@nodaro/shared 3.64.0 → 3.65.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/{chunk-PJJAZQVZ.js → chunk-222QM7CO.js} +3 -3
- package/dist/{chunk-PJJAZQVZ.js.map → chunk-222QM7CO.js.map} +1 -1
- package/dist/chunk-27HA6YOL.cjs +377 -0
- package/dist/chunk-27HA6YOL.cjs.map +1 -0
- package/dist/{chunk-TTGZ2L3O.js → chunk-2NKF476Z.js} +3 -3
- package/dist/{chunk-TTGZ2L3O.js.map → chunk-2NKF476Z.js.map} +1 -1
- package/dist/chunk-3AU52TSW.js +143 -0
- package/dist/chunk-3AU52TSW.js.map +1 -0
- package/dist/{chunk-7AJGUI7L.cjs → chunk-5NVGKXJV.cjs} +34 -17
- package/dist/chunk-5NVGKXJV.cjs.map +1 -0
- package/dist/{chunk-3AYM47GW.js → chunk-5SSMPYE5.js} +99 -9
- package/dist/chunk-5SSMPYE5.js.map +1 -0
- package/dist/{chunk-KHH54GRP.cjs → chunk-6ET3KDO4.cjs} +15 -338
- package/dist/chunk-6ET3KDO4.cjs.map +1 -0
- package/dist/{chunk-ACZZHHFZ.cjs → chunk-6OSIOVF7.cjs} +44 -22
- package/dist/chunk-6OSIOVF7.cjs.map +1 -0
- package/dist/{chunk-OBOOHG3N.js → chunk-6X5PBAC4.js} +3 -3
- package/dist/{chunk-OBOOHG3N.js.map → chunk-6X5PBAC4.js.map} +1 -1
- package/dist/chunk-7BN4EDIC.cjs +185 -0
- package/dist/chunk-7BN4EDIC.cjs.map +1 -0
- package/dist/{chunk-IYSOVS7O.js → chunk-7OWXHW7Z.js} +31 -13
- package/dist/chunk-7OWXHW7Z.js.map +1 -0
- package/dist/{chunk-ZI3QDUHO.js → chunk-AT53ZQUC.js} +3 -3
- package/dist/{chunk-ZI3QDUHO.js.map → chunk-AT53ZQUC.js.map} +1 -1
- package/dist/{chunk-SHH2T4CR.js → chunk-BSTAAGLY.js} +3 -3
- package/dist/{chunk-SHH2T4CR.js.map → chunk-BSTAAGLY.js.map} +1 -1
- package/dist/{chunk-DZ4IOP3Q.js → chunk-CW6MMQNO.js} +277 -3
- package/dist/chunk-CW6MMQNO.js.map +1 -0
- package/dist/{chunk-Q2Y4JHF7.js → chunk-D5Y3EYRM.js} +5 -5
- package/dist/{chunk-Q2Y4JHF7.js.map → chunk-D5Y3EYRM.js.map} +1 -1
- package/dist/{chunk-EXEVT6D2.cjs → chunk-DQ2PARNH.cjs} +5 -5
- package/dist/{chunk-EXEVT6D2.cjs.map → chunk-DQ2PARNH.cjs.map} +1 -1
- package/dist/{chunk-OOZW2MM6.cjs → chunk-E66HSC2E.cjs} +9 -9
- package/dist/{chunk-OOZW2MM6.cjs.map → chunk-E66HSC2E.cjs.map} +1 -1
- package/dist/{chunk-NIAFR5EE.js → chunk-FGBGKBML.js} +32 -3
- package/dist/chunk-FGBGKBML.js.map +1 -0
- package/dist/{chunk-5OBTGNUQ.cjs → chunk-FI3HJZWM.cjs} +294 -2
- package/dist/chunk-FI3HJZWM.cjs.map +1 -0
- package/dist/{chunk-SPAHI7DY.cjs → chunk-FJZTQTF4.cjs} +4 -4
- package/dist/{chunk-SPAHI7DY.cjs.map → chunk-FJZTQTF4.cjs.map} +1 -1
- package/dist/{chunk-J6B7J73J.cjs → chunk-GJ3D5X4X.cjs} +49 -49
- package/dist/{chunk-J6B7J73J.cjs.map → chunk-GJ3D5X4X.cjs.map} +1 -1
- package/dist/{chunk-KB2SQAR3.js → chunk-H745DQ6V.js} +3 -3
- package/dist/{chunk-KB2SQAR3.js.map → chunk-H745DQ6V.js.map} +1 -1
- package/dist/{chunk-7KQ6W5DS.js → chunk-HAH2F7LT.js} +4 -4
- package/dist/{chunk-7KQ6W5DS.js.map → chunk-HAH2F7LT.js.map} +1 -1
- package/dist/{chunk-HDPJNNMF.cjs → chunk-JQMFJAYU.cjs} +21 -21
- package/dist/{chunk-HDPJNNMF.cjs.map → chunk-JQMFJAYU.cjs.map} +1 -1
- package/dist/{chunk-D4KWNDPH.cjs → chunk-JYU5EIRM.cjs} +12 -12
- package/dist/{chunk-D4KWNDPH.cjs.map → chunk-JYU5EIRM.cjs.map} +1 -1
- package/dist/{chunk-AZNZ5WNZ.cjs → chunk-K6RIN3LW.cjs} +142 -45
- package/dist/chunk-K6RIN3LW.cjs.map +1 -0
- package/dist/{chunk-YQV53HRR.cjs → chunk-KTCEKGBJ.cjs} +7 -7
- package/dist/{chunk-YQV53HRR.cjs.map → chunk-KTCEKGBJ.cjs.map} +1 -1
- package/dist/{chunk-W6VJAUGM.cjs → chunk-KWYS4IIM.cjs} +83 -37
- package/dist/chunk-KWYS4IIM.cjs.map +1 -0
- package/dist/{chunk-GTWUS6MQ.cjs → chunk-MIT2USTS.cjs} +61 -61
- package/dist/{chunk-GTWUS6MQ.cjs.map → chunk-MIT2USTS.cjs.map} +1 -1
- package/dist/{chunk-Y4GCJJPF.cjs → chunk-NYZQOIXJ.cjs} +35 -2
- package/dist/chunk-NYZQOIXJ.cjs.map +1 -0
- package/dist/{chunk-FHGRKQVZ.cjs → chunk-PGDR7CO3.cjs} +8 -8
- package/dist/{chunk-FHGRKQVZ.cjs.map → chunk-PGDR7CO3.cjs.map} +1 -1
- package/dist/{chunk-W7KHHNZR.js → chunk-PL7AQQV7.js} +53 -9
- package/dist/chunk-PL7AQQV7.js.map +1 -0
- package/dist/chunk-Q4YOLT4M.js +164 -0
- package/dist/chunk-Q4YOLT4M.js.map +1 -0
- package/dist/{chunk-CHS2SIP6.cjs → chunk-R5ILCNCC.cjs} +90 -76
- package/dist/chunk-R5ILCNCC.cjs.map +1 -0
- package/dist/chunk-RP4KQUNU.cjs +168 -0
- package/dist/chunk-RP4KQUNU.cjs.map +1 -0
- package/dist/{chunk-CHTZZRAL.cjs → chunk-SPU7MUWE.cjs} +6 -6
- package/dist/{chunk-CHTZZRAL.cjs.map → chunk-SPU7MUWE.cjs.map} +1 -1
- package/dist/{chunk-VPTA4Z7G.js → chunk-ST7OBB5J.js} +5 -5
- package/dist/{chunk-VPTA4Z7G.js.map → chunk-ST7OBB5J.js.map} +1 -1
- package/dist/{chunk-TPQJKS7D.js → chunk-V5NLQ4L5.js} +21 -7
- package/dist/{chunk-CHS2SIP6.cjs.map → chunk-V5NLQ4L5.js.map} +1 -1
- package/dist/{chunk-QYU7J5EQ.js → chunk-V6OXSXIX.js} +29 -18
- package/dist/chunk-V6OXSXIX.js.map +1 -0
- package/dist/{chunk-T2OVDF2M.cjs → chunk-WQMPIURW.cjs} +6 -6
- package/dist/{chunk-T2OVDF2M.cjs.map → chunk-WQMPIURW.cjs.map} +1 -1
- package/dist/{chunk-67CMMTLN.cjs → chunk-WZG6GZIL.cjs} +4 -4
- package/dist/{chunk-67CMMTLN.cjs.map → chunk-WZG6GZIL.cjs.map} +1 -1
- package/dist/{chunk-36VWLB4J.js → chunk-XJANKTIL.js} +3 -3
- package/dist/{chunk-36VWLB4J.js.map → chunk-XJANKTIL.js.map} +1 -1
- package/dist/{chunk-O7RZRUMZ.js → chunk-XJWED257.js} +14 -309
- package/dist/chunk-XJWED257.js.map +1 -0
- package/dist/chunk-XSSLNINL.js +346 -0
- package/dist/chunk-XSSLNINL.js.map +1 -0
- package/dist/{chunk-O4RPXBDZ.cjs → chunk-YD4XY74W.cjs} +8 -8
- package/dist/{chunk-O4RPXBDZ.cjs.map → chunk-YD4XY74W.cjs.map} +1 -1
- package/dist/{chunk-SXHHLZ22.js → chunk-YMZUE2IX.js} +9 -9
- package/dist/{chunk-SXHHLZ22.js.map → chunk-YMZUE2IX.js.map} +1 -1
- package/dist/{chunk-4IUE2JB7.js → chunk-Z77M7Q2R.js} +3 -3
- package/dist/{chunk-4IUE2JB7.js.map → chunk-Z77M7Q2R.js.map} +1 -1
- package/dist/{chunk-3KTWYYCN.js → chunk-ZSRXY5YT.js} +3 -3
- package/dist/{chunk-3KTWYYCN.js.map → chunk-ZSRXY5YT.js.map} +1 -1
- package/dist/entries/apps.cjs +42 -42
- package/dist/entries/apps.js +7 -7
- package/dist/entries/captions.d.cts +1 -1
- package/dist/entries/captions.d.ts +1 -1
- package/dist/entries/credits.cjs +230 -214
- package/dist/entries/credits.d.cts +51 -14
- package/dist/entries/credits.d.ts +51 -14
- package/dist/entries/credits.js +10 -10
- package/dist/entries/entities.cjs +202 -193
- package/dist/entries/entities.d.cts +44 -9
- package/dist/entries/entities.d.ts +44 -9
- package/dist/entries/entities.js +2 -1
- package/dist/entries/llm-models.cjs +86 -62
- package/dist/entries/llm-models.d.cts +46 -15
- package/dist/entries/llm-models.d.ts +46 -15
- package/dist/entries/llm-models.js +2 -2
- package/dist/entries/media.cjs +79 -63
- package/dist/entries/media.d.cts +154 -1
- package/dist/entries/media.d.ts +154 -1
- package/dist/entries/media.js +1 -1
- package/dist/entries/model-catalog.cjs +175 -119
- package/dist/entries/model-catalog.d.cts +4 -54
- package/dist/entries/model-catalog.d.ts +4 -54
- package/dist/entries/model-catalog.js +4 -4
- package/dist/entries/models.cjs +491 -358
- package/dist/entries/models.d.cts +86 -7
- package/dist/entries/models.d.ts +86 -7
- package/dist/entries/models.js +6 -5
- package/dist/entries/nodes.cjs +212 -212
- package/dist/entries/nodes.d.cts +1 -1
- package/dist/entries/nodes.d.ts +1 -1
- package/dist/entries/nodes.js +8 -8
- package/dist/entries/pipeline.cjs +153 -153
- package/dist/entries/pipeline.d.cts +3 -3
- package/dist/entries/pipeline.d.ts +3 -3
- package/dist/entries/pipeline.js +3 -3
- package/dist/entries/social.cjs +177 -177
- package/dist/entries/social.js +2 -2
- package/dist/entries/ugc.cjs +71 -71
- package/dist/entries/ugc.js +4 -4
- package/dist/entries/video.cjs +174 -145
- package/dist/entries/video.d.cts +114 -2
- package/dist/entries/video.d.ts +114 -2
- package/dist/entries/video.js +7 -6
- package/dist/entries/workflow.d.cts +1 -1
- package/dist/entries/workflow.d.ts +1 -1
- package/dist/index.cjs +2074 -1793
- package/dist/index.d.cts +10 -9
- package/dist/index.d.ts +10 -9
- package/dist/index.js +24 -23
- package/dist/{model-catalog-Cf9fkEUp.d.ts → model-catalog-BoUKGFNl.d.cts} +167 -1
- package/dist/{model-catalog-Cf9fkEUp.d.cts → model-catalog-BoUKGFNl.d.ts} +167 -1
- package/dist/{model-constants-BxjiWVQ9.d.ts → model-constants-BElS3AeE.d.ts} +55 -6
- package/dist/{model-constants-C_Nyft_i.d.cts → model-constants-CpQz1bGa.d.cts} +55 -6
- package/dist/tts-capabilities-BtnDWK9K.d.ts +133 -0
- package/dist/tts-capabilities-C52UccTI.d.cts +133 -0
- package/package.json +1 -1
- package/src/__tests__/dialogue-capabilities.test.ts +167 -6
- package/src/__tests__/extend-providers.test.ts +87 -0
- package/src/__tests__/kling4-catalog.test.ts +1 -1
- package/src/__tests__/llm-models.test.ts +41 -4
- package/src/__tests__/speech-pricing.test.ts +62 -1
- package/src/__tests__/tts-capabilities.test.ts +160 -4
- package/src/__tests__/video-analysis-pricing.test.ts +13 -1
- package/src/__tests__/video-subjects.test.ts +192 -0
- package/src/__tests__/wired-character-names.test.ts +142 -0
- package/src/credit-estimators/index.ts +4 -0
- package/src/credit-estimators/speech.ts +57 -9
- package/src/credit-identifiers.ts +48 -13
- package/src/dialogue-capabilities.ts +220 -3
- package/src/dubbing-structure.ts +186 -0
- package/src/entries/entities.ts +2 -0
- package/src/entries/llm-models.ts +7 -0
- package/src/entries/media.ts +3 -1
- package/src/entries/model-catalog.ts +17 -0
- package/src/entries/models.ts +39 -0
- package/src/entries/video.ts +8 -1
- package/src/gemini-tts.ts +144 -0
- package/src/llm-models.ts +71 -21
- package/src/model-catalog.ts +259 -0
- package/src/model-constants.ts +101 -8
- package/src/tts-capabilities.ts +193 -1
- package/src/video-analysis-pricing.ts +11 -0
- package/src/video-character-refs.ts +160 -3
- package/src/wired-character-names.ts +134 -18
- package/src/wired-character-video-subjects.ts +63 -0
- package/dist/chunk-3AYM47GW.js.map +0 -1
- package/dist/chunk-5OBTGNUQ.cjs.map +0 -1
- package/dist/chunk-6YIV37V7.js +0 -31
- package/dist/chunk-6YIV37V7.js.map +0 -1
- package/dist/chunk-7AJGUI7L.cjs.map +0 -1
- package/dist/chunk-ACZZHHFZ.cjs.map +0 -1
- package/dist/chunk-AZNZ5WNZ.cjs.map +0 -1
- package/dist/chunk-DZ4IOP3Q.js.map +0 -1
- package/dist/chunk-IYSOVS7O.js.map +0 -1
- package/dist/chunk-KHH54GRP.cjs.map +0 -1
- package/dist/chunk-NIAFR5EE.js.map +0 -1
- package/dist/chunk-O7RZRUMZ.js.map +0 -1
- package/dist/chunk-QYU7J5EQ.js.map +0 -1
- package/dist/chunk-S3UGLZSF.js +0 -42
- package/dist/chunk-S3UGLZSF.js.map +0 -1
- package/dist/chunk-TPQJKS7D.js.map +0 -1
- package/dist/chunk-TQW3R44P.cjs +0 -38
- package/dist/chunk-TQW3R44P.cjs.map +0 -1
- package/dist/chunk-VZ6TE23U.cjs +0 -54
- package/dist/chunk-VZ6TE23U.cjs.map +0 -1
- package/dist/chunk-W6VJAUGM.cjs.map +0 -1
- package/dist/chunk-W7KHHNZR.js.map +0 -1
- package/dist/chunk-Y4GCJJPF.cjs.map +0 -1
package/src/model-catalog.ts
CHANGED
|
@@ -35,6 +35,19 @@ import { isAutoVideoDuration } from "./video-duration-auto.js"
|
|
|
35
35
|
import type { DeclaredSeedanceTaskType } from "./seedance-task-type.js"
|
|
36
36
|
import { SEEDREAM_5_5_DEFAULT_RESOLUTION, SEEDREAM_5_5_RESOLUTIONS } from "./seedream-5-5.js"
|
|
37
37
|
import { FLUX_3_KLEIN_RESOLUTIONS } from "./flux-3-klein.js"
|
|
38
|
+
import {
|
|
39
|
+
GEMINI_3_1_DEFAULT_VOICE,
|
|
40
|
+
GEMINI_3_1_TTS_ACCENTS,
|
|
41
|
+
GEMINI_3_1_TTS_PACES,
|
|
42
|
+
GEMINI_3_1_TTS_STYLES,
|
|
43
|
+
GEMINI_3_1_TTS_VOICES,
|
|
44
|
+
GEMINI_3_8_DEFAULT_VOICE,
|
|
45
|
+
GEMINI_3_8_TONE_TAGS,
|
|
46
|
+
GEMINI_3_8_TTS_VOICES,
|
|
47
|
+
GEMINI_TTS_MAX_CHARS,
|
|
48
|
+
TTS_DIRECTION_MAX_CHARS,
|
|
49
|
+
type TtsNamedVoice,
|
|
50
|
+
} from "./gemini-tts.js"
|
|
38
51
|
|
|
39
52
|
export type ModelKind = "image" | "video" | "audio"
|
|
40
53
|
|
|
@@ -108,6 +121,14 @@ export interface PriceVariant {
|
|
|
108
121
|
*/
|
|
109
122
|
export const SPEECH_UNIT_PRICE_NOTE = "per started 100 characters, at least 8 units — where length pricing is on; a request costs at most its units × this"
|
|
110
123
|
|
|
124
|
+
/**
|
|
125
|
+
* The note on a speech model's `<model>:per-1k-chars` row (`TtsCapabilities.ratePer1kChars`):
|
|
126
|
+
* the same count as {@link SPEECH_UNIT_PRICE_NOTE} — started 100 characters, at
|
|
127
|
+
* least 8 units — quoted per 1,000, so a request costs its hundreds × this ÷ 10,
|
|
128
|
+
* rounded up once (the exact price of N hundreds is the `<model>:per-100-chars:<N>u` id).
|
|
129
|
+
*/
|
|
130
|
+
export const SPEECH_PER_1K_PRICE_NOTE = "per 1,000 characters, counted in started 100s, at least 800 — where length pricing is on; a request costs its hundreds × this ÷ 10, rounded up"
|
|
131
|
+
|
|
111
132
|
/** A voice-setting lever a speech model honours. */
|
|
112
133
|
export type TtsSettingLever = "stability" | "similarity" | "style" | "speed" | "speakerBoost"
|
|
113
134
|
|
|
@@ -160,6 +181,81 @@ export interface TtsCapabilities {
|
|
|
160
181
|
maxChars: number
|
|
161
182
|
/** Languages offered in pickers (ISO 639-1; `fil` has no two-letter code). A curated list, not the vendor's full one. */
|
|
162
183
|
languages: readonly string[]
|
|
184
|
+
/**
|
|
185
|
+
* The model speaks only these named voices, by wire name: a request's
|
|
186
|
+
* `voice` must be one of them (else a 400 before any hold), and none means
|
|
187
|
+
* `defaultVoice`. Absent ⇒ the ElevenLabs voices (premade, library, clones).
|
|
188
|
+
* A client offers this list in place of the ElevenLabs voice browser.
|
|
189
|
+
*/
|
|
190
|
+
voices?: readonly TtsNamedVoice[]
|
|
191
|
+
/** With `voices`: the voice a request that names none runs with. */
|
|
192
|
+
defaultVoice?: string
|
|
193
|
+
/**
|
|
194
|
+
* With `audioTags`: the tags are written `<in angle brackets>`, from the
|
|
195
|
+
* closed list `tags`. The platform's `[square]` tags are rewritten to them
|
|
196
|
+
* where they match and the rest are taken out before sending
|
|
197
|
+
* (`@nodaro/prompts` `toAngleTags`). Absent ⇒ `[square]` tags, sent as written.
|
|
198
|
+
*/
|
|
199
|
+
tagSyntax?: "angle"
|
|
200
|
+
/** With `tagSyntax: "angle"`: the tags the model performs, by their own names (`laugh`, `short pause`). */
|
|
201
|
+
tags?: readonly string[]
|
|
202
|
+
/**
|
|
203
|
+
* Style direction the model takes beside the text — never spoken, never
|
|
204
|
+
* billed. Absent ⇒ none (the request's direction fields are not sent).
|
|
205
|
+
*/
|
|
206
|
+
direction?: TtsDirectionSheet
|
|
207
|
+
/**
|
|
208
|
+
* Dialogue lane only: at most this many speakers (distinct voices) in one
|
|
209
|
+
* script. Absent ⇒ the lane's own limit (ElevenLabs dialogue: 10).
|
|
210
|
+
*/
|
|
211
|
+
dialogueSpeakers?: number
|
|
212
|
+
/**
|
|
213
|
+
* Dialogue lane only: the model can add natural filler words ("um", "you
|
|
214
|
+
* know") between two speakers when asked (`fillerWords` on a request).
|
|
215
|
+
*/
|
|
216
|
+
fillerWords?: true
|
|
217
|
+
/**
|
|
218
|
+
* Dialogue lane only: the model delivers every speech file format
|
|
219
|
+
* (`TTS_OUTPUT_FORMATS`) — it renders a WAV the platform keeps as it is or
|
|
220
|
+
* encodes to MP3. Absent ⇒ the default MP3 only (the ElevenLabs dialogue
|
|
221
|
+
* endpoint, as the platform calls it).
|
|
222
|
+
*/
|
|
223
|
+
dialogueAnyOutputFormat?: true
|
|
224
|
+
/**
|
|
225
|
+
* The model clones a speaker's voice for one request from two recordings
|
|
226
|
+
* (`replica`: a reference and a consent recording, `TtsReplica`), in place of
|
|
227
|
+
* a named voice; nothing is saved as a voice. Gemini 3.8 Flash only: KIE's
|
|
228
|
+
* 3.8 Flash-Lite refuses it ("Inline voice sample replication is not
|
|
229
|
+
* supported", probed 2026-11-13).
|
|
230
|
+
*/
|
|
231
|
+
replica?: true
|
|
232
|
+
/**
|
|
233
|
+
* The model's length-price row is quoted per 1,000 characters
|
|
234
|
+
* (`<id>:per-1k-chars`) instead of per started 100 (`<id>:per-100-chars`).
|
|
235
|
+
* The count is the same — started 100 characters, at least 8 — and a
|
|
236
|
+
* request costs its hundreds × the row ÷ 10, rounded up once
|
|
237
|
+
* (`speechRowCredits`). For a rate below one credit per 100 characters,
|
|
238
|
+
* which a whole-credit row cannot hold.
|
|
239
|
+
*/
|
|
240
|
+
ratePer1kChars?: true
|
|
241
|
+
}
|
|
242
|
+
|
|
243
|
+
/**
|
|
244
|
+
* The style direction a speech model takes (`styleDirection`, `pace`,
|
|
245
|
+
* `accent`, `audioProfile` on a request). A field the sheet does not name is
|
|
246
|
+
* never sent.
|
|
247
|
+
*/
|
|
248
|
+
export interface TtsDirectionSheet {
|
|
249
|
+
/** `styleDirection` is free text — tone, emotion, pace in the person's own words — up to this many characters. */
|
|
250
|
+
styleMaxChars?: number
|
|
251
|
+
/** `styleDirection` is one of these (the model's own names). */
|
|
252
|
+
styles?: readonly string[]
|
|
253
|
+
/** `pace` is one of these. */
|
|
254
|
+
paces?: readonly string[]
|
|
255
|
+
/** `accent` is one of these. */
|
|
256
|
+
accents?: readonly string[]
|
|
257
|
+
/** `audioProfile` (who is speaking, in words) is free text up to this many characters. */
|
|
258
|
+
audioProfileMaxChars?: number
|
|
163
259
|
}
|
|
164
260
|
|
|
165
261
|
/**
|
|
@@ -605,6 +701,7 @@ export const MODEL_RECOMMENDATIONS: readonly ModelRecommendation[] = [
|
|
|
605
701
|
{ intent: "video with start + end frame", modelIds: ["veo3", "veo3.1", "kling-turbo", "minimax", "hailuo-standard", "seedance-2"], note: "All listed support an end frame; VEO uses imageUrls[start, end]." },
|
|
606
702
|
{ intent: "music / song generation", modelIds: ["suno-v6", "suno-v6_wild", "suno-v6_mini", "suno-v5_5"], note: "V6 is the default flagship; V6 Wild for bolder, less predictable results; V6 Mini when speed matters; v5.5 / v5 / v4 keep their own character. Same price." },
|
|
607
703
|
{ intent: "voice over / narration", modelIds: ["elevenlabs-v4", "elevenlabs-v3", "elevenlabs-v4-turbo", "elevenlabs-turbo"], note: "v4 is the default: [audio tags] for emotion and up to 10,000 characters per request. v3 is the previous expressive model, still selectable. v4 Turbo keeps v4's tags at Turbo's credit price; Turbo v2.5 is the cheapest for plain narration without tags." },
|
|
704
|
+
{ intent: "low-cost narration with a named voice", modelIds: ["gemini-3-8-flash-lite-tts", "gemini-3-8-flash-tts", "gemini-3-1-flash-tts", "gemini-2-5-pro-tts"], note: "Google's Gemini voices cost far less per character than ElevenLabs. Each speaks only its own named voices (no clones or library voices) and no language is set: the text decides. 3.8 Flash-Lite is the cheapest; 3.8 Flash is the larger 3.8 model; both take their own tone tags (laugh, sigh, short pause…) and a free-text style. 3.1 Flash and 2.5 Pro take [audio tags] and a style, pace, accent and audio profile." },
|
|
608
705
|
{ intent: "lip-sync a portrait to audio", modelIds: ["kling-avatar-pro", "kling-avatar", "infinitalk"], note: "Pro for best mouth shape; InfiniTalk for resolution control." },
|
|
609
706
|
{ intent: "transcription / captions", modelIds: ["elevenlabs-stt", "incredibly-fast-whisper", "whisper", "nemotron-3-5-transcribe"], note: "Captions need WORD timestamps: ElevenLabs STT (always), Incredibly Fast Whisper or Nemotron 3.5 Transcribe (on request, with a speaker per word). Plain Whisper returns phrase segments only. To translate the speech (to or from English) instead of transcribing it, use Nemotron 3.5 Transcribe with a target language; a translation carries no word timings." },
|
|
610
707
|
{ intent: "motion transfer (drive a subject by another video)", modelIds: ["motion-transfer", "kling-3.0-motion"], note: "Kling 2.6 base is cheap; Kling 3.0 is premium." },
|
|
@@ -4110,6 +4207,29 @@ const VIDEO_MODELS: Record<string, ModelCatalogEntry> = {
|
|
|
4110
4207
|
{ identifier: "video-analysis:gemini-3.1-pro:600s", credits: 1056, note: "10-min ceiling" },
|
|
4111
4208
|
],
|
|
4112
4209
|
},
|
|
4210
|
+
// ULTRA (since 2026-11-13) — the deepest-reasoning analysis model reading
|
|
4211
|
+
// every window, at the same frame sampling as the other single-model tiers.
|
|
4212
|
+
// Priced between the mixed tiers and smart.
|
|
4213
|
+
"gemini-4-carbon-video-analysis": {
|
|
4214
|
+
id: "gemini-4-carbon-video-analysis",
|
|
4215
|
+
kind: "video",
|
|
4216
|
+
modes: ["video-analysis"] as const,
|
|
4217
|
+
family: "Nodaro",
|
|
4218
|
+
maker: "nodaro",
|
|
4219
|
+
label: "Video Analysis (Ultra)",
|
|
4220
|
+
series: "Video Analysis",
|
|
4221
|
+
description: "Deep-reasoning analysis — the most thorough model reads every window of the video. Better on fast cuts and on scenes with many people, and it does not invent cuts. Billed per duration bucket.",
|
|
4222
|
+
// Not `most-accurate`: that tag is Smart's (hand-counted edit list, always
|
|
4223
|
+
// refined). Ultra's strengths are the footage that trips the other tiers.
|
|
4224
|
+
useCases: ["video-analysis", "shot-list", "premium", "fast-cuts", "many-people", "no-invented-cuts"],
|
|
4225
|
+
pricing: [
|
|
4226
|
+
{ identifier: "video-analysis:gemini-4-carbon", credits: 1745, note: "10-min ceiling (no duration given)" },
|
|
4227
|
+
{ identifier: "video-analysis:gemini-4-carbon:60s", credits: 349 },
|
|
4228
|
+
{ identifier: "video-analysis:gemini-4-carbon:180s", credits: 375 },
|
|
4229
|
+
{ identifier: "video-analysis:gemini-4-carbon:360s", credits: 1054 },
|
|
4230
|
+
{ identifier: "video-analysis:gemini-4-carbon:600s", credits: 1745, note: "10-min ceiling" },
|
|
4231
|
+
],
|
|
4232
|
+
},
|
|
4113
4233
|
// Both mixed tiers are variants of the same advanced multi-engine analysis
|
|
4114
4234
|
// and share this ONE credit family (videoAnalysisCreditSegment maps both;
|
|
4115
4235
|
// plan internals live in the private analysis plugin).
|
|
@@ -4244,6 +4364,65 @@ const DIALOGUE_V4_MAX_CHARS = 10000
|
|
|
4244
4364
|
*/
|
|
4245
4365
|
const V4_TURBO_MAX_CHARS = 10000
|
|
4246
4366
|
|
|
4367
|
+
/**
|
|
4368
|
+
* Gemini 3.8 (Flash-Lite, Flash) speech, one speaker: its own 70 voices, its
|
|
4369
|
+
* closed list of `<angle>` tone tags, a free-text style per line. No language
|
|
4370
|
+
* field, no timings, no neighbour text, no ElevenLabs levers. From the
|
|
4371
|
+
* provider's schema (2026-11-12); whether each tag is performed rather than
|
|
4372
|
+
* read aloud is to be checked on staging.
|
|
4373
|
+
*/
|
|
4374
|
+
const GEMINI_3_8_TTS_SHEET: TtsCapabilities = {
|
|
4375
|
+
audioTags: true,
|
|
4376
|
+
tagSyntax: "angle",
|
|
4377
|
+
tags: GEMINI_3_8_TONE_TAGS,
|
|
4378
|
+
ssmlBreaks: false,
|
|
4379
|
+
levers: [],
|
|
4380
|
+
languageCode: false,
|
|
4381
|
+
timestamps: false,
|
|
4382
|
+
stitching: false,
|
|
4383
|
+
maxChars: GEMINI_TTS_MAX_CHARS,
|
|
4384
|
+
// No picker: the language follows the text.
|
|
4385
|
+
languages: [],
|
|
4386
|
+
voices: GEMINI_3_8_TTS_VOICES,
|
|
4387
|
+
defaultVoice: GEMINI_3_8_DEFAULT_VOICE,
|
|
4388
|
+
direction: { styleMaxChars: TTS_DIRECTION_MAX_CHARS },
|
|
4389
|
+
ratePer1kChars: true,
|
|
4390
|
+
// Two-speaker dialogue (decided 2026-11-12): one or two voices, a style per line, filler words on request.
|
|
4391
|
+
dialogueSpeakers: 2,
|
|
4392
|
+
fillerWords: true,
|
|
4393
|
+
// A dialogue's file is chosen as one speaker's is: MP3 or the WAV as rendered (decided 2026-11-13).
|
|
4394
|
+
dialogueAnyOutputFormat: true,
|
|
4395
|
+
}
|
|
4396
|
+
|
|
4397
|
+
/**
|
|
4398
|
+
* Gemini 3.1 Flash and 2.5 Pro speech, one speaker: 30 voices, `[square]`
|
|
4399
|
+
* tone tags written freely (sent as written), and the speaker's style, pace,
|
|
4400
|
+
* accent and audio profile. Otherwise as the 3.8 sheet.
|
|
4401
|
+
*/
|
|
4402
|
+
const GEMINI_3_1_TTS_SHEET: TtsCapabilities = {
|
|
4403
|
+
audioTags: true,
|
|
4404
|
+
ssmlBreaks: false,
|
|
4405
|
+
levers: [],
|
|
4406
|
+
languageCode: false,
|
|
4407
|
+
timestamps: false,
|
|
4408
|
+
stitching: false,
|
|
4409
|
+
maxChars: GEMINI_TTS_MAX_CHARS,
|
|
4410
|
+
languages: [],
|
|
4411
|
+
voices: GEMINI_3_1_TTS_VOICES,
|
|
4412
|
+
defaultVoice: GEMINI_3_1_DEFAULT_VOICE,
|
|
4413
|
+
direction: {
|
|
4414
|
+
styles: GEMINI_3_1_TTS_STYLES,
|
|
4415
|
+
paces: GEMINI_3_1_TTS_PACES,
|
|
4416
|
+
accents: GEMINI_3_1_TTS_ACCENTS,
|
|
4417
|
+
audioProfileMaxChars: TTS_DIRECTION_MAX_CHARS,
|
|
4418
|
+
},
|
|
4419
|
+
ratePer1kChars: true,
|
|
4420
|
+
// Two-speaker dialogue (decided 2026-11-12): one or two voices, the style, pace, accent and profile per speaker.
|
|
4421
|
+
dialogueSpeakers: 2,
|
|
4422
|
+
// A dialogue's file is chosen as one speaker's is: MP3 or the WAV as rendered (decided 2026-11-13).
|
|
4423
|
+
dialogueAnyOutputFormat: true,
|
|
4424
|
+
}
|
|
4425
|
+
|
|
4247
4426
|
const AUDIO_MODELS: Record<string, ModelCatalogEntry> = {
|
|
4248
4427
|
// ── ElevenLabs TTS ──
|
|
4249
4428
|
"elevenlabs-v3": {
|
|
@@ -4392,6 +4571,86 @@ const AUDIO_MODELS: Record<string, ModelCatalogEntry> = {
|
|
|
4392
4571
|
languages: TTS_LANGS_MULTILINGUAL_V2,
|
|
4393
4572
|
},
|
|
4394
4573
|
},
|
|
4574
|
+
|
|
4575
|
+
// ── Google Gemini TTS (through KIE) ──
|
|
4576
|
+
// One id per model for both lanes: the same KIE model takes one speaker
|
|
4577
|
+
// (Text to Speech) and two (Text to Dialogue), and dialogue costs the same per
|
|
4578
|
+
// character as one speaker (decided 2026-11-12) — so `tts` AND `dialogue`,
|
|
4579
|
+
// one sheet and one price row, the same rate by construction.
|
|
4580
|
+
// Named voices of their own (never ElevenLabs'), no language field (the
|
|
4581
|
+
// language follows the text), no timings, no neighbour text. Priced per
|
|
4582
|
+
// 1,000 characters on the same count as every speech model (decided
|
|
4583
|
+
// 2026-11-12, provisional until measured on staging); the flat row is the
|
|
4584
|
+
// price of 1,000 characters.
|
|
4585
|
+
"gemini-3-8-flash-lite-tts": {
|
|
4586
|
+
id: "gemini-3-8-flash-lite-tts",
|
|
4587
|
+
kind: "audio",
|
|
4588
|
+
modes: ["tts", "dialogue"] as const,
|
|
4589
|
+
family: "Google",
|
|
4590
|
+
maker: "google",
|
|
4591
|
+
label: "Gemini 3.8 Flash-Lite TTS",
|
|
4592
|
+
series: "Gemini",
|
|
4593
|
+
description: "Google's lightest Gemini speech model and the cheapest speech on the platform — 70 named voices grouped by use, its own tone tags (laugh, sigh, short pause and more) and a free-text style direction. The language follows the text; up to 10,000 characters per request. Also voices a dialogue of two speakers.",
|
|
4594
|
+
useCases: ["tts", "voice-over", "narration", "fast", "budget"],
|
|
4595
|
+
features: ["audio-tags"],
|
|
4596
|
+
pricing: [
|
|
4597
|
+
{ identifier: "gemini-3-8-flash-lite-tts", credits: 4 },
|
|
4598
|
+
{ identifier: "gemini-3-8-flash-lite-tts:per-1k-chars", credits: 4, note: SPEECH_PER_1K_PRICE_NOTE },
|
|
4599
|
+
],
|
|
4600
|
+
tts: GEMINI_3_8_TTS_SHEET,
|
|
4601
|
+
},
|
|
4602
|
+
"gemini-3-8-flash-tts": {
|
|
4603
|
+
id: "gemini-3-8-flash-tts",
|
|
4604
|
+
kind: "audio",
|
|
4605
|
+
modes: ["tts", "dialogue"] as const,
|
|
4606
|
+
family: "Google",
|
|
4607
|
+
maker: "google",
|
|
4608
|
+
label: "Gemini 3.8 Flash TTS",
|
|
4609
|
+
series: "Gemini",
|
|
4610
|
+
description: "Google's Gemini 3.8 speech model — the same 70 voices, tone tags and free-text style direction as Flash-Lite, from the larger model of the generation. The language follows the text; up to 10,000 characters per request. Also voices a dialogue of two speakers.",
|
|
4611
|
+
useCases: ["tts", "voice-over", "narration", "expressive"],
|
|
4612
|
+
features: ["audio-tags"],
|
|
4613
|
+
pricing: [
|
|
4614
|
+
{ identifier: "gemini-3-8-flash-tts", credits: 6 },
|
|
4615
|
+
{ identifier: "gemini-3-8-flash-tts:per-1k-chars", credits: 6, note: SPEECH_PER_1K_PRICE_NOTE },
|
|
4616
|
+
],
|
|
4617
|
+
// A one-off clone (`replica`, decided 2026-11-13) on 3.8 Flash alone: its Flash-Lite sibling refuses it.
|
|
4618
|
+
tts: { ...GEMINI_3_8_TTS_SHEET, replica: true },
|
|
4619
|
+
},
|
|
4620
|
+
"gemini-3-1-flash-tts": {
|
|
4621
|
+
id: "gemini-3-1-flash-tts",
|
|
4622
|
+
kind: "audio",
|
|
4623
|
+
modes: ["tts", "dialogue"] as const,
|
|
4624
|
+
family: "Google",
|
|
4625
|
+
maker: "google",
|
|
4626
|
+
label: "Gemini 3.1 Flash TTS",
|
|
4627
|
+
series: "Gemini",
|
|
4628
|
+
description: "Google's Gemini 3.1 speech model — 30 named voices, [audio tags] written freely, and a speaking style, pace, accent and audio profile for the speaker. The language follows the text; up to 10,000 characters per request. Also voices a dialogue of two speakers.",
|
|
4629
|
+
useCases: ["tts", "voice-over", "narration", "expressive"],
|
|
4630
|
+
features: ["audio-tags"],
|
|
4631
|
+
pricing: [
|
|
4632
|
+
{ identifier: "gemini-3-1-flash-tts", credits: 13 },
|
|
4633
|
+
{ identifier: "gemini-3-1-flash-tts:per-1k-chars", credits: 13, note: SPEECH_PER_1K_PRICE_NOTE },
|
|
4634
|
+
],
|
|
4635
|
+
tts: GEMINI_3_1_TTS_SHEET,
|
|
4636
|
+
},
|
|
4637
|
+
"gemini-2-5-pro-tts": {
|
|
4638
|
+
id: "gemini-2-5-pro-tts",
|
|
4639
|
+
kind: "audio",
|
|
4640
|
+
modes: ["tts", "dialogue"] as const,
|
|
4641
|
+
family: "Google",
|
|
4642
|
+
maker: "google",
|
|
4643
|
+
label: "Gemini 2.5 Pro TTS",
|
|
4644
|
+
series: "Gemini",
|
|
4645
|
+
description: "Google's Gemini 2.5 Pro speech model — the same 30 voices, [audio tags] and speaker controls (style, pace, accent, audio profile) as Gemini 3.1 Flash TTS, from the Pro model. The language follows the text; up to 10,000 characters per request. Also voices a dialogue of two speakers.",
|
|
4646
|
+
useCases: ["tts", "voice-over", "narration", "expressive", "long-form"],
|
|
4647
|
+
features: ["audio-tags"],
|
|
4648
|
+
pricing: [
|
|
4649
|
+
{ identifier: "gemini-2-5-pro-tts", credits: 16 },
|
|
4650
|
+
{ identifier: "gemini-2-5-pro-tts:per-1k-chars", credits: 16, note: SPEECH_PER_1K_PRICE_NOTE },
|
|
4651
|
+
],
|
|
4652
|
+
tts: GEMINI_3_1_TTS_SHEET,
|
|
4653
|
+
},
|
|
4395
4654
|
"elevenlabs-dialogue": {
|
|
4396
4655
|
id: "elevenlabs-dialogue",
|
|
4397
4656
|
kind: "audio",
|
package/src/model-constants.ts
CHANGED
|
@@ -1805,6 +1805,12 @@ export const TTS_PROVIDERS = [
|
|
|
1805
1805
|
"elevenlabs-v4-turbo",
|
|
1806
1806
|
"elevenlabs-turbo",
|
|
1807
1807
|
"elevenlabs-multilingual",
|
|
1808
|
+
// Google's Gemini speech through KIE (decided 2026-11-12): named voices of
|
|
1809
|
+
// their own (`TtsCapabilities.voices`), never ElevenLabs'.
|
|
1810
|
+
"gemini-3-8-flash-lite-tts",
|
|
1811
|
+
"gemini-3-8-flash-tts",
|
|
1812
|
+
"gemini-3-1-flash-tts",
|
|
1813
|
+
"gemini-2-5-pro-tts",
|
|
1808
1814
|
"elevenlabs",
|
|
1809
1815
|
] as const
|
|
1810
1816
|
export type TtsProvider = typeof TTS_PROVIDERS[number]
|
|
@@ -1831,10 +1837,20 @@ export const TTS_OUTPUT_FORMAT_FILES: Readonly<Record<TtsOutputFormat, { readonl
|
|
|
1831
1837
|
/**
|
|
1832
1838
|
* Text-to-dialogue models (the `provider` of `/v1/text-to-dialogue` and of the
|
|
1833
1839
|
* Text to Dialogue node). Each IS its catalog id and its credit identifier.
|
|
1834
|
-
*
|
|
1835
|
-
*
|
|
1836
|
-
|
|
1837
|
-
|
|
1840
|
+
* The ElevenLabs dialogue models are dialogue-only ids, never members of
|
|
1841
|
+
* TTS_PROVIDERS: their dialogue endpoint is its own model. Google's Gemini
|
|
1842
|
+
* models take one speaker and two under ONE id, so each is in both lists,
|
|
1843
|
+
* priced on its one row (dialogue costs what one speaker costs per character,
|
|
1844
|
+
* decided 2026-11-12); `dialogue-capabilities.test.ts` pins exactly that.
|
|
1845
|
+
*/
|
|
1846
|
+
export const DIALOGUE_PROVIDERS = [
|
|
1847
|
+
"elevenlabs-dialogue",
|
|
1848
|
+
"elevenlabs-dialogue-v4",
|
|
1849
|
+
"gemini-3-8-flash-lite-tts",
|
|
1850
|
+
"gemini-3-8-flash-tts",
|
|
1851
|
+
"gemini-3-1-flash-tts",
|
|
1852
|
+
"gemini-2-5-pro-tts",
|
|
1853
|
+
] as const
|
|
1838
1854
|
export type DialogueProvider = typeof DIALOGUE_PROVIDERS[number]
|
|
1839
1855
|
|
|
1840
1856
|
/** The model a dialogue request runs on when its provider is missing or unknown: v3 dialogue (decided 2026-10-04 — v3 stays the default). */
|
|
@@ -3184,6 +3200,25 @@ export const VIDEO_REF_LIMITS_BY_PROVIDER: Record<
|
|
|
3184
3200
|
* `elements[].voice_url`, no standalone `voices`).
|
|
3185
3201
|
*/
|
|
3186
3202
|
audioBindsToCharacters?: true
|
|
3203
|
+
/**
|
|
3204
|
+
* How many of the `characters` may be VIDEO subjects — a character made
|
|
3205
|
+
* from a short clip of itself (`characterReferences[].videoUrl`) instead of
|
|
3206
|
+
* images. The clip counts as one of the `videos` too: the provider caps
|
|
3207
|
+
* reference videos and video subjects together. Absent ⇒ none (any other
|
|
3208
|
+
* model refuses a character video). Each clip's own length is
|
|
3209
|
+
* `VIDEO_CHARACTER_CLIP_DURATION_LIMITS`.
|
|
3210
|
+
*/
|
|
3211
|
+
videoCharacters?: number
|
|
3212
|
+
/**
|
|
3213
|
+
* The provider bills a video subject's clip as reference-video seconds
|
|
3214
|
+
* (its with-video rate on output + input), so the run is priced as a
|
|
3215
|
+
* reference-video run with the clip among its input seconds. Absent ⇒ a
|
|
3216
|
+
* video subject is priced like an image subject: it adds no input seconds
|
|
3217
|
+
* and does not move the run onto the reference-video rate (its clip still
|
|
3218
|
+
* counts toward the `videos` cap and the reference videos' total length).
|
|
3219
|
+
* Read through `videoCharacterClipsBilledAsReferenceVideo`.
|
|
3220
|
+
*/
|
|
3221
|
+
videoCharactersBillAsReferenceVideo?: true
|
|
3187
3222
|
} | undefined
|
|
3188
3223
|
> = {
|
|
3189
3224
|
// Seedance 2 — full multimodal caps.
|
|
@@ -3228,7 +3263,12 @@ export const VIDEO_REF_LIMITS_BY_PROVIDER: Record<
|
|
|
3228
3263
|
// (docs.kie.ai/market/kling/v4-reference-to-video). A wired start frame takes
|
|
3229
3264
|
// one of the 10 image slots. No `voices` key: that names Gemini Omni's
|
|
3230
3265
|
// preset personas, which Kling does not take.
|
|
3231
|
-
|
|
3266
|
+
// Of the 7 subjects, at most 3 may be video subjects (a 3-8 s clip each), and
|
|
3267
|
+
// reference videos plus video subjects are at most 5. A video subject is
|
|
3268
|
+
// priced as a reference video here until a measured run confirms how Kling
|
|
3269
|
+
// 4.0 bills one (decided 2026-11-14): Flash, measured, bills it as an image
|
|
3270
|
+
// subject (no flag below).
|
|
3271
|
+
"kling-4": { images: 10, videos: 5, audio: 3, characters: 7, materials: 15, videoCharacters: 3, videoCharactersBillAsReferenceVideo: true },
|
|
3232
3272
|
// Kling 4.0 Keyframes — the Image References are the in-between frames
|
|
3233
3273
|
// (`keyframe_urls`, max 8; the clip's duration lowers the cap, see
|
|
3234
3274
|
// kling4KeyframeFrameCap). No reference videos or audio.
|
|
@@ -3238,8 +3278,10 @@ export const VIDEO_REF_LIMITS_BY_PROVIDER: Record<
|
|
|
3238
3278
|
// standalone voices: reference audio clip N becomes wired character N's voice
|
|
3239
3279
|
// (`voice_url`) when that character is a subject (2-4 images), so at most one
|
|
3240
3280
|
// per character (7) and none takes a material seat. A wired start frame takes one of the 10 image slots when it
|
|
3241
|
-
// runs on this endpoint.
|
|
3242
|
-
|
|
3281
|
+
// runs on this endpoint. A video subject is billed like an image subject
|
|
3282
|
+
// (measured on a staging run, decided 2026-11-14): no
|
|
3283
|
+
// `videoCharactersBillAsReferenceVideo`.
|
|
3284
|
+
"kling-4-flash": { images: 10, videos: 5, audio: 7, characters: 7, materials: 15, audioBindsToCharacters: true, videoCharacters: 3 },
|
|
3243
3285
|
"grok-i2v": { images: 7 }, // backend kie/models.ts maxRefImages: 7
|
|
3244
3286
|
"happyhorse-ref2v": { images: 9 }, // backend kie/models.ts maxRefImages: 9
|
|
3245
3287
|
// VEO 3.x — REFERENCE_2_VIDEO path caps refs at 3 (kie/video.ts slices to
|
|
@@ -3320,6 +3362,56 @@ export const VIDEO_REF_VIDEO_DURATION_LIMITS: Record<string, RefVideoDurationLim
|
|
|
3320
3362
|
"kling-4-flash": { minSec: 2, maxSec: 20, maxTotalSec: 20 },
|
|
3321
3363
|
}
|
|
3322
3364
|
|
|
3365
|
+
/**
|
|
3366
|
+
* The length a character's own clip (`characterReferences[].videoUrl`, a video
|
|
3367
|
+
* subject) must have, per provider. Kling 4.0 and Kling 4.0 Flash: "one 3-8
|
|
3368
|
+
* second single-shot character video" (docs.kie.ai/market/kling/v4-reference-to-video,
|
|
3369
|
+
* v4-flash-reference-to-video, read 2026-11-13). The clip is also one of the
|
|
3370
|
+
* run's reference videos, so `VIDEO_REF_VIDEO_DURATION_LIMITS` (its per-clip
|
|
3371
|
+
* bound and the combined `maxTotalSec`) applies to it as well.
|
|
3372
|
+
*/
|
|
3373
|
+
export const VIDEO_CHARACTER_CLIP_DURATION_LIMITS: Record<string, { minSec: number; maxSec: number } | undefined> = {
|
|
3374
|
+
"kling-4": { minSec: 3, maxSec: 8 },
|
|
3375
|
+
"kling-4-flash": { minSec: 3, maxSec: 8 },
|
|
3376
|
+
}
|
|
3377
|
+
|
|
3378
|
+
/**
|
|
3379
|
+
* Validate probed character-clip durations against
|
|
3380
|
+
* {@link VIDEO_CHARACTER_CLIP_DURATION_LIMITS}. Unlike a reference video
|
|
3381
|
+
* ({@link checkRefVideoDurations}, which ignores a clip the probe could not
|
|
3382
|
+
* read and charges its worst case), a character's clip must be measured within
|
|
3383
|
+
* its bounds to be accepted (decided 2026-11-13): one the probe could not read
|
|
3384
|
+
* (NaN, ≤0) is refused too, as `character_video_unreadable`.
|
|
3385
|
+
*/
|
|
3386
|
+
export function checkCharacterClipDurations(
|
|
3387
|
+
provider: string,
|
|
3388
|
+
durationsSec: readonly number[],
|
|
3389
|
+
): { ok: true } | { ok: false; code: "video_too_long" | "character_video_unreadable"; message: string } {
|
|
3390
|
+
const limit = VIDEO_CHARACTER_CLIP_DURATION_LIMITS[provider]
|
|
3391
|
+
if (!limit) return { ok: true }
|
|
3392
|
+
if (durationsSec.some((d) => !Number.isFinite(d) || d <= 0)) {
|
|
3393
|
+
return {
|
|
3394
|
+
ok: false,
|
|
3395
|
+
code: "character_video_unreadable",
|
|
3396
|
+
message:
|
|
3397
|
+
`A character's video could not be measured, and it must be ${limit.minSec}–${limit.maxSec} seconds long. ` +
|
|
3398
|
+
"Check that the link is a playable video file anyone can open, and run again. Nothing was charged.",
|
|
3399
|
+
}
|
|
3400
|
+
}
|
|
3401
|
+
const offender = durationsSec.find((d) => d < limit.minSec || d > limit.maxSec)
|
|
3402
|
+
if (offender === undefined) return { ok: true }
|
|
3403
|
+
// The measured length, rounded UP to hundredths, so a clip just past the
|
|
3404
|
+
// bound never reads as inside it (8.03 s, never "8.0 s").
|
|
3405
|
+
const shown = (Math.ceil(offender * 100 - 1e-9) / 100).toFixed(2)
|
|
3406
|
+
return {
|
|
3407
|
+
ok: false,
|
|
3408
|
+
code: "video_too_long",
|
|
3409
|
+
message:
|
|
3410
|
+
`A character's video must be between ${limit.minSec} and ${limit.maxSec} seconds — one is ${shown} s. ` +
|
|
3411
|
+
`Use a single-shot clip of that length (a Trim Video node upstream works) and run again.`,
|
|
3412
|
+
}
|
|
3413
|
+
}
|
|
3414
|
+
|
|
3323
3415
|
/**
|
|
3324
3416
|
* Validate probed reference-video durations against the provider's limits.
|
|
3325
3417
|
*
|
|
@@ -4492,7 +4584,8 @@ export const VIDEO_DURATION_TIERS: Record<string, Array<{ maxSeconds: number; su
|
|
|
4492
4584
|
// The length tiers of the earlier Seedance Extend rows
|
|
4493
4585
|
// (`seedance-2-extend:<tier>:<res>`). No extend is priced on them any more —
|
|
4494
4586
|
// each Seedance extend provider has a row per second (`tailExtendCreditId`)
|
|
4495
|
-
// —
|
|
4587
|
+
// — and the rows themselves were removed (migration 648); the tiers stay,
|
|
4588
|
+
// unread, as part of this published table.
|
|
4496
4589
|
"seedance-2-extend": [
|
|
4497
4590
|
{ maxSeconds: 4, suffix: "4s" },
|
|
4498
4591
|
{ maxSeconds: 8, suffix: "8s" },
|
package/src/tts-capabilities.ts
CHANGED
|
@@ -13,7 +13,8 @@
|
|
|
13
13
|
* text-to-speech request that names one runs as the fallback model, so these
|
|
14
14
|
* helpers answer for the fallback there.
|
|
15
15
|
*/
|
|
16
|
-
import { MODEL_CATALOG, type TtsCapabilities, type TtsSettingLever } from "./model-catalog.js"
|
|
16
|
+
import { MODEL_CATALOG, type TtsCapabilities, type TtsDirectionSheet, type TtsSettingLever } from "./model-catalog.js"
|
|
17
|
+
import { REPLICA_CONSENT_ATTESTATION, type TtsNamedVoice, type TtsReplica } from "./gemini-tts.js"
|
|
17
18
|
|
|
18
19
|
/** Legacy provider ids that are another model under an old name. They have no catalog entry of their own. */
|
|
19
20
|
export const TTS_PROVIDER_ALIASES: Readonly<Record<string, string>> = {
|
|
@@ -87,3 +88,194 @@ export function ttsSupportsTimestamps(provider: string | undefined): boolean {
|
|
|
87
88
|
export function ttsLanguageCodes(provider: string | undefined): readonly string[] {
|
|
88
89
|
return getTtsCapabilities(provider).languages
|
|
89
90
|
}
|
|
91
|
+
|
|
92
|
+
/**
|
|
93
|
+
* The named voices the model speaks, or `undefined` when it speaks the
|
|
94
|
+
* ElevenLabs voices (premade, library, clones). See `TtsCapabilities.voices`.
|
|
95
|
+
*/
|
|
96
|
+
export function ttsNamedVoices(provider: string | undefined): readonly TtsNamedVoice[] | undefined {
|
|
97
|
+
return getTtsCapabilities(provider).voices
|
|
98
|
+
}
|
|
99
|
+
|
|
100
|
+
/**
|
|
101
|
+
* Whether `voice` is one the model can speak: on a model with named voices,
|
|
102
|
+
* one of them by its exact wire name; on any other model, any id (the
|
|
103
|
+
* ElevenLabs voice checks apply there, as before).
|
|
104
|
+
*/
|
|
105
|
+
export function ttsAcceptsVoice(provider: string | undefined, voice: string): boolean {
|
|
106
|
+
const named = ttsNamedVoices(provider)
|
|
107
|
+
return !named || named.some((v) => v.id === voice)
|
|
108
|
+
}
|
|
109
|
+
|
|
110
|
+
/**
|
|
111
|
+
* The voice a request on a named-voice model runs with: the one it names when
|
|
112
|
+
* the model speaks it, else the model's default. `undefined` on an ElevenLabs
|
|
113
|
+
* model (its own default applies downstream).
|
|
114
|
+
*/
|
|
115
|
+
export function ttsNamedVoiceFor(provider: string | undefined, voice: unknown): string | undefined {
|
|
116
|
+
const caps = getTtsCapabilities(provider)
|
|
117
|
+
if (!caps.voices) return undefined
|
|
118
|
+
return typeof voice === "string" && caps.voices.some((v) => v.id === voice) ? voice : caps.defaultVoice
|
|
119
|
+
}
|
|
120
|
+
|
|
121
|
+
/** The style direction the model takes (`TtsCapabilities.direction`), or `undefined` when it takes none. */
|
|
122
|
+
export function ttsDirection(provider: string | undefined): TtsDirectionSheet | undefined {
|
|
123
|
+
return getTtsCapabilities(provider).direction
|
|
124
|
+
}
|
|
125
|
+
|
|
126
|
+
/** The model writes its tone tags `<in angle brackets>` from a closed list (`TtsCapabilities.tagSyntax`). */
|
|
127
|
+
export function ttsAngleTags(provider: string | undefined): readonly string[] | undefined {
|
|
128
|
+
const caps = getTtsCapabilities(provider)
|
|
129
|
+
return caps.audioTags && caps.tagSyntax === "angle" ? (caps.tags ?? []) : undefined
|
|
130
|
+
}
|
|
131
|
+
|
|
132
|
+
/**
|
|
133
|
+
* The style direction of a request, as the model takes it: each field kept
|
|
134
|
+
* only when the model's sheet names it and the value is one it accepts (a
|
|
135
|
+
* listed choice, or free text within its cap — trimmed, an empty one
|
|
136
|
+
* dropped). What the provider funnel sends and the workflow engine passes on;
|
|
137
|
+
* the REST route rejects instead of dropping (`ttsDirectionProblem`).
|
|
138
|
+
*/
|
|
139
|
+
export interface TtsDirectionValues {
|
|
140
|
+
styleDirection?: string
|
|
141
|
+
pace?: string
|
|
142
|
+
accent?: string
|
|
143
|
+
audioProfile?: string
|
|
144
|
+
}
|
|
145
|
+
|
|
146
|
+
export function ttsDirectionFor(provider: string | undefined, values: Readonly<Record<string, unknown>>): TtsDirectionValues {
|
|
147
|
+
const sheet = ttsDirection(provider)
|
|
148
|
+
if (!sheet) return {}
|
|
149
|
+
const text = (value: unknown, max: number | undefined): string | undefined => {
|
|
150
|
+
if (max === undefined || typeof value !== "string") return undefined
|
|
151
|
+
const trimmed = value.trim()
|
|
152
|
+
return trimmed && trimmed.length <= max ? trimmed : undefined
|
|
153
|
+
}
|
|
154
|
+
const choice = (value: unknown, choices: readonly string[] | undefined): string | undefined =>
|
|
155
|
+
choices && typeof value === "string" && choices.includes(value) ? value : undefined
|
|
156
|
+
const out: TtsDirectionValues = {}
|
|
157
|
+
const style = sheet.styles ? choice(values.styleDirection, sheet.styles) : text(values.styleDirection, sheet.styleMaxChars)
|
|
158
|
+
if (style !== undefined) out.styleDirection = style
|
|
159
|
+
const pace = choice(values.pace, sheet.paces)
|
|
160
|
+
if (pace !== undefined) out.pace = pace
|
|
161
|
+
const accent = choice(values.accent, sheet.accents)
|
|
162
|
+
if (accent !== undefined) out.accent = accent
|
|
163
|
+
const profile = text(values.audioProfile, sheet.audioProfileMaxChars)
|
|
164
|
+
if (profile !== undefined) out.audioProfile = profile
|
|
165
|
+
return out
|
|
166
|
+
}
|
|
167
|
+
|
|
168
|
+
/**
|
|
169
|
+
* Why a request's style direction cannot run on the model, or `undefined`
|
|
170
|
+
* when it can. Only the fields the model takes are judged; on a model that
|
|
171
|
+
* takes none, or a field it does not take, the value is never sent (as the
|
|
172
|
+
* neighbour text on a model that does not stitch), so it is no problem.
|
|
173
|
+
*/
|
|
174
|
+
export function ttsDirectionProblem(provider: string | undefined, values: Readonly<Record<string, unknown>>): { field: keyof TtsDirectionValues; message: string } | undefined {
|
|
175
|
+
const sheet = ttsDirection(provider)
|
|
176
|
+
if (!sheet) return undefined
|
|
177
|
+
const textProblem = (field: keyof TtsDirectionValues, max: number | undefined) => {
|
|
178
|
+
const value = values[field]
|
|
179
|
+
if (max === undefined || value === undefined) return undefined
|
|
180
|
+
if (typeof value !== "string") return { field, message: `${field} must be a string` }
|
|
181
|
+
return value.trim().length > max ? { field, message: `${field} must be at most ${max} characters` } : undefined
|
|
182
|
+
}
|
|
183
|
+
const choiceProblem = (field: keyof TtsDirectionValues, choices: readonly string[] | undefined) => {
|
|
184
|
+
const value = values[field]
|
|
185
|
+
if (!choices || value === undefined || value === "") return undefined
|
|
186
|
+
return typeof value === "string" && choices.includes(value) ? undefined : { field, message: `${field} must be one of: ${choices.join(", ")}` }
|
|
187
|
+
}
|
|
188
|
+
return (
|
|
189
|
+
(sheet.styles ? choiceProblem("styleDirection", sheet.styles) : textProblem("styleDirection", sheet.styleMaxChars)) ??
|
|
190
|
+
choiceProblem("pace", sheet.paces) ??
|
|
191
|
+
choiceProblem("accent", sheet.accents) ??
|
|
192
|
+
textProblem("audioProfile", sheet.audioProfileMaxChars)
|
|
193
|
+
)
|
|
194
|
+
}
|
|
195
|
+
|
|
196
|
+
/** Every voice name a named-voice model speaks (all of them, across models). Built once from the catalog. */
|
|
197
|
+
let namedVoiceIds: ReadonlySet<string> | undefined
|
|
198
|
+
function anyModelsNamedVoice(voice: string): boolean {
|
|
199
|
+
namedVoiceIds ??= new Set(
|
|
200
|
+
Object.values(MODEL_CATALOG).flatMap((m) => (m.modes.includes("tts") ? (m.tts?.voices ?? []).map((v) => v.id) : [])),
|
|
201
|
+
)
|
|
202
|
+
return namedVoiceIds.has(voice)
|
|
203
|
+
}
|
|
204
|
+
|
|
205
|
+
/**
|
|
206
|
+
* The voice a text-to-speech run on `provider` speaks, when the model and the
|
|
207
|
+
* voice may have been chosen apart — a published app's Model card, a field
|
|
208
|
+
* mapping onto the model, an upstream Character's voice, a preset's voice with
|
|
209
|
+
* a model the caller named (decided 2026-11-13: the editor's model-switch rule,
|
|
210
|
+
* applied wherever the editor is not there to apply it):
|
|
211
|
+
* - a model with named voices (Gemini): the voice when it is one of them, its
|
|
212
|
+
* default voice for any other voice, and none (its default) for none;
|
|
213
|
+
* - an ElevenLabs model: none (the deployment's default) for a voice that only
|
|
214
|
+
* a named-voice model speaks (`Fola`, `Kore`…), else the voice as given.
|
|
215
|
+
* Returns `undefined` for "no voice named"; never a voice the model refuses.
|
|
216
|
+
*/
|
|
217
|
+
export function ttsVoiceForModel(provider: string | undefined, voice: unknown): string | undefined {
|
|
218
|
+
const named = typeof voice === "string" && voice.length > 0 ? voice : undefined
|
|
219
|
+
const caps = getTtsCapabilities(provider)
|
|
220
|
+
if (caps.voices) {
|
|
221
|
+
if (named === undefined) return undefined
|
|
222
|
+
return caps.voices.some((v) => v.id === named) ? named : caps.defaultVoice
|
|
223
|
+
}
|
|
224
|
+
return named !== undefined && anyModelsNamedVoice(named) ? undefined : named
|
|
225
|
+
}
|
|
226
|
+
|
|
227
|
+
/** The model clones a speaker's voice for one request (`TtsCapabilities.replica`: Gemini 3.8 Flash). */
|
|
228
|
+
export function ttsTakesReplica(provider: string | undefined): boolean {
|
|
229
|
+
return getTtsCapabilities(provider).replica === true
|
|
230
|
+
}
|
|
231
|
+
|
|
232
|
+
/** Every model that takes a one-off clone, in catalog order (for a refusal's sentence). */
|
|
233
|
+
export function replicaModels(): string[] {
|
|
234
|
+
return Object.values(MODEL_CATALOG).filter((m) => m.modes.includes("tts") && m.tts?.replica === true).map((m) => m.id)
|
|
235
|
+
}
|
|
236
|
+
|
|
237
|
+
/** Whether `value` is an https link (the only scheme a recording is fetched from). */
|
|
238
|
+
function isHttpsLink(value: unknown): value is string {
|
|
239
|
+
if (typeof value !== "string") return false
|
|
240
|
+
try {
|
|
241
|
+
return new URL(value).protocol === "https:"
|
|
242
|
+
} catch {
|
|
243
|
+
return false
|
|
244
|
+
}
|
|
245
|
+
}
|
|
246
|
+
|
|
247
|
+
/** A request's one-off clone as the platform reads it: two https links, or undefined when absent or not that shape. */
|
|
248
|
+
export function replicaOf(value: unknown): TtsReplica | undefined {
|
|
249
|
+
if (!value || typeof value !== "object") return undefined
|
|
250
|
+
const { sourceAudioUrl, consentAudioUrl } = value as Record<string, unknown>
|
|
251
|
+
return isHttpsLink(sourceAudioUrl) && isHttpsLink(consentAudioUrl) ? { sourceAudioUrl, consentAudioUrl } : undefined
|
|
252
|
+
}
|
|
253
|
+
|
|
254
|
+
/**
|
|
255
|
+
* Why a request's one-off clone (`replica`) cannot run on the model, or
|
|
256
|
+
* undefined (none asked, or one it takes). Refused, never dropped: a clone
|
|
257
|
+
* that is not sent is someone else's voice. The recordings themselves (WAV,
|
|
258
|
+
* size, length) are judged where they are fetched (backend
|
|
259
|
+
* `lib/voice-replica.ts`).
|
|
260
|
+
*/
|
|
261
|
+
export function ttsReplicaProblem(provider: string | undefined, value: unknown): string | undefined {
|
|
262
|
+
if (value === undefined || value === null) return undefined
|
|
263
|
+
if (!ttsTakesReplica(provider)) {
|
|
264
|
+
return `${provider ?? "This model"} does not clone a voice; a one-off clone runs on ${replicaModels().join(", ")}`
|
|
265
|
+
}
|
|
266
|
+
return replicaOf(value) ? undefined : "replica needs sourceAudioUrl and consentAudioUrl, each an https link to a WAV recording"
|
|
267
|
+
}
|
|
268
|
+
|
|
269
|
+
/**
|
|
270
|
+
* Why an API or MCP clone lacks its consent, or undefined: a `replica` asked
|
|
271
|
+
* must say `consent: true` — the caller's confirmation of
|
|
272
|
+
* `REPLICA_CONSENT_ATTESTATION`, what the editor's checkbox says (decided
|
|
273
|
+
* 2026-11-13). Refused before anything is read or held.
|
|
274
|
+
*/
|
|
275
|
+
export function replicaConsentProblem(value: unknown): string | undefined {
|
|
276
|
+
if (value === undefined || value === null) return undefined
|
|
277
|
+
return (value as { consent?: unknown }).consent === true
|
|
278
|
+
? undefined
|
|
279
|
+
: `replica.consent must be true: it confirms "${REPLICA_CONSENT_ATTESTATION}"`
|
|
280
|
+
}
|
|
281
|
+
|