modality routermodality-router: A machine frame with one line splitting into two — each input sent to the model that handles its kind.
Motion transfermotion-transfer: The camcorder with a stick figure in it — one body's movement given to another.
Multimodalmultimodal: A multimodal model — one that handles images, audio and text together, not text alone.
Multimodal promptmultimodal-prompt: A speech bubble with a picture and words in it — a prompt that carries more than text.
Music generationmusic-generate: A speech bubble with a note in it — music made from a description.
Noise suppressionnoise-suppress: A speech bubble with a wave and a cross — the background taken out of the voice.
OCRocr: OCR — optical character recognition, reading the words out of a picture as text.
OCR extractionocr-extract: The photo frame with lines of text in it — the words read off a picture.
Phonemesphoneme: A speech bubble with a zigzag in it — speech broken into its smallest sounds.
Posepose: Pose — where the body's joints are, a skeleton of keypoints found in an image.
Prosodyprosody: A speech bubble with a pitch line that rises — the tune of how something is said.
Render queuerender-queue: The camcorder with frames lined up in it — clips waiting their turn to be rendered.
Scene detectionscene-detect: The camcorder with two cuts across it — where one scene ends and the next begins.
Segment anythingsegment-anything: The photo frame with a point inside a mask — click once, get the thing around it.
Shot boundaryshot-boundary: The camcorder with one cut across it — the frame where the camera changed.
Speaker diarisationspeaker-diarise: A speech bubble with two heads in it — each stretch of speech given to its speaker.
Speechspeech: Speech — sound with words in it, a spoken voice recorded or generated.
Speech alignmentspeech-align: A speech bubble with a line and ticks under it — each word matched to its moment in the audio.
Speech to textspeech-to-text: Speech to text — what was said, transcribed into written words.
Burn-in subtitlessubtitle-burn: The camcorder with a solid bar in it — subtitles drawn into the frames themselves.
Table extractiontable-extract-ai: The photo frame with a grid in it — rows and cells pulled out of a picture of a table.
Text to speechtext-to-speech: Text to speech — written words turned into a spoken voice by a model.
Texture generationtexture-generate: The cube's outline hatched — the surface a shape is dressed in.
3D generationthree-d-generate: A cube — an object made in three dimensions from a prompt or a picture.
Thumbnail pickthumbnail-pick: The photo frame with a small picture ticked — the one frame chosen to stand for the rest.
Upscale imageupscale-image: The photo frame with a small picture and an arrow to the corner — more pixels than it came with.
Video captionvideo-caption: The camcorder with a line along the bottom — words for what is being said and seen.
Video generationvideo-gen: Video generation — moving pictures that never happened, a clip made by a model.
Video generationvideo-generate: The camcorder with a bolt in it — moving pictures made from a prompt.
Video summaryvideo-summarise: The camcorder with two lines in it — a long recording told in a few sentences.
Video trackingvideo-track: The camcorder with a subject and its trail — the same thing followed frame to frame.
Vision modelvision-model: A vision model — a model that looks at images and understands what it sees.
Voice presetvoice-preset: A speech bubble with a saved pill in it — a voice chosen from the ones on offer.
Wake wordwake-word: A speech bubble with a bolt in it — the one word that wakes the device.

In code, each is one import — import { AspectCrop } from "@iconmind/react/icons/aspect-crop" — and the same name in Vue, Svelte, Solid, Preact, React Native, Astro, Blade and Flutter.

Narrow it by tag

image19 here · 29 in all speech16 here · 22 in all video12 here · 20 in all text6 here · 26 in all generate5 here · 15 in all voice5 here · 13 in all 3d4 here · 9 in all audio3 here · 19 in all extract3 here · 6 in all model3 here · 19 in all

Every multimodal icon in AI & LLM, free to ship

63 icons in outline and duotone at three weights, generated from one grid so nothing in the set can drift out of step. MIT licensed — commercial use, no attribution, no seat count.

All AI & LLM icons Install the package Request an icon

Other groups in AI & LLM

model59 inference115 prompt11 token7 context53 embedding7 training61 evaluation16 safety58