@henjicc/ai-sdk 0.9.0 → 0.10.0

package/dist/catalog/shared/subjectElements.js

@@ -0,0 +1,93 @@
+/**
+ * 元素（角色 / 物体主体）参数的值结构与校验，供 Kling 4.0 在 Fal 与 KIE 上共用。
+ *
+ * 值结构：`[{ id?, name?, kind: 'image' | 'video', images: string[], videos: string[], videoSeconds? }]`，
+ * `videoSeconds` 是上传时读到的视频时长，只用于估价，不发给平台。
+ * 图片元素第 1 张是正面主图、其余是其他角度；视频元素只用一段 3–8 秒单镜头视频。
+ * Fal 在提示词里按顺序写 `@Element1`，KIE 写元素名 `@名称`，所以名称只对 KIE 有意义。
+ * 媒体值是本地受管路径，由公共上传预处理按字段名（`frontal_image_url`、`image_urls`、`video_url`）上传替换。
+ */
+/** 两家对元素的共同上限：最多 7 个，其中视频元素最多 3 个；图片元素 2–4 张图。 */
+export const SUBJECT_ELEMENT_LIMITS = { maxItems: 7, maxVideoItems: 3, minImages: 2, maxImages: 4 };
+/** 视频元素时长读不到时按官方上限 8 秒估价，宁可估高，避免付费授权金额偏低。 */
+const SUBJECT_ELEMENT_VIDEO_MAX_SECONDS = 8;
+const KIE_NAME_PATTERN = /^[一-龥A-Za-z0-9_]{1,20}$/;
+const KIE_RESERVED_NAME = /^(?:(?:image|video|keyframe|object|audio)_\d+|first_frame|last_frame)$/i;
+function isObject(value) {
+    return typeof value === 'object' && value !== null && !Array.isArray(value);
+}
+function sources(value) {
+    if (typeof value === 'string')
+        return value.trim() ? [value.trim()] : [];
+    return Array.isArray(value) ? value.filter((item) => typeof item === 'string' && item.trim().length > 0).map(item => item.trim()) : [];
+}
+export function readSubjectElements(value) {
+    if (!Array.isArray(value))
+        return [];
+    return value.filter(isObject).map((element) => {
+        const images = sources(element.images);
+        const videos = sources(element.videos);
+        const kind = element.kind === 'video' || (element.kind !== 'image' && videos.length > 0 && images.length === 0) ? 'video' : 'image';
+        return {
+            name: typeof element.name === 'string' ? element.name.trim() : '',
+            kind,
+            images: kind === 'image' ? images : [],
+            ...(kind === 'video' && videos[0] ? { video: videos[0] } : {}),
+            ...(kind === 'video' && typeof element.videoSeconds === 'number' && element.videoSeconds > 0 ? { videoSeconds: element.videoSeconds } : {}),
+        };
+    });
+}
+/** 视频元素按视频输入计费的总秒数：用上传时读到的时长，读不到按上限计。 */
+export function subjectElementVideoSeconds(value) {
+    return readSubjectElements(value)
+        .filter(element => element.kind === 'video')
+        .reduce((total, element) => total + Math.min(SUBJECT_ELEMENT_VIDEO_MAX_SECONDS, element.videoSeconds ?? SUBJECT_ELEMENT_VIDEO_MAX_SECONDS), 0);
+}
+function validate(elements, modelName) {
+    const limits = SUBJECT_ELEMENT_LIMITS;
+    if (elements.length > limits.maxItems)
+        throw new Error(`${modelName} 最多 ${limits.maxItems} 个元素`);
+    if (elements.filter(element => element.kind === 'video').length > limits.maxVideoItems)
+        throw new Error(`${modelName} 最多 ${limits.maxVideoItems} 个视频元素`);
+    elements.forEach((element, index) => {
+        const label = `${modelName} 第 ${index + 1} 个元素`;
+        if (element.kind === 'video') {
+            if (!element.video)
+                throw new Error(`${label}还没有上传视频`);
+        }
+        else if (element.images.length < limits.minImages || element.images.length > limits.maxImages) {
+            throw new Error(`${label}需要 ${limits.minImages}–${limits.maxImages} 张图（第 1 张为正面主图）`);
+        }
+    });
+}
+/** Fal：图片元素拆成正面主图与其他角度，视频元素只给视频；提示词按顺序引用 `@Element1`。 */
+export function buildFalSubjectElements(value, modelName) {
+    const elements = readSubjectElements(value);
+    validate(elements, modelName);
+    return elements.map((element) => element.kind === 'video'
+        ? { video_url: element.video }
+        : { frontal_image_url: element.images[0], reference_image_urls: element.images.slice(1) });
+}
+/**
+ * KIE：每个元素必须有唯一名称，提示词用 `@名称` 引用。
+ * Flash 端点用 `type` 区分图片与视频元素，标准版没有这个字段。
+ */
+export function buildKieSubjectElements(value, modelName, options) {
+    const elements = readSubjectElements(value);
+    validate(elements, modelName);
+    const seen = new Set();
+    return elements.map((element, index) => {
+        const label = `${modelName} 第 ${index + 1} 个元素`;
+        if (!KIE_NAME_PATTERN.test(element.name))
+            throw new Error(`${label}的名称须为 1–20 个汉字、字母、数字或下划线`);
+        if (KIE_RESERVED_NAME.test(element.name))
+            throw new Error(`${label}的名称“${element.name}”是保留名，请换一个`);
+        const key = element.name.toLowerCase();
+        if (seen.has(key))
+            throw new Error(`${label}的名称“${element.name}”与前面的元素重复`);
+        seen.add(key);
+        if (element.kind === 'video')
+            return { name: element.name, video_url: element.video, ...(options.sendType ? { type: 'video' } : {}) };
+        return { name: element.name, image_urls: element.images };
+    });
+}

package/dist/catalog/siliconflow/moss-ttsd-0.5.model.js

@@ -3,7 +3,7 @@ import { speechCharacterPrice, speechString, speechText } from '../audio/tts-com
 export const siliconflowMossTtsd05Model = defineModel({
     meta: {
         id: 'siliconflow-moss-ttsd-0.5', canonicalModelId: 'moss-ttsd-v0.5', provider: 'siliconflow', type: 'audio',
-        tags: ['text-to-audio'],
+        tags: ['text-to-audio', 'multi-speaker-dialogue'],
     },
     inputLimits: { images: { max: 0 }, videos: { max: 0 } },
     params: [

package/dist/catalog/validate.js

@@ -8,6 +8,7 @@
  * 同样的报错文案），应用侧改为调用本模块，不再维护第二份判断，避免源码与副本
  * 静默分叉。
  */
+import { isModelTag } from '../types/modelTags.js';
 export class ModelRuntimeValidationError extends Error {
     constructor(message) {
         super(message);
@@ -55,6 +56,10 @@ function validateMeta(model, fail) {
         if (!Array.isArray(meta.tags)) {
             fail('Model meta.tags must be an array');
         }
+        const unknown = meta.tags.filter((tag) => typeof tag !== 'string' || !isModelTag(tag));
+        if (unknown.length) {
+            fail(`Model meta.tags has tags outside MODEL_TAG_DEFINITIONS: ${unknown.join(', ')}`);
+        }
     }
     if (meta.polling) {
         if (typeof meta.polling.interval !== 'number' || meta.polling.interval <= 0) {
@@ -276,6 +281,9 @@ function validatePricing(model, fail) {
         fail('Model pricing.currency is required and must be a string');
     }
     const hasFixed = typeof pricing.fixed === 'number';
+    if (pricing.billingMode !== undefined && pricing.billingMode !== 'estimated') {
+        fail('Model pricing.billingMode must be estimated');
+    }
     const hasCalculator = typeof pricing.calculator === 'function';
     if (!hasFixed && !hasCalculator) {
         fail('Model pricing must have either fixed or calculator');
@@ -323,10 +331,15 @@ function validatePricing(model, fail) {
             ? ['megapixels', 'width', 'height', 'fileSizeBytes'].includes(requirement.metric)
             : requirement.mediaType === 'video'
                 ? ['durationSeconds', 'megapixels', 'width', 'height'].includes(requirement.metric)
-                : false;
+                : requirement.mediaType === 'audio'
+                    ? requirement.metric === 'durationSeconds'
+                    : false;
         if (!validMetric) {
             fail(`${prefix} has an unsupported mediaType/metric combination`);
         }
+        if (requirement.when !== undefined && typeof requirement.when !== 'function') {
+            fail(`${prefix}.when must be a function`);
+        }
         const multiplier = requirement.multiplier;
         if (!multiplier)
             return;

package/dist/catalog/volcengine/seedance-2.0-fast.model.d.ts

@@ -0,0 +1,2 @@
+export declare const volcengineSeedance20FastModel: import("../../index.js").ModelRuntimeDefinition;
+export default volcengineSeedance20FastModel;

package/dist/catalog/volcengine/seedance-2.0-fast.model.js

@@ -0,0 +1,11 @@
+import { createVolcengineSeedanceModel } from './seedance.js';
+// 价格：https://www.volcengine.com/docs/82379/1544106 （在线推理 token 单价，分输入含 / 不含视频两档）
+export const volcengineSeedance20FastModel = createVolcengineSeedanceModel({
+    id: 'volcengine-seedance-2.0-fast', canonicalModelId: 'seedance-2.0-fast', seriesRank: 2.1,
+    apiModel: 'doubao-seedance-2-0-fast-260128', paramPrefix: 'volcengineSeedance20Fast', generation: '2.0',
+    resolutions: ['480p', '720p'],
+    tokenPrices: { '480p': 37, '720p': 37 },
+    videoTokenPrices: { '480p': 22, '720p': 22 },
+    priceDescription: '按 token 计费：480p/720p ¥37 每百万 token；token ≈ 时长 × 宽 × 高 × 24 / 1024（720p 16:9 5 秒约 ¥4.00）；带参考视频时 480p/720p ¥22 每百万 token，输入视频时长也计入（不足 4 秒按 4 秒）',
+});
+export default volcengineSeedance20FastModel;

package/dist/catalog/volcengine/seedance-2.0-mini.model.d.ts

@@ -0,0 +1,2 @@
+export declare const volcengineSeedance20MiniModel: import("../../index.js").ModelRuntimeDefinition;
+export default volcengineSeedance20MiniModel;

package/dist/catalog/volcengine/seedance-2.0-mini.model.js

@@ -0,0 +1,11 @@
+import { createVolcengineSeedanceModel } from './seedance.js';
+// 价格：https://www.volcengine.com/docs/82379/1544106 （在线推理 token 单价，分输入含 / 不含视频两档）
+export const volcengineSeedance20MiniModel = createVolcengineSeedanceModel({
+    id: 'volcengine-seedance-2.0-mini', canonicalModelId: 'seedance-2.0-mini', seriesRank: 2.05,
+    apiModel: 'doubao-seedance-2-0-mini-260615', paramPrefix: 'volcengineSeedance20Mini', generation: '2.0',
+    resolutions: ['480p', '720p'],
+    tokenPrices: { '480p': 23, '720p': 23 },
+    videoTokenPrices: { '480p': 14, '720p': 14 },
+    priceDescription: '按 token 计费：480p/720p ¥23 每百万 token；token ≈ 时长 × 宽 × 高 × 24 / 1024（720p 16:9 5 秒约 ¥2.48）；带参考视频时 480p/720p ¥14 每百万 token，输入视频时长也计入（不足 4 秒按 4 秒）',
+});
+export default volcengineSeedance20MiniModel;

package/dist/catalog/volcengine/seedance-2.0.model.d.ts

@@ -0,0 +1,2 @@
+export declare const volcengineSeedance20Model: import("../../index.js").ModelRuntimeDefinition;
+export default volcengineSeedance20Model;

package/dist/catalog/volcengine/seedance-2.0.model.js

@@ -0,0 +1,11 @@
+import { createVolcengineSeedanceModel } from './seedance.js';
+// 价格：https://www.volcengine.com/docs/82379/1544106 （在线推理 token 单价，分输入含 / 不含视频两档）
+export const volcengineSeedance20Model = createVolcengineSeedanceModel({
+    id: 'volcengine-seedance-2.0', canonicalModelId: 'seedance-2.0', seriesRank: 2,
+    apiModel: 'doubao-seedance-2-0-260128', paramPrefix: 'volcengineSeedance20', generation: '2.0',
+    resolutions: ['480p', '720p', '1080p', '4k'],
+    tokenPrices: { '480p': 46, '720p': 46, '1080p': 51, '4k': 26 },
+    videoTokenPrices: { '480p': 28, '720p': 28, '1080p': 31, '4k': 16 },
+    priceDescription: '按 token 计费：480p/720p ¥46、1080p ¥51、4K ¥26 每百万 token；token ≈ 时长 × 宽 × 高 × 24 / 1024（720p 16:9 5 秒约 ¥4.97）；带参考视频时 480p/720p ¥28、1080p ¥31、4K ¥16 每百万 token，输入视频时长也计入（不足 4 秒按 4 秒）',
+});
+export default volcengineSeedance20Model;

package/dist/catalog/volcengine/seedance-2.5.model.d.ts

@@ -0,0 +1,2 @@
+export declare const volcengineSeedance25Model: import("../../index.js").ModelRuntimeDefinition;
+export default volcengineSeedance25Model;

package/dist/catalog/volcengine/seedance-2.5.model.js

@@ -0,0 +1,11 @@
+import { createVolcengineSeedanceModel } from './seedance.js';
+// 价格：https://www.volcengine.com/docs/82379/1544106 （在线推理 token 单价，分输入含 / 不含视频两档）
+export const volcengineSeedance25Model = createVolcengineSeedanceModel({
+    id: 'volcengine-seedance-2.5', canonicalModelId: 'seedance-2.5', seriesRank: 2.5,
+    apiModel: 'doubao-seedance-2-5-260628', paramPrefix: 'volcengineSeedance25', generation: '2.5',
+    resolutions: ['480p', '720p', '1080p'],
+    tokenPrices: { '480p': 70, '720p': 70, '1080p': 77 },
+    videoTokenPrices: { '480p': 42, '720p': 42, '1080p': 46 },
+    priceDescription: '按 token 计费：480p/720p ¥70、1080p ¥77 每百万 token；token ≈ 时长 × 宽 × 高 × 24 / 1024（720p 16:9 5 秒约 ¥7.56）；自动时长按 30 秒预估；带参考视频时 480p/720p ¥42、1080p ¥46 每百万 token，输入视频时长也计入（不足 4 秒按 4 秒）',
+});
+export default volcengineSeedance25Model;

package/dist/catalog/volcengine/seedance.d.ts

@@ -0,0 +1,17 @@
+/** 火山方舟官方 Seedance 2.5 / 2.0 系列视频生成模型的共用运行时契约。 */
+export interface VolcengineSeedanceConfig {
+    id: string;
+    canonicalModelId: string;
+    seriesRank: number;
+    apiModel: string;
+    /** 参数 ID 前缀，例如 `volcengineSeedance25`。 */
+    paramPrefix: string;
+    generation: '2.5' | '2.0';
+    resolutions: string[];
+    /** 输入不含视频时的在线推理 token 单价（元/百万 token），按输出分辨率。 */
+    tokenPrices: Record<string, number>;
+    /** 输入含参考视频时的 token 单价（元/百万 token）；输入视频时长同样计入 token 用量。 */
+    videoTokenPrices: Record<string, number>;
+    priceDescription: string;
+}
+export declare function createVolcengineSeedanceModel(config: VolcengineSeedanceConfig): import("../../index.js").ModelRuntimeDefinition;

package/dist/catalog/volcengine/seedance.js

@@ -0,0 +1,192 @@
+/** 火山方舟官方 Seedance 2.5 / 2.0 系列视频生成模型的共用运行时契约。 */
+import { defineModel } from '../defineModel.js';
+import { hasUploadedImage, resolveUploadedImageSources, resolveUploadedVideoDurationSeconds, resolveUploadedVideoSources } from '../shared/mediaPresence.js';
+const RATIOS = ['16:9', '4:3', '1:1', '3:4', '9:16', '21:9'];
+const MIN_BILLED_INPUT_VIDEO_SECONDS = 4;
+/** 官方「不同宽高比对应的宽高像素值」表；2.5 与 2.0 系列在 480p 档取值不同。 */
+const PIXELS = {
+    '2.5': {
+        '480p': { '16:9': [854, 480], '4:3': [752, 560], '1:1': [640, 640], '3:4': [560, 752], '9:16': [480, 854], '21:9': [992, 432] },
+        '720p': { '16:9': [1280, 720], '4:3': [1112, 834], '1:1': [960, 960], '3:4': [834, 1112], '9:16': [720, 1280], '21:9': [1470, 630] },
+        '1080p': { '16:9': [1920, 1080], '4:3': [1664, 1248], '1:1': [1440, 1440], '3:4': [1248, 1664], '9:16': [1080, 1920], '21:9': [2206, 946] },
+    },
+    '2.0': {
+        '480p': { '16:9': [864, 496], '4:3': [752, 560], '1:1': [640, 640], '3:4': [560, 752], '9:16': [496, 864], '21:9': [992, 432] },
+        '720p': { '16:9': [1280, 720], '4:3': [1112, 834], '1:1': [960, 960], '3:4': [834, 1112], '9:16': [720, 1280], '21:9': [1470, 630] },
+        '1080p': { '16:9': [1920, 1080], '4:3': [1664, 1248], '1:1': [1440, 1440], '3:4': [1248, 1664], '9:16': [1080, 1920], '21:9': [2206, 946] },
+        '4k': { '16:9': [3840, 2160], '4:3': [3326, 2494], '1:1': [2880, 2880], '3:4': [2494, 3326], '9:16': [2160, 3840], '21:9': [4398, 1886] },
+    },
+};
+function audioSources(params) {
+    const pick = (value) => Array.isArray(value)
+        ? value.filter((item) => typeof item === 'string' && item.trim().length > 0) : [];
+    const uploaded = pick(params.uploadedAudioFilePaths);
+    return uploaded.length > 0 ? uploaded : pick(params.audios);
+}
+export function createVolcengineSeedanceModel(config) {
+    const key = (name) => config.paramPrefix + name;
+    const is25 = config.generation === '2.5';
+    const maxDuration = is25 ? 30 : 15;
+    const maxReferenceImages = is25 ? 30 : 9;
+    const maxReferenceAudios = is25 ? 10 : 3;
+    // 参考视频：2.5 最多 10 段、合计 ≤ 30 秒；2.0 系列最多 3 段、合计 ≤ 15 秒。只收公网 URL，本地文件经通用文件上传服务托管。
+    const maxReferenceVideos = is25 ? 10 : 3;
+    const maxReferenceVideoSeconds = is25 ? 30 : 15;
+    // 2.5 才有显式任务类型；编辑、延长都要求至少一段参考视频且比例自适应，编辑还不能指定时长。
+    // 默认显式发 reference：提交时就校验限制，也不会因提示词里的“替换”被平台判成编辑任务后异步失败。
+    const taskType = (params) => {
+        const value = String(params[key('TaskType')] ?? 'reference');
+        return is25 && (value === 'auto' || value === 'edit' || value === 'extend') ? value : 'reference';
+    };
+    const editing = `${key('Mode')} === "reference-to-video" && ${key('TaskType')} === "edit"`;
+    const resolution = (params) => {
+        const value = String(params[key('Resolution')] ?? '720p');
+        return config.resolutions.includes(value) ? value : '720p';
+    };
+    const autoDuration = (params) => is25 && params[key('AutoDuration')] === true;
+    const duration = (params) => Math.min(maxDuration, Math.max(4, Math.round(Number(params[key('Duration')] || 5))));
+    const params = [
+        {
+            id: key('Mode'), type: 'dropdown', order: 1,
+            // 参考模式不设素材下限（2.5 可以只传音频），默认放在文/图生视频，带多图或音频时由按素材选模式切过去。
+            default: 'text-image-to-video',
+            options: [{ value: 'text-image-to-video' }, { value: 'reference-to-video' }],
+        },
+        ...(is25 ? [{
+                id: key('TaskType'), type: 'dropdown', order: 2, default: 'reference',
+                visible: { condition: `${key('Mode')} === "reference-to-video"` },
+                options: [{ value: 'reference' }, { value: 'edit' }, { value: 'extend' }, { value: 'auto' }],
+            }] : []),
+        {
+            id: key('AspectRatio'), type: 'dropdown', order: 3,
+            // 2.5 的首帧 / 首尾帧与编辑、延长任务只接受 adaptive，比例跟随素材。
+            ...(is25 ? { visible: { condition: (params) => params[key('Mode')] === 'reference-to-video'
+                        ? !['edit', 'extend'].includes(String(params[key('TaskType')]))
+                        : !hasUploadedImage(params) } } : {}),
+            default: 'smart',
+            options: [{ value: 'smart' }, ...['1:1', '4:3', '3:4', '16:9', '9:16', '21:9'].map((value) => ({ value }))],
+        },
+        {
+            id: key('Resolution'), type: 'dropdown', order: 4,
+            default: '720p',
+            options: config.resolutions.map((value) => ({ value })),
+        },
+        {
+            id: key('Duration'), type: 'number', order: 5,
+            default: 5, min: 4, max: maxDuration, step: 1,
+            ...(is25 ? { visible: { condition: `!(${editing})` } } : {}),
+        },
+        ...(is25 ? [{ id: key('AutoDuration'), type: 'switch', order: 6, default: false, visible: { condition: `!(${editing})` } }] : []),
+        { id: key('GenerateAudio'), type: 'switch', order: 7, default: true },
+        { id: key('ReturnLastFrame'), type: 'switch', order: 8, default: false },
+        { id: key('WebSearch'), type: 'switch', order: 9, default: false },
+    ];
+    return defineModel({
+        meta: {
+            id: config.id, canonicalModelId: config.canonicalModelId, seriesId: 'seedance', seriesRank: config.seriesRank,
+            provider: 'volcengine', type: 'video',
+            tags: is25
+                ? ['text-to-video', 'image-to-video', 'start-end-frame', 'reference-mode', 'video-to-video', 'video-extension']
+                : ['text-to-video', 'image-to-video', 'start-end-frame', 'reference-mode'],
+            polling: { interval: 5000, maxAttempts: 400, expectedAttempts: 60 },
+        },
+        inputLimits: {
+            images: { max: 2 }, videos: { max: 0 }, audios: { max: 0 },
+            rules: [
+                {
+                    when: `${key('Mode')} === "reference-to-video"`,
+                    images: { max: maxReferenceImages }, videos: { max: maxReferenceVideos }, audios: { max: maxReferenceAudios },
+                },
+                ...(is25 ? [{
+                        when: `${key('Mode')} === "reference-to-video" && (${key('TaskType')} === "edit" || ${key('TaskType')} === "extend")`,
+                        images: { max: maxReferenceImages }, videos: { min: 1, max: maxReferenceVideos }, audios: { max: maxReferenceAudios },
+                    }] : []),
+                { imageConstraints: { minShortSidePx: 300, maxLongSidePx: 6000, minAspectRatio: 0.4, maxAspectRatio: 2.5 } },
+            ],
+        },
+        params,
+        endpoints: '/api/v3/contents/generations/tasks',
+        request: {
+            builder: (input) => {
+                const images = resolveUploadedImageSources(input);
+                const audios = audioSources(input);
+                const videos = resolveUploadedVideoSources(input);
+                const reference = input[key('Mode')] !== 'text-image-to-video';
+                const rawRatio = String(input[key('AspectRatio')] || 'smart');
+                const prompt = typeof input.prompt === 'string' ? input.prompt.trim() : '';
+                const content = [];
+                if (prompt)
+                    content.push({ type: 'text', text: prompt });
+                const body = {
+                    model: config.apiModel,
+                    content,
+                    resolution: resolution(input),
+                    ratio: RATIOS.includes(rawRatio) ? rawRatio : 'adaptive',
+                    duration: autoDuration(input) ? -1 : duration(input),
+                    generate_audio: input[key('GenerateAudio')] !== false,
+                    return_last_frame: input[key('ReturnLastFrame')] === true,
+                    watermark: false,
+                };
+                if (input[key('WebSearch')] === true)
+                    body.tools = [{ type: 'web_search' }];
+                if (reference) {
+                    if (images.length + audios.length + videos.length === 0)
+                        throw new Error(`Seedance ${config.generation} 参考生视频至少需要一张参考图、一段参考视频或参考音频`);
+                    if (!is25 && audios.length > 0 && images.length + videos.length === 0)
+                        throw new Error('Seedance 2.0 的参考音频必须与参考图或参考视频一起使用');
+                    const task = taskType(input);
+                    if ((task === 'edit' || task === 'extend') && videos.length === 0) {
+                        throw new Error(`Seedance 2.5 的${task === 'edit' ? '视频编辑' : '视频延长'}至少需要一段参考视频`);
+                    }
+                    if (is25)
+                        body.omni_reference_task_type = task;
+                    if (task === 'edit' || task === 'extend')
+                        body.ratio = 'adaptive';
+                    // 编辑任务的输出时长跟随原视频，官方只接受 -1。
+                    if (task === 'edit')
+                        body.duration = -1;
+                    for (const url of images.slice(0, maxReferenceImages))
+                        content.push({ type: 'image_url', image_url: { url }, role: 'reference_image' });
+                    for (const url of videos.slice(0, maxReferenceVideos))
+                        content.push({ type: 'video_url', video_url: { url }, role: 'reference_video' });
+                    for (const url of audios.slice(0, maxReferenceAudios))
+                        content.push({ type: 'audio_url', audio_url: { url }, role: 'reference_audio' });
+                    return body;
+                }
+                if (!prompt && images.length === 0)
+                    throw new Error(`Seedance ${config.generation} 文生视频需要填写提示词`);
+                images.slice(0, 2).forEach((url, index) => {
+                    content.push({ type: 'image_url', image_url: { url }, role: index === 0 ? 'first_frame' : 'last_frame' });
+                });
+                // 2.5 首帧、首尾帧任务只接受 adaptive，输出比例跟随首帧。
+                if (is25 && images.length > 0)
+                    body.ratio = 'adaptive';
+                return body;
+            },
+        },
+        pricing: {
+            currency: '¥',
+            calculator: (input) => {
+                const res = resolution(input);
+                const rawRatio = String(input[key('AspectRatio')] || 'smart');
+                const table = PIXELS[config.generation][res] ?? PIXELS[config.generation]['720p'];
+                const [width, height] = table[rawRatio] ?? table['16:9'];
+                const withVideo = input[key('Mode')] !== 'text-image-to-video' && resolveUploadedVideoSources(input).length > 0;
+                const task = taskType(input);
+                // 读不到参考视频时长时按输出时长估，并封顶到平台的合计上限。
+                const inputSeconds = withVideo
+                    ? Math.min(maxReferenceVideoSeconds, resolveUploadedVideoDurationSeconds(input, duration(input)))
+                    : 0;
+                // 编辑任务输出时长跟随原视频；自动时长按上限预估。
+                const outputSeconds = task === 'edit' && withVideo ? inputSeconds : autoDuration(input) ? maxDuration : duration(input);
+                // 官方 token 用量估算：(输入视频时长 + 输出时长) × 宽 × 高 × 24fps / 1024。
+                // 含视频时另有最低 token 用量：官方价格示例里输入 2–4 秒同价，即输入视频不足 4 秒按 4 秒计。
+                const billedInputSeconds = withVideo ? Math.max(MIN_BILLED_INPUT_VIDEO_SECONDS, inputSeconds) : 0;
+                const tokens = (billedInputSeconds + outputSeconds) * width * height * 24 / 1024;
+                const prices = withVideo ? config.videoTokenPrices : config.tokenPrices;
+                return tokens * (prices[res] ?? prices['720p']) / 1_000_000;
+            },
+            description: config.priceDescription,
+        },
+    });
+}

package/dist/catalog/volcengine/seedream-5.0-lite.model.js

@@ -6,7 +6,8 @@ export const volcengineSeedream50LiteModel = defineModel({
     meta: {
         id: 'volcengine-seedream-5.0-lite', canonicalModelId: 'seedream-5.0-lite', seriesId: 'seedream', seriesRank: 5,
         provider: 'volcengine', type: 'image',
-        tags: ['text-to-image', 'image-to-image', 'supports-image-editing', 'supports-multi-image', 'max-images-14', 'multi-output', 'supports-4k', 'provider-volcengine'],
+        deprecated: '火山方舟 2026-11-24 关停 doubao-seedream-5-0-260128，官方替代为 5.0 Pro',
+        tags: ['text-to-image', 'image-to-image'],
         aliases: ['seedream-5-lite-official']
     },
     inputLimits: { images: { max: 14 }, videos: { max: 0 } },

package/dist/catalog/volcengine/seedream-5.0.js

@@ -9,7 +9,7 @@ export function createSeedream50Model(config) {
         meta: {
             id: config.id, canonicalModelId: config.canonicalModelId, seriesId: 'seedream', seriesRank: 5.1,
             provider: 'volcengine', type: 'image',
-            tags: ['text-to-image', 'image-to-image', 'supports-image-editing', 'supports-multi-image', 'supports-layer-decomposition', 'max-images-10', 'provider-volcengine'],
+            tags: ['text-to-image', 'image-to-image', 'layer-decomposition', 'transparent-background'],
             aliases: config.aliases
         },
         inputLimits: {

package/dist/catalog/volcengine-speech/seed-icl-2.0.model.js

@@ -7,7 +7,7 @@ export const VOLC_CLONE_LANGUAGES = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 1
 export const volcengineSeedIcl20Model = defineModel({
     meta: {
         id: 'volcengine-seed-icl-2.0', canonicalModelId: 'seed-icl-2.0', provider: 'volcengine-speech', type: 'audio',
-        tags: ['text-to-audio', 'voice-clone', 'provider-volcengine'],
+        tags: ['text-to-audio', 'voice-clone', 'timestamps'],
         polling: { interval: 3000, maxAttempts: 120 },
     },
     inputLimits: { images: { max: 0 }, videos: { max: 0 } },
@@ -59,7 +59,7 @@ export const volcengineSeedIcl20Model = defineModel({
                 throw new Error('后付费音色首次正式合成将收取 138 元槽位费并锁定音色，请先确认费用');
             return {
                 resource_id: systemVoice ? 'seed-tts-2.0' : 'seed-icl-2.0',
-                req_params: { text: speechText(params), ...(!systemVoice ? { model: 'seed-tts-2.0-standard' } : {}), speaker, audio_params: { format: 'mp3', sample_rate: 24000 } },
+                req_params: { text: speechText(params), ...(!systemVoice ? { model: 'seed-tts-2.0-standard' } : {}), speaker, audio_params: { format: 'mp3', sample_rate: 24000, enable_subtitle: true } },
             };
         },
     },

package/dist/catalog/volcengine-speech/seed-tts-2.0.model.js

@@ -3,7 +3,7 @@ import { speechCharacterPrice, speechString, speechText } from '../audio/tts-com
 export const volcengineSeedTts20Model = defineModel({
     meta: {
         id: 'volcengine-seed-tts-2.0', canonicalModelId: 'seed-tts-2.0', provider: 'volcengine-speech', type: 'audio',
-        tags: ['text-to-audio', 'provider-volcengine'],
+        tags: ['text-to-audio', 'timestamps'],
     },
     inputLimits: { images: { max: 0 }, videos: { max: 0 } },
     params: [
@@ -16,7 +16,7 @@ export const volcengineSeedTts20Model = defineModel({
             req_params: {
                 text: speechText(params),
                 speaker: speechString(params.volcSeedTtsSpeaker) ?? 'zh_female_vv_uranus_bigtts',
-                audio_params: { format: 'mp3', sample_rate: 24000 },
+                audio_params: { format: 'mp3', sample_rate: 24000, enable_subtitle: true },
             },
         }),
     },

package/dist/discovery/index.js

@@ -1,3 +1,4 @@
+import { getModelTagDefinition, isModelTag } from '../types/modelTags.js';
 import { AiRuntimeError } from '../runtime/AiRuntimeError.js';
 import { normalizeCapabilityStableId, validateCapabilityDescriptor, } from '../capabilities/validation.js';
 import { validateLlmModuleDescriptor } from '../llm/modules/validation.js';
@@ -20,9 +21,13 @@ export function createModelCapabilityDiscovery(input) {
         ...(input.generationPacks ?? []).flatMap((pack) => pack.models),
     ]);
     for (const model of generation) {
+        if (model.meta.deprecated)
+            continue;
         add({ sourceKind: 'generation-model', id: model.meta.id, profile: profileGenerationModel(model), source: model });
     }
     for (const entry of input.llmEntries ?? []) {
+        if (entry.deprecated)
+            continue;
         add({ sourceKind: 'llm-model', id: entry.id, profile: profileLlmCatalogEntry(entry), source: entry });
     }
     for (const model of input.llmModels ?? []) {
@@ -48,7 +53,8 @@ export function createModelCapabilityDiscovery(input) {
 }
 export function profileGenerationModel(model) {
     const tags = [...new Set(model.meta.tags ?? [])];
-    const operations = deriveGenerationOperations(model.meta.type, tags);
+    // 未经 defineModel 校验的模型可能带词表外的旧标签：照样留在 tags 里，只是不参与推导操作
+    const operations = deriveGenerationOperations(model.meta.type, tags.filter(isModelTag));
     const inputs = new Set();
     if (model.acceptsPrompt !== false)
         inputs.add('text');
@@ -64,7 +70,7 @@ export function profileGenerationModel(model) {
         operations,
         acceptedInputContentKinds: [...inputs],
         outputContentKinds: [output],
-        features: tags.filter((tag) => !tag.startsWith('provider-') && !operations.includes(tag)),
+        features: tags.filter((tag) => !operations.includes(tag)),
         tags,
     };
 }
@@ -143,18 +149,19 @@ function deriveGenerationOperations(type, tags) {
         operations.add('image-generation');
         if (tags.includes('text-to-image'))
             operations.add('text-to-image');
-        if (tags.some((tag) => ['image-to-image', 'supports-image-editing', 'image-edit'].includes(tag)))
+        // 发现接口的图片操作只分生成与编辑，专项都归入编辑。
+        if (tags.some((tag) => tag === 'image-to-image' || getModelTagDefinition(tag).layer === 'specialty'))
             operations.add('image-edit');
     }
     else if (type === 'video') {
         operations.add('video-generation');
         if (tags.includes('text-to-video'))
             operations.add('text-to-video');
-        if (tags.some((tag) => ['image-to-video', 'start-end-frame'].includes(tag)))
+        if (tags.includes('image-to-video'))
             operations.add('image-to-video');
-        if (tags.some((tag) => ['reference-mode', 'video-reference', 'multi-image-reference'].includes(tag)))
+        if (tags.includes('reference-mode'))
             operations.add('reference-to-video');
-        if (tags.some((tag) => ['supports-video-editing', 'video-to-video', 'video-extension'].includes(tag)))
+        if (tags.some((tag) => tag === 'video-to-video' || tag === 'video-extension'))
             operations.add('video-edit');
     }
     else if (type === 'audio') {

package/dist/index.d.ts

@@ -20,6 +20,7 @@ export * from './protocols/index.js';
 export * from './upload/index.js';
 export * from './providers/index.js';
 export * from './types/index.js';
+export type { SpeechTimings, GenerationOutputTimings } from './types/speechTimings.js';
 export * from './structured-output/index.js';
 export * from './catalog/index.js';
 export * from './pricing/index.js';
@@ -31,3 +32,5 @@ export * from './capabilities/speech-recognition/bailian/index.js';
 export * from './capabilities/speech-recognition/volcengine/index.js';
 export * from './capabilities/speech-recognition/siliconflow/index.js';
 export * from './capabilities/speech-recognition/groq/index.js';
+export * from './capabilities/translation/index.js';
+export * from './capabilities/translation/bailian/index.js';

package/dist/index.js

@@ -45,3 +45,5 @@ export * from './capabilities/speech-recognition/bailian/index.js';
 export * from './capabilities/speech-recognition/volcengine/index.js';
 export * from './capabilities/speech-recognition/siliconflow/index.js';
 export * from './capabilities/speech-recognition/groq/index.js';
+export * from './capabilities/translation/index.js';
+export * from './capabilities/translation/bailian/index.js';

package/dist/llm/bigmodel/preset.js

@@ -1,5 +1,6 @@
 import { createLlmCapabilitiesForModel } from '../defaults.js';
 import { resolveLlmEndpointIdentity } from '../endpointProfiles.js';
+import { findLlmModelCatalogEntry } from '../modelCatalog.js';
 import { BIGMODEL_ENDPOINT_PROFILE_FAMILY, BIGMODEL_PROVIDER_FAMILY_ID, } from './profiles.js';
 export function createBigmodelProvider(options = {}) {
     const profileId = options.endpointProfile ?? 'cn';
@@ -36,7 +37,7 @@ export function createBigmodelModels(provider) {
     const profile = BIGMODEL_ENDPOINT_PROFILE_FAMILY.profiles.find(item => item.id === identity.endpointProfile);
     if (!profile)
         throw new Error('[llm_endpoint_profile_unknown] resolved BigModel endpoint profile is unavailable');
-    return profile.modelIds.map(modelId => ({
+    return profile.modelIds.filter(modelId => !findLlmModelCatalogEntry(modelId)?.deprecated).map(modelId => ({
         providerId: identity.providerId,
         providerFamilyId: identity.providerFamilyId,
         endpointProfile: identity.endpointProfile,

package/dist/llm/bigmodel/profiles.js

@@ -42,12 +42,6 @@ export const BIGMODEL_GLM_5_3_FLASH_PRICING = {
         inputPerMillionTokens: 0.8,
         outputPerMillionTokens: 2.8,
         cacheReadPerMillionTokens: 0.23,
-        promotion: {
-            inputPerMillionTokens: 0.4,
-            outputPerMillionTokens: 1.4,
-            cacheReadPerMillionTokens: 0.115,
-            observedAt: '2026-08-28',
-        },
     },
     global: {
         currency: 'USD',

package/dist/llm/chat.d.ts

@@ -27,6 +27,10 @@ export interface LlmChatCompletedInfo {
     output: string;
     reasoningOutput: string;
     usage: LlmUsageDto | null;
+    /** 发出请求到第一个字（含思考内容）的毫秒数；没有任何输出时为 null。 */
+    firstTokenMs: number | null;
+    /** 首字之后的输出速度（输出 token / 秒，含思考 token）；供应商没给用量时为 null。 */
+    outputTokensPerSecond: number | null;
     finishReason: string | null;
     truncated: boolean;
     toolCalls?: LlmStreamToolCall[];
@@ -47,6 +51,10 @@ export interface LlmChatStreamOutcome {
     output: string;
     reasoningOutput: string;
     usage: LlmUsageDto | null;
+    /** 发出请求到第一个字（含思考内容）的毫秒数；没有任何输出时为 null。 */
+    firstTokenMs: number | null;
+    /** 首字之后的输出速度（输出 token / 秒，含思考 token）；供应商没给用量时为 null。 */
+    outputTokensPerSecond: number | null;
     finishReason: string | null;
     truncated: boolean;
     toolCalls?: LlmStreamToolCall[];