v148 · origin trial · on-device AI · Prompt API

Creative Writing Studio

See how temperature and topK reshape the same prompt. High temperature produces wild, creative variations; low temperature returns consistent, predictable text. Use the presets or dial in your own values.

Checking Prompt API with sampling parameter support…

Presets

Manual tune

temperature 1.0

topK 40

temperature: 1.0
topK: 40

Writing prompt

Run A

—

Click "Generate ×2" to see two outputs with the current parameters.

Run B

—

Same prompt, same parameters — will the outputs differ?

How sampling parameters work

// Chrome 148: temperature and topK on LanguageModel.create()
const session = await LanguageModel.create({
  temperature: 0.1,  // 0 = deterministic, 2 = chaotic
  topK: 1,           // 1 = only the most likely token, 128 = full distribution
});

// Read back the clamped values the model actually uses
console.log(session.temperature, session.topK);

// Static defaults from the model
const caps = await LanguageModel.capabilities();
console.log(caps.defaultTemperature, caps.defaultTopK);

references

implementation reference

Need the exact API surface, compatibility boundaries, errors, lifecycle, and source links? Read the matching gendn reference ↗