@tanstack/ai-grok 0.14.10 → 0.15.0

package/dist/esm/video/video-provider-options.js

@@ -28,6 +28,15 @@ function parseGrokVideoSize(size) {
 	};
 }
 /**
+* Models that accept native 1080p on text-to-video and image-to-video.
+* Reference-to-video stays capped at 720p even on these models.
+*
+* @experimental Video generation is an experimental feature and may change.
+*/
+function isGrokVideoNative1080pModel(model) {
+	return model === "grok-imagine-video-1.5";
+}
+/**
 * Validate the `size` template for a given grok video model.
 *
 * @experimental Video generation is an experimental feature and may change.
@@ -37,6 +46,7 @@ function validateVideoSize(model, size) {
 	const parsed = parseGrokVideoSize(size);
 	if (!parsed || !GROK_VIDEO_ASPECT_RATIOS.includes(parsed.aspectRatio)) throw new Error(`Size "${size}" is not supported by model "${model}". Expected "aspectRatio" or "aspectRatio_resolution" (e.g. "16:9_720p") with aspect ratio one of: ${GROK_VIDEO_ASPECT_RATIOS.join(", ")}`);
 	if (parsed.resolution !== void 0 && !GROK_VIDEO_RESOLUTIONS.includes(parsed.resolution)) throw new Error(`Resolution "${parsed.resolution}" is not supported by model "${model}". Supported resolutions: ${GROK_VIDEO_RESOLUTIONS.join(", ")}`);
+	if (parsed.resolution === "1080p" && !isGrokVideoNative1080pModel(model)) throw new Error(`Resolution "1080p" is not supported by model "${model}". Use 'grok-imagine-video-1.5' for native 1080p text-to-video / image-to-video.`);
 }
 /**
 * Runtime duration table backing `availableDurations()` / `snapDuration()`.
@@ -70,24 +80,40 @@ function getGrokVideoDurationOptions(model) {
 	return GROK_VIDEO_DURATIONS[model];
 }
 /**
-* Models that only support image-to-video — a starting-frame image is
-* required and text-to-video is rejected by the Imagine API. Used by the
-* adapter to fail fast with a clear message instead of surfacing the raw
-* "Text-to-video is not supported for this model" 400.
+* Models that support reference-to-video inputs (`reference_images` /
+* `reference_audios`). The per-model provider-options map hides the fields
+* from other models at compile time; this backs the runtime gate for
+* untyped callers (e.g. deserialized JSON) so they get a clear error
+* instead of a raw API 400.
+*
+* @experimental Video generation is an experimental feature and may change.
+*/
+var GROK_VIDEO_REFERENCE_MODELS = /* @__PURE__ */ new Set(["grok-imagine-video-1.5"]);
+/**
+* True when the model accepts reference-to-video inputs.
+*
+* @experimental Video generation is an experimental feature and may change.
+*/
+function isGrokVideoReferenceModel(model) {
+	return GROK_VIDEO_REFERENCE_MODELS.has(model);
+}
+/**
+* Models that accept a source-video prompt part for `/v1/videos/edits`
+* and `/v1/videos/extensions`. xAI lists video input only on
+* grok-imagine-video (v1.0).
 *
 * @experimental Video generation is an experimental feature and may change.
 */
-var GROK_VIDEO_IMAGE_TO_VIDEO_ONLY = /* @__PURE__ */ new Set(["grok-imagine-video-1.5"]);
+var GROK_VIDEO_SOURCE_MODELS = /* @__PURE__ */ new Set(["grok-imagine-video"]);
 /**
-* True when the model only supports image-to-video (a starting frame is
-* required).
+* True when the model accepts edit / extend source-video jobs.
 *
 * @experimental Video generation is an experimental feature and may change.
 */
-function isImageToVideoOnlyModel(model) {
-	return GROK_VIDEO_IMAGE_TO_VIDEO_ONLY.has(model);
+function isGrokVideoSourceModel(model) {
+	return GROK_VIDEO_SOURCE_MODELS.has(model);
 }
 //#endregion
-export { GROK_VIDEO_DURATIONS, getGrokVideoDurationOptions, isImageToVideoOnlyModel, parseGrokVideoSize, validateVideoSize };
+export { GROK_VIDEO_DURATIONS, getGrokVideoDurationOptions, isGrokVideoNative1080pModel, isGrokVideoReferenceModel, isGrokVideoSourceModel, parseGrokVideoSize, validateVideoSize };
 
 //# sourceMappingURL=video-provider-options.js.map

package/dist/esm/video/video-provider-options.js.map

@@ -1 +1 @@
-{"version":3,"file":"video-provider-options.js","names":[],"sources":["../../../src/video/video-provider-options.ts"],"sourcesContent":["/**\n * Grok Video Generation Provider Options (xAI Imagine API)\n *\n * Based on https://docs.x.ai/docs/guides/video-generations\n *\n * @experimental Video generation is an experimental feature and may change.\n */\n\nimport type { DurationOptions } from '@tanstack/ai/adapters'\nimport type { GrokVideoModel } from '../model-meta'\n\n/**\n * Aspect ratios accepted by the grok-imagine video models.\n *\n * Note: this is a narrower set than the grok-imagine image models — the\n * video endpoint rejects the phone-screen ratios ('9:19.5', '9:20', …) and\n * 'auto'.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoAspectRatio =\n  | '1:1'\n  | '16:9'\n  | '9:16'\n  | '4:3'\n  | '3:4'\n  | '3:2'\n  | '2:3'\n\n/**\n * Resolution tiers for the grok-imagine video models.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoResolution = '480p' | '720p' | '1080p'\n\n/**\n * Size strings for grok-imagine video models. The Imagine API is\n * aspect-ratio based rather than pixel-size based; like the grok-imagine\n * image models, the generic `size` option uses an\n * `aspectRatio_resolution` template (\"16:9_720p\") — the resolution suffix\n * is optional (\"16:9\" uses the API default).\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoSize =\n  | GrokVideoAspectRatio\n  | `${GrokVideoAspectRatio}_${GrokVideoResolution}`\n\nconst GROK_VIDEO_ASPECT_RATIOS: ReadonlyArray<string> = [\n  '1:1',\n  '16:9',\n  '9:16',\n  '4:3',\n  '3:4',\n  '3:2',\n  '2:3',\n]\n\nconst GROK_VIDEO_RESOLUTIONS: ReadonlyArray<string> = ['480p', '720p', '1080p']\n\n/**\n * Video duration limits enforced by the Imagine API (seconds).\n */\nexport const GROK_VIDEO_MIN_DURATION = 1\nexport const GROK_VIDEO_MAX_DURATION = 15\n\n/**\n * Parses a grok video size string into its components.\n * Format: \"aspectRatio\" or \"aspectRatio_resolution\",\n * e.g. \"16:9_720p\" → { aspectRatio: \"16:9\", resolution: \"720p\" }.\n * Returns undefined when the string doesn't match the template.\n */\nexport function parseGrokVideoSize(\n  size: string,\n): { aspectRatio: string; resolution?: string } | undefined {\n  const match = size.match(/^([\\d.]+:[\\d.]+)(?:_(.+))?$/)\n  const [, aspectRatio, resolution] = match ?? []\n  if (aspectRatio === undefined) return undefined\n  return { aspectRatio, ...(resolution !== undefined && { resolution }) }\n}\n\n/**\n * Validate the `size` template for a given grok video model.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function validateVideoSize(\n  model: string,\n  size?: string,\n): asserts size is GrokVideoSize | undefined {\n  if (size === undefined) return\n  const parsed = parseGrokVideoSize(size)\n  if (!parsed || !GROK_VIDEO_ASPECT_RATIOS.includes(parsed.aspectRatio)) {\n    throw new Error(\n      `Size \"${size}\" is not supported by model \"${model}\". Expected ` +\n        `\"aspectRatio\" or \"aspectRatio_resolution\" (e.g. \"16:9_720p\") with ` +\n        `aspect ratio one of: ${GROK_VIDEO_ASPECT_RATIOS.join(', ')}`,\n    )\n  }\n  if (\n    parsed.resolution !== undefined &&\n    !GROK_VIDEO_RESOLUTIONS.includes(parsed.resolution)\n  ) {\n    throw new Error(\n      `Resolution \"${parsed.resolution}\" is not supported by model \"${model}\". ` +\n        `Supported resolutions: ${GROK_VIDEO_RESOLUTIONS.join(', ')}`,\n    )\n  }\n}\n\n/**\n * Per-model duration type. The Imagine API accepts any integer second in the\n * 1–15 range, so this is a continuous range expressed as `number` (a literal\n * union can't represent it). `snapDuration()` coerces a raw seconds value into\n * the valid range at runtime.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelDurationByName = {\n  'grok-imagine-video': number\n  'grok-imagine-video-1.5': number\n}\n\n/**\n * Runtime duration table backing `availableDurations()` / `snapDuration()`.\n * Both grok-imagine video models accept the same continuous 1–15 integer-second\n * range.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport const GROK_VIDEO_DURATIONS: {\n  readonly [TModel in GrokVideoModel]: DurationOptions<\n    GrokVideoModelDurationByName[TModel]\n  >\n} = {\n  'grok-imagine-video': {\n    kind: 'range',\n    min: GROK_VIDEO_MIN_DURATION,\n    max: GROK_VIDEO_MAX_DURATION,\n    step: 1,\n    unit: 'seconds',\n  },\n  'grok-imagine-video-1.5': {\n    kind: 'range',\n    min: GROK_VIDEO_MIN_DURATION,\n    max: GROK_VIDEO_MAX_DURATION,\n    step: 1,\n    unit: 'seconds',\n  },\n}\n\n/**\n * Look up the duration options for a grok video model.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function getGrokVideoDurationOptions<TModel extends GrokVideoModel>(\n  model: TModel,\n): DurationOptions<GrokVideoModelDurationByName[TModel]> {\n  return GROK_VIDEO_DURATIONS[model]\n}\n\n/**\n * Provider-specific options for grok video generation. These map directly\n * onto the Imagine API request body and take precedence over the generic\n * `size` / `duration` options when both are provided.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport interface GrokVideoProviderOptions {\n  /**\n   * Output aspect ratio.\n   */\n  aspect_ratio?: GrokVideoAspectRatio\n\n  /**\n   * Output resolution tier.\n   */\n  resolution?: GrokVideoResolution\n\n  /**\n   * Video duration in integer seconds (1–15).\n   */\n  duration?: number\n}\n\n/**\n * Type-only map from model name to its specific provider options.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelProviderOptionsByName = {\n  'grok-imagine-video': GrokVideoProviderOptions\n  'grok-imagine-video-1.5': GrokVideoProviderOptions\n}\n\n/**\n * Type-only map from model name to its supported `size` strings.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelSizeByName = {\n  'grok-imagine-video': GrokVideoSize\n  'grok-imagine-video-1.5': GrokVideoSize\n}\n\n/**\n * Type-only map from model name to the non-text prompt modalities it accepts.\n * Both models accept an `image` prompt part as the starting frame:\n * `grok-imagine-video` (v1.0) does text-to-video and image-to-video, while\n * `grok-imagine-video-1.5` is image-to-video only (the image is required).\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelInputModalitiesByName = {\n  'grok-imagine-video': readonly ['image']\n  'grok-imagine-video-1.5': readonly ['image']\n}\n\n/**\n * Models that only support image-to-video — a starting-frame image is\n * required and text-to-video is rejected by the Imagine API. Used by the\n * adapter to fail fast with a clear message instead of surfacing the raw\n * \"Text-to-video is not supported for this model\" 400.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nconst GROK_VIDEO_IMAGE_TO_VIDEO_ONLY: ReadonlySet<string> = new Set([\n  'grok-imagine-video-1.5',\n])\n\n/**\n * True when the model only supports image-to-video (a starting frame is\n * required).\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function isImageToVideoOnlyModel(model: string): boolean {\n  return GROK_VIDEO_IMAGE_TO_VIDEO_ONLY.has(model)\n}\n"],"mappings":";AAiDA,IAAM,2BAAkD;CACtD;CACA;CACA;CACA;CACA;CACA;CACA;AACF;AAEA,IAAM,yBAAgD;CAAC;CAAQ;CAAQ;AAAO;;;;;;;AAc9E,SAAgB,mBACd,MAC0D;CAE1D,MAAM,GAAG,aAAa,cADR,KAAK,MAAM,6BACW,KAAS,CAAC;CAC9C,IAAI,gBAAgB,KAAA,GAAW,OAAO,KAAA;CACtC,OAAO;EAAE;EAAa,GAAI,eAAe,KAAA,KAAa,EAAE,WAAW;CAAG;AACxE;;;;;;AAOA,SAAgB,kBACd,OACA,MAC2C;CAC3C,IAAI,SAAS,KAAA,GAAW;CACxB,MAAM,SAAS,mBAAmB,IAAI;CACtC,IAAI,CAAC,UAAU,CAAC,yBAAyB,SAAS,OAAO,WAAW,GAClE,MAAM,IAAI,MACR,SAAS,KAAK,+BAA+B,MAAM,qGAEzB,yBAAyB,KAAK,IAAI,GAC9D;CAEF,IACE,OAAO,eAAe,KAAA,KACtB,CAAC,uBAAuB,SAAS,OAAO,UAAU,GAElD,MAAM,IAAI,MACR,eAAe,OAAO,WAAW,+BAA+B,MAAM,4BAC1C,uBAAuB,KAAK,IAAI,GAC9D;AAEJ;;;;;;;;AAsBA,IAAa,uBAIT;CACF,sBAAsB;EACpB,MAAM;EACN,KAAA;EACA,KAAA;EACA,MAAM;EACN,MAAM;CACR;CACA,0BAA0B;EACxB,MAAM;EACN,KAAA;EACA,KAAA;EACA,MAAM;EACN,MAAM;CACR;AACF;;;;;;AAOA,SAAgB,4BACd,OACuD;CACvD,OAAO,qBAAqB;AAC9B;;;;;;;;;AAmEA,IAAM,iDAAsD,IAAI,IAAI,CAClE,wBACF,CAAC;;;;;;;AAQD,SAAgB,wBAAwB,OAAwB;CAC9D,OAAO,+BAA+B,IAAI,KAAK;AACjD"}
+{"version":3,"file":"video-provider-options.js","names":[],"sources":["../../../src/video/video-provider-options.ts"],"sourcesContent":["/**\n * Grok Video Generation Provider Options (xAI Imagine API)\n *\n * Based on https://docs.x.ai/developers/model-capabilities/video/generation\n * (plus the image-to-video, reference-to-video, editing, and extension pages\n * under the same section).\n *\n * @experimental Video generation is an experimental feature and may change.\n */\n\nimport type { DurationOptions } from '@tanstack/ai/adapters'\nimport type { GrokVideoModel } from '../model-meta'\n\n/**\n * Aspect ratios accepted by the grok-imagine video models.\n *\n * Note: this is a narrower set than the grok-imagine image models — the\n * video endpoint rejects the phone-screen ratios ('9:19.5', '9:20', …) and\n * 'auto'.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoAspectRatio =\n  | '1:1'\n  | '16:9'\n  | '9:16'\n  | '4:3'\n  | '3:4'\n  | '3:2'\n  | '2:3'\n\n/**\n * Resolution tiers for the grok-imagine video models.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoResolution = '480p' | '720p' | '1080p'\n\n/**\n * Resolutions accepted by grok-imagine-video (v1.0). Native 1080p is a\n * grok-imagine-video-1.5 generation feature.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoResolutionV1 = '480p' | '720p'\n\n/**\n * Size strings for grok-imagine video models. The Imagine API is\n * aspect-ratio based rather than pixel-size based; like the grok-imagine\n * image models, the generic `size` option uses an\n * `aspectRatio_resolution` template (\"16:9_720p\") — the resolution suffix\n * is optional (\"16:9\" uses the API default).\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoSize =\n  | GrokVideoAspectRatio\n  | `${GrokVideoAspectRatio}_${GrokVideoResolution}`\n\n/**\n * Size strings for grok-imagine-video (v1.0) — 1080p is not in the type.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoSizeV1 =\n  | GrokVideoAspectRatio\n  | `${GrokVideoAspectRatio}_${GrokVideoResolutionV1}`\n\nconst GROK_VIDEO_ASPECT_RATIOS: ReadonlyArray<string> = [\n  '1:1',\n  '16:9',\n  '9:16',\n  '4:3',\n  '3:4',\n  '3:2',\n  '2:3',\n]\n\nconst GROK_VIDEO_RESOLUTIONS: ReadonlyArray<string> = ['480p', '720p', '1080p']\n\n/**\n * Video duration limits enforced by the Imagine API (seconds).\n */\nexport const GROK_VIDEO_MIN_DURATION = 1\nexport const GROK_VIDEO_MAX_DURATION = 15\n\n/**\n * Parses a grok video size string into its components.\n * Format: \"aspectRatio\" or \"aspectRatio_resolution\",\n * e.g. \"16:9_720p\" → { aspectRatio: \"16:9\", resolution: \"720p\" }.\n * Returns undefined when the string doesn't match the template.\n */\nexport function parseGrokVideoSize(\n  size: string,\n): { aspectRatio: string; resolution?: string } | undefined {\n  const match = size.match(/^([\\d.]+:[\\d.]+)(?:_(.+))?$/)\n  const [, aspectRatio, resolution] = match ?? []\n  if (aspectRatio === undefined) return undefined\n  return { aspectRatio, ...(resolution !== undefined && { resolution }) }\n}\n\n/**\n * Models that accept native 1080p on text-to-video and image-to-video.\n * Reference-to-video stays capped at 720p even on these models.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function isGrokVideoNative1080pModel(model: string): boolean {\n  return model === 'grok-imagine-video-1.5'\n}\n\n/**\n * Validate the `size` template for a given grok video model.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function validateVideoSize(\n  model: string,\n  size?: string,\n): asserts size is GrokVideoSize | undefined {\n  if (size === undefined) return\n  const parsed = parseGrokVideoSize(size)\n  if (!parsed || !GROK_VIDEO_ASPECT_RATIOS.includes(parsed.aspectRatio)) {\n    throw new Error(\n      `Size \"${size}\" is not supported by model \"${model}\". Expected ` +\n        `\"aspectRatio\" or \"aspectRatio_resolution\" (e.g. \"16:9_720p\") with ` +\n        `aspect ratio one of: ${GROK_VIDEO_ASPECT_RATIOS.join(', ')}`,\n    )\n  }\n  if (\n    parsed.resolution !== undefined &&\n    !GROK_VIDEO_RESOLUTIONS.includes(parsed.resolution)\n  ) {\n    throw new Error(\n      `Resolution \"${parsed.resolution}\" is not supported by model \"${model}\". ` +\n        `Supported resolutions: ${GROK_VIDEO_RESOLUTIONS.join(', ')}`,\n    )\n  }\n  if (parsed.resolution === '1080p' && !isGrokVideoNative1080pModel(model)) {\n    throw new Error(\n      `Resolution \"1080p\" is not supported by model \"${model}\". ` +\n        `Use 'grok-imagine-video-1.5' for native 1080p text-to-video / image-to-video.`,\n    )\n  }\n}\n\n/**\n * Per-model duration type. The Imagine API accepts any integer second in the\n * 1–15 range, so this is a continuous range expressed as `number` (a literal\n * union can't represent it). `snapDuration()` coerces a raw seconds value into\n * the valid range at runtime.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelDurationByName = {\n  'grok-imagine-video': number\n  'grok-imagine-video-1.5': number\n}\n\n/**\n * Runtime duration table backing `availableDurations()` / `snapDuration()`.\n * Both grok-imagine video models accept the same continuous 1–15 integer-second\n * range.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport const GROK_VIDEO_DURATIONS: {\n  readonly [TModel in GrokVideoModel]: DurationOptions<\n    GrokVideoModelDurationByName[TModel]\n  >\n} = {\n  'grok-imagine-video': {\n    kind: 'range',\n    min: GROK_VIDEO_MIN_DURATION,\n    max: GROK_VIDEO_MAX_DURATION,\n    step: 1,\n    unit: 'seconds',\n  },\n  'grok-imagine-video-1.5': {\n    kind: 'range',\n    min: GROK_VIDEO_MIN_DURATION,\n    max: GROK_VIDEO_MAX_DURATION,\n    step: 1,\n    unit: 'seconds',\n  },\n}\n\n/**\n * Look up the duration options for a grok video model.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function getGrokVideoDurationOptions<TModel extends GrokVideoModel>(\n  model: TModel,\n): DurationOptions<GrokVideoModelDurationByName[TModel]> {\n  return GROK_VIDEO_DURATIONS[model]\n}\n\n/**\n * Request mode for a source-video job. `'edit'` posts to `/v1/videos/edits`\n * (modify the source clip in place); `'extend'` posts to\n * `/v1/videos/extensions` (continue the source clip — `duration` is the\n * length of the **added tail**, not the total). Both require exactly one\n * video prompt part carrying the source clip, and both are\n * `grok-imagine-video` (v1.0) only.\n *\n * Output geometry (aspect ratio / resolution) is inherited from the source\n * clip in both modes, capped at 720p, and edit outputs also inherit the\n * source length — the adapter rejects `size`, `aspect_ratio`, `resolution`,\n * and (in edit mode) `duration` rather than sending fields the API ignores.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoMode = 'edit' | 'extend'\n\n/**\n * Provider options shared by both grok-imagine video models. These map\n * directly onto the Imagine API request body and take precedence over the\n * generic `size` / `duration` options when both are provided.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport interface GrokVideoBaseProviderOptions {\n  /**\n   * Output aspect ratio. Generation only — edit / extend outputs inherit\n   * the source clip's geometry and the adapter rejects this in those modes.\n   */\n  aspect_ratio?: GrokVideoAspectRatio\n\n  /**\n   * Output resolution tier. Generation only — edit / extend outputs inherit\n   * the source clip's geometry and the adapter rejects this in those modes.\n   * `1080p` is grok-imagine-video-1.5 generation only; reference-to-video\n   * is capped at 720p.\n   */\n  resolution?: GrokVideoResolution\n\n  /**\n   * Video duration in integer seconds (1–15). In `'extend'` mode this is\n   * the length of the added tail only, not the total output length. Not\n   * valid in `'edit'` mode (the output inherits the source clip's length).\n   */\n  duration?: number\n}\n\n/**\n * Provider options for grok-imagine-video (v1.0), which is the only model\n * that accepts a source-video edit / extend job.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport interface GrokVideoSourceProviderOptions extends GrokVideoBaseProviderOptions {\n  /**\n   * Selects the request mode for a source-video prompt part: `'edit'`\n   * (`/v1/videos/edits`) or `'extend'` (`/v1/videos/extensions`). Required\n   * when the prompt carries a video part; not valid without one. Omit for\n   * plain generation (`/v1/videos/generations`). grok-imagine-video only.\n   */\n  mode?: GrokVideoMode\n}\n\n/**\n * Provider options for grok-imagine-video-1.5, which adds the\n * reference-to-video inputs on top of the shared options.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport interface GrokVideoProviderOptions extends GrokVideoBaseProviderOptions {\n  /**\n   * Reference images for reference-to-video generation (output capped at\n   * 720p). Usually populated from image prompt parts with\n   * `metadata.role: 'reference'` (or `'character'`); set explicitly to\n   * replace the part-derived list. Reference images are addressed from the\n   * prompt text as `<IMAGE_0>`, `<IMAGE_1>`, … in request order, and do not\n   * lock the first frame.\n   */\n  reference_images?: Array<{ url: string }>\n\n  /**\n   * Preset TTS voices to reference for generated speech (max 3). Voice ids\n   * come from the xAI TTS voice roster (e.g. 'eve', 'rex') or a custom\n   * voice id, and are addressed from the prompt text as `<AUDIO_0>`,\n   * `<AUDIO_1>`, `<AUDIO_2>`.\n   */\n  reference_audios?: Array<{ voice_id: string }>\n}\n\n/**\n * Widest option surface. Used when `modelOptions` arrives as deserialized\n * JSON and the adapter must validate fields the per-model map already\n * hides at compile time.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoRuntimeOptions = GrokVideoSourceProviderOptions &\n  GrokVideoProviderOptions\n\n/**\n * Maximum reference voices accepted by the Imagine video endpoint.\n */\nexport const GROK_VIDEO_MAX_REFERENCE_AUDIOS = 3\n\n/**\n * Maximum reference images accepted by the Imagine video endpoint.\n */\nexport const GROK_VIDEO_MAX_REFERENCE_IMAGES = 7\n\n/**\n * Model names whose per-model options declare the reference fields. Keeps\n * the runtime set below provably in sync with\n * {@link GrokVideoModelProviderOptionsByName} — a typo or a new\n * reference-capable model missing from the set is a compile error.\n */\ntype GrokVideoReferenceModel = {\n  [TModel in GrokVideoModel]: 'reference_images' extends keyof GrokVideoModelProviderOptionsByName[TModel]\n    ? TModel\n    : never\n}[GrokVideoModel]\n\n/**\n * Models that support reference-to-video inputs (`reference_images` /\n * `reference_audios`). The per-model provider-options map hides the fields\n * from other models at compile time; this backs the runtime gate for\n * untyped callers (e.g. deserialized JSON) so they get a clear error\n * instead of a raw API 400.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nconst GROK_VIDEO_REFERENCE_MODELS: ReadonlySet<string> =\n  new Set<GrokVideoReferenceModel>(['grok-imagine-video-1.5'])\n\n/**\n * True when the model accepts reference-to-video inputs.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function isGrokVideoReferenceModel(model: string): boolean {\n  return GROK_VIDEO_REFERENCE_MODELS.has(model)\n}\n\n/**\n * Model names whose per-model options declare `mode`. Same\n * provably-in-sync construction as {@link GrokVideoReferenceModel}.\n */\ntype GrokVideoSourceModel = {\n  [TModel in GrokVideoModel]: 'mode' extends keyof GrokVideoModelProviderOptionsByName[TModel]\n    ? TModel\n    : never\n}[GrokVideoModel]\n\n/**\n * Models that accept a source-video prompt part for `/v1/videos/edits`\n * and `/v1/videos/extensions`. xAI lists video input only on\n * grok-imagine-video (v1.0).\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nconst GROK_VIDEO_SOURCE_MODELS: ReadonlySet<string> =\n  new Set<GrokVideoSourceModel>(['grok-imagine-video'])\n\n/**\n * True when the model accepts edit / extend source-video jobs.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport function isGrokVideoSourceModel(model: string): boolean {\n  return GROK_VIDEO_SOURCE_MODELS.has(model)\n}\n\n/**\n * Type-only map from model name to its specific provider options. Only\n * grok-imagine-video-1.5 exposes the reference-to-video fields. Only\n * grok-imagine-video (v1.0) exposes `mode` for edit / extend.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelProviderOptionsByName = {\n  'grok-imagine-video': GrokVideoSourceProviderOptions\n  'grok-imagine-video-1.5': GrokVideoProviderOptions\n}\n\n/**\n * Type-only map from model name to its supported `size` strings.\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelSizeByName = {\n  'grok-imagine-video': GrokVideoSizeV1\n  'grok-imagine-video-1.5': GrokVideoSize\n}\n\n/**\n * Type-only map from model name to the non-text prompt modalities it accepts.\n * Both models support text-to-video and accept an optional `image` prompt\n * part as the starting frame; image parts with `metadata.role: 'reference'`\n * or `'character'` become `reference_images` (grok-imagine-video-1.5 only).\n * A `video` prompt part carries the source clip for edit / extension mode\n * on grok-imagine-video only (`modelOptions.mode: 'edit' | 'extend'`).\n *\n * @experimental Video generation is an experimental feature and may change.\n */\nexport type GrokVideoModelInputModalitiesByName = {\n  'grok-imagine-video': readonly ['image', 'video']\n  'grok-imagine-video-1.5': readonly ['image']\n}\n"],"mappings":";AAoEA,IAAM,2BAAkD;CACtD;CACA;CACA;CACA;CACA;CACA;CACA;AACF;AAEA,IAAM,yBAAgD;CAAC;CAAQ;CAAQ;AAAO;;;;;;;AAc9E,SAAgB,mBACd,MAC0D;CAE1D,MAAM,GAAG,aAAa,cADR,KAAK,MAAM,6BACW,KAAS,CAAC;CAC9C,IAAI,gBAAgB,KAAA,GAAW,OAAO,KAAA;CACtC,OAAO;EAAE;EAAa,GAAI,eAAe,KAAA,KAAa,EAAE,WAAW;CAAG;AACxE;;;;;;;AAQA,SAAgB,4BAA4B,OAAwB;CAClE,OAAO,UAAU;AACnB;;;;;;AAOA,SAAgB,kBACd,OACA,MAC2C;CAC3C,IAAI,SAAS,KAAA,GAAW;CACxB,MAAM,SAAS,mBAAmB,IAAI;CACtC,IAAI,CAAC,UAAU,CAAC,yBAAyB,SAAS,OAAO,WAAW,GAClE,MAAM,IAAI,MACR,SAAS,KAAK,+BAA+B,MAAM,qGAEzB,yBAAyB,KAAK,IAAI,GAC9D;CAEF,IACE,OAAO,eAAe,KAAA,KACtB,CAAC,uBAAuB,SAAS,OAAO,UAAU,GAElD,MAAM,IAAI,MACR,eAAe,OAAO,WAAW,+BAA+B,MAAM,4BAC1C,uBAAuB,KAAK,IAAI,GAC9D;CAEF,IAAI,OAAO,eAAe,WAAW,CAAC,4BAA4B,KAAK,GACrE,MAAM,IAAI,MACR,iDAAiD,MAAM,iFAEzD;AAEJ;;;;;;;;AAsBA,IAAa,uBAIT;CACF,sBAAsB;EACpB,MAAM;EACN,KAAA;EACA,KAAA;EACA,MAAM;EACN,MAAM;CACR;CACA,0BAA0B;EACxB,MAAM;EACN,KAAA;EACA,KAAA;EACA,MAAM;EACN,MAAM;CACR;AACF;;;;;;AAOA,SAAgB,4BACd,OACuD;CACvD,OAAO,qBAAqB;AAC9B;;;;;;;;;;AAoIA,IAAM,8CACJ,IAAI,IAA6B,CAAC,wBAAwB,CAAC;;;;;;AAO7D,SAAgB,0BAA0B,OAAwB;CAChE,OAAO,4BAA4B,IAAI,KAAK;AAC9C;;;;;;;;AAmBA,IAAM,2CACJ,IAAI,IAA0B,CAAC,oBAAoB,CAAC;;;;;;AAOtD,SAAgB,uBAAuB,OAAwB;CAC7D,OAAO,yBAAyB,IAAI,KAAK;AAC3C"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tanstack/ai-grok",
-  "version": "0.14.10",
+  "version": "0.15.0",
   "description": "xAI Grok adapter for TanStack AI chat, image generation, realtime, and structured outputs.",
   "author": "Tanner Linsley",
   "license": "MIT",
@@ -50,17 +50,17 @@
   ],
   "dependencies": {
     "openai": "^6.41.0",
-    "@tanstack/ai-utils": "0.4.0",
-    "@tanstack/openai-base": "0.9.10"
+    "@tanstack/ai-utils": "^0.4.0",
+    "@tanstack/openai-base": "^0.9.13"
   },
   "devDependencies": {
-    "@vitest/coverage-v8": "4.0.14",
-    "vite": "^8.1.4",
-    "@tanstack/ai": "0.43.0"
+    "@vitest/coverage-v8": "4.1.10",
+    "vite": "^8.2.1",
+    "@tanstack/ai": "0.45.0"
   },
   "peerDependencies": {
     "zod": "^4.0.0",
-    "@tanstack/ai": "^0.43.0"
+    "@tanstack/ai": "^0.45.0"
   },
   "scripts": {
     "build": "vite build",

package/src/adapters/image.ts

@@ -129,7 +129,8 @@ export class GrokImageAdapter<
         throw new Error(
           `grok: model "${model}" does not support image prompt parts. ` +
             `Image-conditioned generation requires an Imagine API model ` +
-            `('grok-imagine-image' or 'grok-imagine-image-quality').`,
+            `('grok-imagine-image', 'grok-imagine-image-2.0' or ` +
+            `'grok-imagine-image-quality').`,
         )
       }
       return await this.editImages(options, resolved)