Smooth Reveal

The smoothing presets that turn bursty LLM chunks into a steady per-character reveal, and the hook behind them.

Models and networks deliver text in bursts, so rendering each chunk as it lands makes an answer jump forward a sentence at a time. Streamdown buffers incoming text and reveals it at an adaptive characters-per-second rate: faster when the buffer grows, draining the remainder once input goes quiet.

Presets

smoothing picks one of three presets; the default is balanced.

Preset	Target buffer	Min commit interval
`realtime`	40ms	32ms
`balanced`	120ms	48ms
`silky`	170ms	56ms

realtime keeps the smallest buffer for the lowest latency; silky holds the longest buffer for the steadiest flow. The same bursty stream feeds all three panes below.

One stream, three presets

Loading preview

useSmoothStreamContent

The pacing is available as a hook for custom renderers:

TSX

import { useSmoothStreamContent } from "@lobehub/streamdown";

const visible = useSmoothStreamContent(content, { preset: "balanced" });

Options are enabled (default true), preset (default 'balanced'), remend (options for the remend pass on the open tail) and tailUnitsRef, through which the component reports the real length of the tail block.