New TTS models LINNET_4 (1 credit per second) and FINCH (0.2 credits per second, resolution multiplier applies; Pro plan and above, otherwise 403 VT40314). Supported languages are listed per language in supportedTtsModels. LINNET_4 is now the default when ttsModel is omitted (previously ORIOLE); languages that do not support LINNET_4 still default to ORIOLE.

Send ttsModel explicitly if you rely on ORIOLE; omitted ttsModel now resolves to LINNET_4 where the language supports it.

TTS model DODO renamed to LINNET_3. Responses (supportedTtsModels, project ttsModel) return LINNET_3; requests still accept DODO and ELEVEN_V3 as deprecated aliases.

Compare against LINNET_3 when reading responses. Send LINNET_3 in requests.

TTS models renamed: AUDIO_ENGINE_V3 → ORIOLE, ELEVEN_V2 → WREN, ELEVEN_V3 → DODO. Responses (supportedTtsModels, project ttsModel) return the new names; requests still accept the old names.

Compare against the new names when reading responses. Send the new names in requests.

NIGHTINGALE TTS model (end-to-end dubbing with cloned voices, 6 credits per second) with targetLanguages[].cloningStrength. Estimate Quota Usage gains translateType, ttsModelCounts, projectSeq and exportCount.

Bulk Change Speakers endpoint (POST .../speakers/bulk-change) moves every script of one speaker to another in a single call.

Request examples now use AUDIO_ENGINE_V3 as the recommended ttsModel (Expressive, supported by all languages). ttsModel and supportedTtsModels now show a description per value.

Validate Media minimum media duration corrected from 5 seconds to 1 second (error F4009).

Add Speaker from Sentence endpoint (POST .../speakers/from-sentence) clones a voice from a reference sentence and adds it as a new speaker.

ttsModel must be one of the target language's supportedTtsModels, otherwise 400 VT4009. AUDIO_ENGINE_V3 added to the ttsModel values; supportedTtsModels and languageTag documented on the Language API.

Error formats documented as actually returned: gateway G0001/A0010, domain code/detailCode, and RFC 7807 parameter errors. 14 response specs aligned with production (creation status 200, projectSpeakerSeq, relative /perso-storage/... paths).

/connect page lets the perso-dubbing plugin obtain an API key in one click.

Response examples no longer show a result wrapper; they match the top-level shape the API returns.

Download target values realigned: video, originalSubtitle, translatedSubtitle removed; audioScript, scriptTimestamps, originalSubBackground added. media_type values are lowercase (video, audio). File register step is PUT.

Translate request fields targetLanguageCodes and top-level ttsModel (deprecated since 2026-05-14).

Use targetLanguages with a per-language ttsModel.

Initial public release of the Perso AI API documentation.