Live demos
These demos run the real library in your browser. Models download once from Hugging Face and are cached; nothing you type or say leaves the page. Each button says how much it downloads.
generate: text
LFM2.5 350M writes a reply, streamed token by token.
evaluate: route a prompt
A zero-shot NLI judge scores the text against each lane.
evaluate: judge a photo or a voice clip
LiquidAI d1-omni answers a question about an image, or up to 30 s of speech, in one pass. Pick a file, or run it on the sample picture.
embed: similarity
MiniLM embeds each line; the bars show how close each is to the first.
speak: text to speech
Kokoro 82M speaks, starting with the first sentence.
generate: speech to text
Record a few seconds; Moonshine Tiny transcribes it.
forecast: a time series
Chronos-Bolt Tiny continues a noisy daily pattern, with an 80% range.
paint: text to image
SD-Turbo paints a 512×512 image in one step. It needs WebGPU and downloads about 2.5 GB.